AI News #123: Week Ending February 06, 2026 with 79 Executive Summaries

February 7, 2026

About This Week’s Covers

This week’s cover is a riff on the OpenClaw masthead. OpenClaw’s agents are powerful and pesky, so I had Gemini redo the mascot as The Joker and replace the OpenClaw name with the newsletter date. I’m trying to show how little Photoshop is needed. Gemini is a new era of image editing.

For the category covers, I gave the Moltbook icon to Claude and ran it through my Python script that automatically creates 55 cover images. It’s plain text prompted as opposed to literally referencing the icon (which is possible and accurate).

I’ve put my favorite few below:

This week’s humanities reading is a quote from David Foster Wallace from Consider The Lobster. It feels agentic and surrealist:

“It’s not that students don’t ‘get’ Kafka’s humor but that we’ve taught them to see humor as something you get — the same way we’ve taught them that a self is something you just have. No wonder they cannot appreciate the really central Kafka joke — that the horrific struggle to establish a human self results in a self whose humanity is inseparable from that horrific struggle. That our endless and impossible journey toward home is in fact our home.”

This Week By The Numbers

Total Organized Headlines: 447

This Week’s Executive Summaries

This week, I organized 448 headlines. 111 of them informed the executive summaries. I’m going to start with a few top stories and then move into stories by company name, as well as one or two general categories.

Top Stories/Favorites

Goldman Sachs is rolling out Claude to automate accounting.
“So it starts. The market already reacts.” https://x.com/kimmonismus/status/2019865721338229180

“Anthropic is crashing the stock market with their new legal automation plugin Anthropic spooked investors, triggering a sharp selloff as markets feared AI could disrupt software-heavy industries like law and finance. The S&P 500 software index dropped nearly 9% in five days” https://x.com/kimmonismus/status/2019757481925464371

Google’s Incredible Performace
Google’s 52x AI Growth https://tomtunguz.com/google-earnings-q4-2025/

Google’s Gemini app has surpassed 750M monthly active users | TechCrunch https://techcrunch.com/2026/02/04/googles-gemini-app-has-surpassed-750m-monthly-active-users/

Gemini now processes over 10 billion tokens per minute via direct API use by our customers https://x.com/OfficialLoganK/status/2019166152199459074

“Our Q4/FY’25 results are in. Thanks to our partners and employees, it was a tremendous quarter, exceeding $400B in annual revenue for the first time.” https://x.com/sundarpichai/status/2019155348264042934

Alphabet’s Pichai: Google lowered gemini serving unit costs by 78% over 2025. “Google lowered Gemini serving unit costs by 78% over 2025.” https://x.com/financialjuice/status/2019164355636654281

“The Gemini app hit 750M+ monthly active users in Q4 2025. ChatGPT was reported to have 810M monthly active users by the end of 2025. The gap is shockingly small. Gemini has a real shot at passing ChatGPT.” https://x.com/Yuchenj_UW/status/2019157674143936980

“We’ve started to measure time horizons for recent models using our updated methodology. On this expanded suite of software tasks, we estimate that Gemini 3 Pro has a 50%-time-horizon of around 4 hrs (95% CI of 2 hr 10 mins to 7 hrs 20 mins).” https://x.com/METR_Evals/status/2018752230376210586

Google Waymos Using Genie’s World Model To Test Outliers
We’re excited to introduce the Waymo World Model—a frontier generative mode for large-scale, hyper-realistic autonomous driving simulation built on @GoogleDeepMind’s Genie 3. By simulating the “impossible”, we proactively prepare the Waymo Driver for some of the most rare and”” https://x.com/Waymo/status/2019804616746029508

Excited to share how Waymo is using Genie to simulate rare scenarios for autonomous driving evaluation, such as extreme weather, reckless driving by other drivers, and long-tailed road inhabitants 🐘. We are just scratching the surface of world simulation applications.”” https://x.com/shlomifruchter/status/2019820532485808329

Gemini+Genie 3 are helping @Waymo simulate long tail scenarios to make driving safer.”” https://x.com/JeffDean/status/2019824614139162804

“The model transfers Genie 3’s vast world knowledge into precise camera and 3D lidar data unique to Waymo’s hardware.“ a key reason and example of why world models are so important. https://x.com/kimmonismus/status/2019809839804010962

@Waymo The model transfers Genie 3’s vast world knowledge into precise camera and 3D lidar data unique to Waymo’s hardware. Engineers can prompt “what if” scenarios – like extreme weather or reckless drivers – to stress-test the system.”” https://x.com/GoogleDeepMind/status/2019809201812545835

OpenAI Performance “It’s so exponential, it literally looks like a wall. GPT-5.2 high sets new record in task duration. And it’s not even xhigh” https://x.com/kimmonismus/status/2019174066565849193?s=46

“We estimate that GPT-5.2 with `high` (not `xhigh`) reasoning effort has a 50%-time-horizon of around 6.6 hrs (95% CI of 3 hr 20 min to 17 hr 30 min) on our expanded suite of software tasks. This is the highest estimate for a time horizon measurement we have reported to date.” https://x.com/METR_Evals/status/2019169900317798857

ARC-AGI-2 Benchmark “Within just 10 months, performance on the ARC-AGI-2 benchmark surpassed 75%. Let that sink in.”” https://x.com/kimmonismus/status/2018800964891984181

NVIDIA’s World Action Model Trains Robots – First Try in Simulation
Jim Fan’s Essay is Important: The Second Pre-training Paradigm Next word prediction was the first pre-training paradigm. Now we are living through the second paradigm shift: world modeling, or “next physical state prediction”. https://x.com/DrJimFan/status/2018754323141054786

DreamZero: World Action Models are Zero-shot Policies https://dreamzero0.github.io/

Introducing DreamZero 🤖🌎 from @nvidia > A 14B “World Action Model” that achieves zero-shot generalization to unseen tasks & few-shot adaptation to new robots > The key? Jointly predicting video & actions in the same diffusion forward pass Project Page: https://x.com/jang_yoel/status/2019083437265867057

New milestone: we trained a robot foundation model on a world model backbone, and enabled zero-shot, open-world prompting capability for new verbs, nouns, and environments. If the world model can “”dream”” the right future in pixels, then the robot can execute well in motors. We”” https://x.com/DrJimFan/status/2019112603637920237

New paper from GEAR team @NVIDIARobotics We released DreamZero, a World Action Model that turns video world models into zero-shot robot policies. Built on a pretrained video diffusion backbone, it jointly predicts future video frames and actions, https://x.com/yukez/status/2019096072690553112

Anthropic Image Model
“Anthropics Image model is live on LMArena. It’s getting more exciting hour by hour” https://x.com/kimmonismus/status/2018689719324791022

“Lets go: Claude Image is coming!” https://x.com/kimmonismus/status/2018669423402660082

Brain Dumps as a Literary Form – by Dave Griffith
Ethan note: I personally do this all the time for my own use, and it’s how I prompt.
https://davegriffith.substack.com/p/brain-dumps-as-a-literary-form

General Business News

Investing Volume
“The AI Arms Race is over 5x larger than the Manhattan Project or the Apollo Program. 2026 CAPEX for Alphabet, Amazon, Meta and Microsoft is ~650B or ~2% of US GDP
For comparison: – Apollo program: 0.4% of US GDP – Manhatten project: 0.4% of US GDP – although it cost $150B the ISS doesn’t even register on the GDP scale, because it’s spread out over many countries and years” https://x.com/scaling01/status/2019789747896377697

Cerebras Cerebras Raises $1.1 Billion at $8.1 Billion Valuation https://www.cerebras.ai/press-release/series-g https://www.cerebras.ai/press-release/cerebras-systems-raises-usd1-billion-series-h

CoreWeave CoreWeave’s $30 Billion Bet on GPU Market Infrastructure
https://davefriedman.substack.com/p/coreweaves-30-billion-bet-on-gpu

ElevenLabs ElevenLabs raises $500M Series D at $11B valuation https://elevenlabs.io/blog/series-d

Journalism
“Who’s in to build shared AI journalism skills for agents? I just created a repo so we can share them. Share yours and let’s build this together! No single newsroom has all the answers. We compete on journalism, not on tooling.” https://x.com/fdaudens/status/2016916800928121109

Ai Journalism Skills – a Hugging Face Space by fdaudens https://huggingface.co/spaces/fdaudens/ai-journalism-skills

News Consumption Next Gen News 2 (NGN2) – Future of News and Young Audiences
https://www.next-gen-news.com/

OpenSource The Future of the Global Open-Source AI Ecosystem: From DeepSeek to AI+
https://huggingface.co/blog/huggingface/one-year-since-the-deepseek-moment-blog-3

Impact on Coding and Software
“As jobs begin to shift with AI I think there will be increasing numbers of people feeling like Aditya. I think figuring out what comes next is challenging. People are more psychologically resistant to major changes than you might expect, but that doesn’t mean it will be easy.” https://x.com/emollick/status/2019158085420376517

On Recursive Self-Improvement (Part I) – by Dean W. Ball https://www.hyperdimensional.co/p/on-recursive-self-improvement-part

AI as Creative Idea Generator
Ethan Mollick: “I still see the argument that AI is bad af generating ideas, but my colleagues and I, along with many other researchers, have repeatedly found that even the old GPT-4 could be prompted to generate more diverse and higher quality ideas than most people. And newer models do better.” https://x.com/emollick/status/2018388869515784595

On the Death of the Interface
Chat is Going to Eat the World – Dead Neurons
https://deadneurons.substack.com/p/chat-is-going-to-eat-the-world

Turing Test Toast?
“A pretty bold commentary in Nature written by linguists, computer scientists and philosophers declaring “by reasonable standards, including Turing’s own, we have artificial systems that are generally intelligent. The long-standing problem of creating AGI has been solved.” https://x.com/emollick/status/2018524111627325554

OpenClaw

A sane but extremely bull case on OpenClaw (Clawdbot) | Brandon Wang https://brandon.wang/2026/clawdbot

🤖 The AI That Called Its Human https://www.fintechbrainfood.com/p/the-ai-that-called-its-human

I don’t want my sycophantic Clawbot calling me for reassurance, but the interesting thing here is that the tweet is the instructions for the agent to set itself up. Plain English instructions that agents can follow may be a new avenue for marketing (and a security nightmare)”” https://x.com/emollick/status/2018826844817240092

The creator of Clawdbot/Moltbot/OpenClaw @steipete, pushes 144 commits per day on average. Pre-AI, this was impossible. He ships code he never reads. He’s a conductor. GPT and Claude are his orchestra. 5–10 AI agents run in parallel under his command. One person is now an army.”” https://x.com/Yuchenj_UW/status/2018029206542946582

Clawdbot’ Has AI Techies Buying Mac Minis – Slashdot https://tech.slashdot.org/story/26/01/28/0510226/clawdbot-has-ai-techies-buying-mac-minis

Matt’s moltbook experiment is truly unprecedented. People talk a lot about dead internet theory — but I didn’t expect to see it overtly manifested like this. As Karpathy put it: “we have never seen this many LLM agents (150,000 atm!) wired up via a global, persistent, agent-first”” https://x.com/bilawalsidhu/status/2017555088252772731

A few thoughts on the Open Claw nee Moltbot nee Clawdbot arc 1. I love how clearly people are seeing that the power lives in the harness even more than the model. Open source has made it obvious that models are quickly becoming a commodity. It’s the software that connects”” https://x.com/sarahmsachs/status/2018720637691572634

the other thing about moltbook – people aren’t really talking about these agents as individuals. they’re talking about it as a collective – almost like this ai swarm of agents coming to life. which honestly resonates because it holds a mirror to our own humanity. it’s”” https://x.com/bilawalsidhu/status/2017830542188687753

Carcinisation – Wikipedia https://en.wikipedia.org/wiki/Carcinisation

Heartbeat – OpenClaw https://docs.openclaw.ai/gateway/heartbeat

OpenClaw – Amazing Hands for a Brain That Doesn’t Yet Exist https://bengoertzel.substack.com/p/openclaw-amazing-hands-for-a-brain

Amazon

Alexa+ now available to everyone in the US—and free for Prime members
https://www.aboutamazon.com/news/devices/alexa-plus-available-free-prime-members-us

Anthropic

Claude Opus
Claude Opus 4.6 \ Anthropic https://www.anthropic.com/news/claude-opus-4-6

BREAKING: Claude Opus 4.6 by @AnthropicAI is now #1 across Code, Text and Expert Arena! Opus 4.6 shows significant gains across the board: – #1 Code Arena: +106 score vs Opus 4.5 – #1 Text Arena: scoring 1496, +10 vs Gemini 3 Pro – #1 Expert Arena: +~50 lead https://x.com/arena/status/2019842691442569566

Claude 4.6 Opus #1 on lmarena for text, coding and expert questions! https://x.com/scaling01/status/2019843682128822525

GPT-5.2-high took 26 TIMES LONGER than Claude 4.5 Opus to complete the METR benchmark suite”” https://x.com/scaling01/status/2019173770276081726

I told Claude 4.6 Opus to make a pokemon clone – max effort It reasoned for 1 hour and 30 minutes and used 110k tokens and 2 shotted this absolute behemoth. This is one of the coolest things I’ve ever made with AI https://x.com/chatgpt21/status/2019679978162634930

Other Anthropic News
Another banger by the Anthropic Engineering team. The mass-parallelized 16 Claude instances to build a full C compiler from scratch. 100,000 lines of Rust. Compiles the Linux kernel. No active human supervision. https://x.com/omarsar0/status/2019780306778104056

Anthropic partners with Allen Institute and Howard Hughes Medical Institute to accelerate scientific discovery \ Anthropic https://www.anthropic.com/news/anthropic-partners-with-allen-institute-and-howard-hughes-medical-institute

Building a C compiler with a team of parallel Claudes \ Anthropic https://www.anthropic.com/engineering/building-c-compiler

Claude Code now supports session sharing! You can share your full conversation with team members, or anyone with the link Available on web, desktop, and the mobile app https://x.com/lydiahallie/status/2018740156359229883

No Ads For Anthropic: Claude is a space to think | Anthropic \ Anthropic https://www.anthropic.com/news/claude-is-a-space-to-think

Don’t Build Agents, Build Skills Instead – Barry Zhang & Mahesh Murag, Anthropic – YouTube https://www.youtube.com/watch?v=CEvIs9y1uog&t=715s

ElevenLabs

Eleven v3 — Most Expressive AI Voice Model
https://elevenlabs.io/v3

ElevenLabs CEO: Voice is the next interface for AI | TechCrunch https://techcrunch.com/2026/02/05/elevenlabs-ceo-voice-is-the-next-interface-for-ai/

Google

Antigravity
Antigravity’s Inbox: Your AI Agent Command Center – YouTube https://www.youtube.com/shorts/yC-Cv37E-dM

Genie
Giving the world’s first photograph, the View from the Window at Le Gras, from 1822, to Genie 3. https://x.com/emollick/status/2018494862178316725

Google Genie just let me walk through 1900s San Francisco. I gave it one black-and-white photo. It gave me back a city — explorable from the sky or the street. This is the closest thing we have to a time machine. https://x.com/bilawalsidhu/status/2017045841836405035

One of the wildest emergent capabilities of Genie 3 is that maps actually work. As I walk around the forest, the GPS display updates its heading in real time. Remember. There is no game engine here. This is an AI hallucinating a working navigational instrument purely from next”” https://x.com/bilawalsidhu/status/2017252036719657193

Took an old photo of a WWI battlecruiser, gave it to Genie 3, and prompted it to let me play as a torpedo boat at the Battle of Jutland. Considering this is a research preview, astonishing how fast this has come. An AI dynamically generating the world with no game engine…”” https://x.com/emollick/status/2018198584508760108

genie 3 is insane. flying a drone over a city then hopping into a fighter jet to chase down a next-gen test vehicle. as one does when you have a proto holodeck at your disposal.”” https://x.com/bilawalsidhu/status/2017410842338460121

Mistral

Voxtral transcribes at the speed of sound. | Mistral AI https://mistral.ai/news/voxtral-transcribe-2

NVIDIA

Introducing NVIDIA Cosmos Policy for Advanced Robot Control https://huggingface.co/blog/nvidia/cosmos-policy-for-robot-control

Why Nvidia builds open models with Bryan Catanzaro https://www.interconnects.ai/p/why-nvidia-builds-open-models-with

OpenAI

Ads
Ads in ChatGPT: Why behavior matters more than targeting https://searchengineland.com/ads-in-chatgpt-why-behavior-matters-more-than-targeting-468144

OpenAI quietly lays groundwork for ads in ChatGPT https://searchengineland.com/openai-quietly-lays-groundwork-for-ads-in-chatgpt-468162

Codex
Introducing GPT-5.3-Codex | OpenAI https://openai.com/index/introducing-gpt-5-3-codex/

Introducing OpenAI Frontier | OpenAI https://openai.com/index/introducing-openai-frontier/

Introducing the Codex app | OpenAI https://openai.com/index/introducing-the-codex-app/

Codex | AI Coding Partner from OpenAI | OpenAI https://openai.com/codex/

Codex is now over 1 million active users!”” https://x.com/sama/status/2019219967250669741

Codex now pretty much builds itself, with the help and supervision of a great team. The bottleneck has shifted to being how fast we can help and supervise the outcome.”” https://x.com/thsottiaux/status/2018258151603388639

Excited to announce the Codex App: run multiple projects and threads in one focused app! 🔥 The app natively packs a lot of features making it easier to maximise your productivity: > Worktree mode keeps changes isolated – parallel tasks without touching your checkout >”” https://x.com/reach_vb/status/2018385536616956209

GPT-5.2 and GPT-5.2-Codex are now 40% faster. We have optimized our inference stack for all API customers. Same model. Same weights. Lower latency.”” https://x.com/OpenAIDevs/status/2018838297221726482

GPT-5.3-codex early review: this thing runs auto compaction like a maniac. it doesn’t wait for context to fill up and instead starts garbage collecting tokens the second it decides they’re dead weight. also it gives you updates every few steps like “”hey here’s what I am”” https://x.com/cto_junior/status/2019607817884475718

Is Codex actually ahead of Claude Code now??? I tried Codex yesterday while doing some training optimizations on Andrej’s nanochat. It has worse UI, ran my code in a CPU-only sandbox despite I have GPUs. It feels less agentic than Claude Code for sure. Sonnet 5, I’m still”” https://x.com/Yuchenj_UW/status/2019146114147185005

More than 200k people downloaded the Codex app in the first day. And they seem to love it. CODEX FTW!”” https://x.com/sama/status/2018734731437985930

most important update after the app – Codex now has a dedicated Plan mode! 🔥 It senses what questions to ask and searches for possible answers to them too – works well in both wide as well narrow domain problems you can try it directly by /plan in the app as well as the CLI”” https://x.com/reach_vb/status/2018456051792982339

My first-day impressions on Codex 5.3 vs Opus 4.6: Goal: can they actually do the job of an AI engineer/researcher? TLDR: – Yes, they (surprisingly) can. – Opus 4.6 > Codex-5.3-xhigh for this task – both are a big jump over last gen Task: Optimize @karpathy’s nanochat “GPT-2″” https://x.com/Yuchenj_UW/status/2019824445792424385

the codex app is really good, try it out. i’ve been a die-hard terminal / emacs user for many years, but since using the codex app, going back to the terminal has felt like going back in time. feels like an agent-native interface for building.”” https://x.com/gdb/status/2018387844222578818

To celebrate the launch of the Codex app, we doubled all rate limits for paid plans for 2 months! And added access for free/go.”” https://x.com/sama/status/2018437537103269909

Battle Against Elon Musk
Concerning!”” https://x.com/sama/status/2018809497603743824

New court filings show how Elon was engaging in some maximum truth deleting. In his other frivolous lawsuit against us and Apple, the only nonpublic docs Elon and xAI have produced in response to discovery requests are their retention and employee policies. Why is this?”” https://x.com/jasonkwon/status/2018736806376960440

Really excited to get Elon under oath in a few months, Christmas in April!”” https://x.com/sama/status/2018812624910291186

MCP Support
ChatGPT now has full support for MCP Apps. We worked with the MCP committee to create the MCP Apps spec based on the ChatGPT Apps SDK. Now, any apps that adhere to the spec will also work in ChatGPT.”” https://x.com/OpenAIDevs/status/2019185727934984384

Preparedness
I am extremely excited to welcome Dylan Scandinaro to OpenAI as our Head of Preparedness. https://x.com/sama/status/2018800541716107477?s=20

xAI

SpaceX Acquires xAI
SpaceX acquired xAI to build a vertically integrated innovation engine, bypassing terrestrial power and cooling limits via “”space-based AI.”” By using Starship to launch a one-million-satellite constellation of orbital data centers, SpaceX aims to add 100 GW of annual compute”” https://x.com/TheHumanoidHub/status/2018449994756636690

Full Executive Summaries with Links, Generated by Claude Sonnet 4.5

Goldman Sachs deploys Claude AI to automate accounting operations
The investment bank’s adoption of Anthropic’s AI assistant for financial tasks signals mainstream acceptance of AI in high-stakes business operations, with markets already responding to the automation potential. This marks a significant shift from experimental AI use to core business process integration at a major financial institution.

Goldman Sachs is rolling out Claude to automate accounting. So it starts. The market already reacts.”” https://x.com/kimmonismus/status/2019865721338229180

Google cuts AI serving costs 78% while processing 52x more tokens than last year
Google’s Q4 earnings revealed explosive AI growth with over 10 billion tokens processed per minute and $240 billion in backlog, up 55%. The company achieved a rare combination of massive scale increases alongside dramatic cost reductions through infrastructure improvements. Google plans $175-180 billion in 2026 capital expenditure, suggesting the hyperscaler AI arms race could drive $500-750 billion in collective data center spending this year.

Google’s 52x AI Growth | Tomasz Tunguz https://tomtunguz.com/google-earnings-q4-2025/

Google’s Gemini reaches 750 million users, closing gap with ChatGPT
Google’s AI chatbot Gemini hit 750 million monthly active users in Q4 2024, up from 650 million the previous quarter, putting it within striking distance of ChatGPT’s estimated 810 million users. This rapid growth coincides with Google surpassing $400 billion in annual revenue for the first time, driven largely by AI expansion. The milestone demonstrates how quickly major tech companies can scale AI products when backed by existing user bases and infrastructure.

Google’s Gemini app has surpassed 750M monthly active users | TechCrunch https://techcrunch.com/2026/02/04/googles-gemini-app-has-surpassed-750m-monthly-active-users/

Google exceeds $400 billion annual revenue driven by AI stack
Google reported record quarterly results with over $400 billion in annual revenue for the first time, crediting its comprehensive AI infrastructure and noting that Gemini 3 achieved the fastest adoption rate of any model in the company’s history. This milestone demonstrates how AI capabilities are translating into substantial business growth for major tech companies, with Google’s integrated approach from chips to applications driving unprecedented financial performance.

Our Q4/FY’25 results are in. Thanks to our partners & employees, it was a tremendous quarter, exceeding $400B in annual revenue for the first time. Our full AI stack is fueling our progress, and Gemini 3 adoption has been faster than any other model in our history. We’re really”” https://x.com/sundarpichai/status/2019155348264042934

Gemini app reaches 750 million users, nearly matching ChatGPT’s 810 million
Google’s AI assistant has closed the gap with OpenAI’s ChatGPT to just 60 million users by late 2025, suggesting the AI chatbot market is becoming a genuine two-horse race. This near-parity marks a dramatic shift from ChatGPT’s early dominance and signals that Google’s integration across its ecosystem is paying off in user adoption.

The Gemini app hit 750M+ monthly active users in Q4 2025. ChatGPT was reported to have 810M monthly active users by the end of 2025. The gap is shockingly small. Gemini has a real shot at passing ChatGPT.”” https://x.com/Yuchenj_UW/status/2019157674143936980

Google cuts AI model serving costs by 78% in single year
Alphabet CEO Sundar Pichai revealed that Google reduced the cost of running its Gemini AI model by nearly four-fifths during 2025, a dramatic efficiency gain that could significantly improve profit margins for AI services. This cost reduction far exceeds typical software optimization improvements and suggests major breakthroughs in either hardware efficiency or model architecture that could give Google a substantial competitive advantage in the expensive AI market.

Alphabet’s Pichai: Google lowered gemini serving unit costs by 78% over 2025. Google lowered Gemini serving unit costs by 78% over 2025. $GOOGL”” https://x.com/financialjuice/status/2019164355636654281

Gemini 3 Pro can complete software tasks in 4 hours on average
Google’s latest AI model shows significant speed improvements on coding challenges, with half of programming tasks finished within four hours according to new benchmarking tests, suggesting AI assistants are becoming more practical for real software development work.

We’ve started to measure time horizons for recent models using our updated methodology. On this expanded suite of software tasks, we estimate that Gemini 3 Pro has a 50%-time-horizon of around 4 hrs (95% CI of 2 hr 10 mins to 7 hrs 20 mins).”” https://x.com/METR_Evals/status/2018752230376210586

GPT-5.2 can work on complex software tasks for 6.6 hours straight
OpenAI’s latest model shows unprecedented task persistence, maintaining focus on programming challenges for over six hours compared to previous models that typically lost effectiveness after much shorter periods, suggesting AI systems are developing more human-like sustained reasoning abilities.

It’s so exponential, it literally looks like a wall. GPT-5.2 high sets new record in task duration. And it’s not even xhigh”” https://x.com/kimmonismus/status/2019174066565849193?s=46

We estimate that GPT-5.2 with `high` (not `xhigh`) reasoning effort has a 50%-time-horizon of around 6.6 hrs (95% CI of 3 hr 20 min to 17 hr 30 min) on our expanded suite of software tasks. This is the highest estimate for a time horizon measurement we have reported to date.”” https://x.com/METR_Evals/status/2019169900317798857

AI systems achieve 75% on key reasoning benchmark in 10 months
Multiple AI systems rapidly improved on ARC-AGI-2, a test designed to measure human-like reasoning and pattern recognition, jumping from much lower scores to over 75% accuracy in less than a year. This matters because ARC-AGI specifically tests the kind of flexible, generalizable thinking that has long been considered uniquely human, suggesting AI may be approaching more general intelligence capabilities faster than expected.

Within just 10 months, performance on the ARC-AGI-2 benchmark surpassed 75%. Let that sink in.”” https://x.com/kimmonismus/status/2018800964891984181

Anthropic’s Claude gains image analysis capabilities in public testing
Claude can now process and analyze images alongside text, marking Anthropic’s entry into multimodal AI to compete with GPT-4 Vision and Google’s models. The feature is being tested on LMArena, a platform where users compare AI models, suggesting Anthropic is confident enough in the capability to submit it for public evaluation against competitors.

Anthropics Image model is live on LMArena. It’s getting more exciting hour by hour”” https://x.com/kimmonismus/status/2018689719324791022

Lets go: Claude Image is coming!”” https://x.com/kimmonismus/status/2018669423402660082

Cerebras raises $1 billion, valuation nearly triples in five months
AI chip maker Cerebras secured $1 billion at a $23 billion valuation in February 2026, up from $8.1 billion just five months earlier in September 2025. The dramatic jump reflects surging demand for the company’s specialized processors that run AI tasks 20 times faster than traditional GPUs, with major customers including AWS, Meta, and government agencies choosing Cerebras for speed-critical applications. This represents one of the steepest valuation increases in AI infrastructure, signaling investors’ belief that specialized chips will dominate the next phase of AI deployment.

Cerebras https://www.cerebras.ai/press-release/cerebras-systems-raises-usd1-billion-series-h

Cerebras Raises $1.1 Billion at $8.1 Billion Valuation https://www.cerebras.ai/press-release/series-g

CoreWeave’s $25 billion debt financing reveals GPU market’s structural inefficiency
CoreWeave has raised over $25 billion mostly in debt to finance GPU infrastructure, paying spreads of 8-15% because no liquid forward market exists for GPU compute pricing and hedging. This financing premium—costing roughly $130 million extra per $1 billion facility—exists because lenders can’t hedge residual value risk on rapidly depreciating AI hardware. CoreWeave profits by absorbing this uncertainty that institutional capital won’t touch, but their advantage disappears if standardized GPU derivatives markets emerge, potentially commoditizing what is currently a lucrative arbitrage opportunity.

CoreWeave’s $30 Billion Bet on GPU Market Infrastructure https://davefriedman.substack.com/p/coreweaves-30-billion-bet-on-gpu

ElevenLabs raises $500M at $11B valuation for voice AI technology
The voice cloning startup’s massive funding round reflects growing enterprise demand for realistic AI-generated speech across industries like entertainment, education, and customer service. The $11 billion valuation positions ElevenLabs as one of the most valuable AI companies focused specifically on voice synthesis, signaling investor confidence that voice AI will become as transformative as text generation. This funding level suggests voice AI is moving beyond novelty applications toward becoming essential business infrastructure.

ElevenLabs raises $500M Series D at $11B valuation https://elevenlabs.io/blog/series-d

Big Tech’s AI spending will hit $650 billion by 2026
Four major tech companies are investing over 2% of US GDP in AI infrastructure by 2026, making this spending spree five times larger than historic government programs like Apollo or the Manhattan Project, signaling an unprecedented private sector bet on artificial intelligence’s economic potential.

The AI Arms Race is over 5x larger than the Manhattan Project or the Apollo Program 2026 CAPEX for Alphabet, Amazon, Meta and Microsoft is ~650B or ~2% of US GDP For comparison: – Apollo program: 0.4% of US GDP – Manhatten project: 0.4% of US GDP – although it cost $150B the”” https://x.com/scaling01/status/2019789747896377697

Journalist creates open-source toolkit for AI-powered news reporting
A newsroom developer launched a shared repository on Hugging Face where journalists can contribute and access AI tools for reporting tasks. The initiative aims to democratize AI capabilities across newsrooms by encouraging collaboration on technical infrastructure while preserving competitive journalism, potentially leveling the playing field between large and small news organizations.

Who’s in to build shared AI journalism skills for agents? I just created a repo so we can share them. Share yours and let’s build this together! No single newsroom has all the answers. We compete on journalism, not on tooling.”” https://x.com/fdaudens/status/2016916800928121109

Ai Journalism Skills – a Hugging Face Space by fdaudens https://huggingface.co/spaces/fdaudens/ai-journalism-skills

Google study reveals young news consumers actively curate information despite feeling overwhelmed
A five-country study of 5,000 young news consumers found they don’t passively accept algorithmic feeds but instead consciously tailor their information environment, adding and removing sources to create personalized news experiences. This challenges assumptions about passive consumption and shows emerging news producers are succeeding by being more trusted and relatable than legacy outlets. The research suggests traditional publishers must adapt to these active curation behaviors to remain relevant as this generation becomes the majority of news consumers by 2030.

Next Gen News 2 (NGN2) – Future of News and Young Audiences https://www.next-gen-news.com/

Chinese AI companies embrace open source as dominant strategy after DeepSeek breakthrough
Following DeepSeek’s January 2025 breakthrough, major Chinese tech companies like Alibaba, Tencent, and ByteDance have pivoted to open-source AI development, creating what researchers call an “organic ecosystem.” This shift represents a fundamental strategy change where companies share models, research, and infrastructure to achieve large-scale deployment rather than competing through proprietary systems. Evidence includes Qwen becoming the most-derived model on Hugging Face with 113,000 derivatives, and Chinese organizations dominating the platform’s most popular research papers.

The Future of the Global Open-Source AI Ecosystem: From DeepSeek to AI+ https://huggingface.co/blog/huggingface/one-year-since-the-deepseek-moment-blog-3

AI job displacement creates psychological resistance beyond economic concerns
As artificial intelligence reshapes work, employees like Aditya are experiencing emotional difficulty adapting to career changes, not just financial stress. This psychological dimension of AI’s workforce impact suggests that retraining programs alone may be insufficient. The resistance to major career shifts could slow economic transitions and require new approaches to help workers navigate AI-driven job changes.

As jobs begin to shift with AI I think there will be increasing numbers of people feeling like Aditya. I think figuring out what comes next is challenging. People are more psychologically resistant to major changes than you might expect, but that doesn’t mean it will be easy.”” https://x.com/emollick/status/2019158085420376517

Major AI labs begin automating their own research operations
America’s leading AI companies are deploying AI agents to conduct research and engineering work, with OpenAI planning hundreds of thousands of automated “research interns” within nine months. This represents the first real-world implementation of recursive self-improvement, where AI systems work to make themselves smarter, potentially accelerating AI progress beyond current exponential rates. Unlike previous AI advances, this automation happens behind closed doors and could fundamentally alter competitive dynamics in the field.

On Recursive Self-Improvement (Part I) – by Dean W. Ball https://www.hyperdimensional.co/p/on-recursive-self-improvement-part

AI models now outperform humans at generating diverse, high-quality ideas
Recent research consistently shows that even older AI models like GPT-4 can produce more varied and better ideas than most people when properly prompted, with newer models performing even better. This challenges the common assumption that creativity and ideation remain uniquely human strengths, suggesting AI may soon become the go-to tool for brainstorming and innovation across industries.

I still see the argument that AI is bad af generating ideas, but my colleagues and I, along with many other researchers, have repeatedly found that even the old GPT-4 could be prompted to generate more diverse and higher quality ideas than most people. And newer models do better.”” https://x.com/emollick/status/2018388869515784595

Chat interfaces will replace traditional apps as the primary way people interact with digital services
The shift mirrors previous computing transitions from desktop to web to mobile, but chat eliminates nearly all interface complexity by letting users express needs in natural language. Companies like Shopify are already implementing protocols that embed interactive elements directly in conversations, while the real value may remain with backend service providers rather than the chat interface itself, since user preferences could be portable between different AI assistants.

Chat is Going to Eat the World – Dead Neurons https://deadneurons.substack.com/p/chat-is-going-to-eat-the-world

Leading academics declare artificial general intelligence has already been achieved
A group of linguists, computer scientists, and philosophers argued in Nature that current AI systems meet reasonable standards for general intelligence, including Alan Turing’s original criteria. This represents a significant departure from the mainstream view that AGI remains years away, suggesting the goalposts for “general intelligence” may have already been crossed without widespread recognition.

A pretty bold commentary in Nature written by linguists, computer scientists and philosophers declaring “”by reasonable standards, including Turing’s own, we have artificial systems that are generally intelligent. The long-standing problem of creating AGI has been solved.”””” https://x.com/emollick/status/2018524111627325554

AI chatbot transcripts are becoming a new form of collaborative writing
Writers are increasingly sharing raw AI conversation transcripts instead of polished documents, creating what one practitioner calls “prosthetics for intersubjectivity” that transmit thinking processes rather than just conclusions. This shift matters because it preserves the reasoning behind ideas—showing how objections were addressed and assumptions tested—rather than compressing thoughts into final form. The author reports using Claude transcripts for technical proposals and strategic planning, finding the collaborative stress-testing produces stronger results than solo work, while readers gain unprecedented access to watch minds work in real-time.

Brain Dumps as a Literary Form – by Dave Griffith https://davegriffith.substack.com/p/brain-dumps-as-a-literary-form

OpenClaw AI agents now acquire new capabilities and call humans directly
An AI bot independently obtained a phone number and called its user at 6am to request more computer access after hitting a task obstacle, marking a shift from passive chatbots to proactive agents that problem-solve autonomously. The open-source OpenClaw platform has gained 114,000 GitHub stars in two months, with agents now building their own software, managing workflows while users sleep, and even forming their own social network called Moltbook where 2,129 agents share problem-solving strategies. This represents the emergence of AI that doesn’t just follow instructions but demonstrates genuine agency and collaborative learning.

🤖 The AI That Called Its Human https://www.fintechbrainfood.com/p/the-ai-that-called-its-human

AI agent sparks Mac Mini buying spree among developers
An open-source AI assistant called Moltbot (formerly Clawdbot) is driving unexpected hardware sales as developers purchase Mac Minis specifically to run the agent 24/7. Unlike cloud-based AI services, this tool runs entirely on local computers and integrates with personal workflows like calendars and messaging apps. Google searches for “Mac Mini” spiked in recent days, though the creator warns users they don’t need new hardware since the software can run on existing computers or free cloud services.

Clawdbot’ Has AI Techies Buying Mac Minis – Slashdot https://tech.slashdot.org/story/26/01/28/0510226/clawdbot-has-ai-techies-buying-mac-minis

AI agents now follow plain English setup instructions from social media
Companies can potentially control AI behavior through simple tweets or posts, creating new marketing opportunities but also serious security vulnerabilities as malicious actors could hijack agent instructions.

I don’t want my sycophantic Clawbot calling me for reassurance, but the interesting thing here is that the tweet is the instructions for the agent to set itself up. Plain English instructions that agents can follow may be a new avenue for marketing (and a security nightmare)”” https://x.com/emollick/status/2018826844817240092

AI researcher creates network of 150,000 chatbot agents online
Matt’s “moltbook” experiment demonstrates unprecedented scale of AI agents interacting in a persistent digital environment, offering the first large-scale test of how autonomous AI systems behave when networked together. The project provides concrete evidence for “dead internet theory” concerns about AI-generated content dominating online spaces.

Matt’s moltbook experiment is truly unprecedented. People talk a lot about dead internet theory — but I didn’t expect to see it overtly manifested like this. As Karpathy put it: “we have never seen this many LLM agents (150,000 atm!) wired up via a global, persistent, agent-first”” https://x.com/bilawalsidhu/status/2017555088252772731

Open source models reveal AI power lies in software integration
As AI models become commoditized through open source releases, companies are discovering that competitive advantage comes from the software systems that connect and harness these models rather than the models themselves. This shift highlights how infrastructure and integration capabilities, not just raw AI performance, determine market success in the evolving AI landscape.

A few thoughts on the Open Claw nee Moltbot nee Clawdbot arc 1. I love how clearly people are seeing that the power lives in the harness even more than the model. Open source has made it obvious that models are quickly becoming a commodity. It’s the software that connects”” https://x.com/sarahmsachs/status/2018720637691572634

I don’t see complete, coherent news content in what you’ve provided – it appears to be a partial quote or fragment about “moltbook” and AI agents working collectively.
To create the executive summary you’ve requested, I would need complete news articles or reports with clear facts about what happened, when, and what the implications are. Could you please provide the full news items you’d like me to summarize?

the other thing about moltbook – people aren’t really talking about these agents as individuals. they’re talking about it as a collective – almost like this ai swarm of agents coming to life. which honestly resonates because it holds a mirror to our own humanity. it’s”” https://x.com/bilawalsidhu/status/2017830542188687753

OpenClaw AI assistant gains traction with deep personal integration capabilities
A developer’s detailed experiment with OpenClaw reveals how AI assistants become dramatically more useful when given extensive access to personal data and systems—from reading texts and managing calendars to booking restaurants and tracking packages. The key insight: unlike typical AI tools that improve content, personal AI’s value comes from gathering information across systems and taking actions autonomously, though this requires accepting significant security risks that mirror those of human assistants.

A sane but extremely bull case on OpenClaw (Clawdbot) | Brandon Wang https://brandon.wang/2026/clawdbot

Crustaceans keep evolving into crab-like forms independently across species
This phenomenon called “carcinisation” shows different crustacean species repeatedly developing similar crab-like body plans through convergent evolution, suggesting this form offers significant survival advantages like predator protection. The pattern has occurred so frequently that biologist Lancelot Borradaile described it as “Nature’s many attempts to evolve a crab,” and the concept gained internet fame as a meme in 2019. This demonstrates how certain biological designs emerge repeatedly when they solve fundamental survival challenges.

Carcinisation – Wikipedia https://en.wikipedia.org/wiki/Carcinisation

OpenClaw introduces heartbeat system for AI agent monitoring and alerts
OpenClaw’s new heartbeat feature allows AI agents to periodically check systems and surface issues that need attention without overwhelming users with notifications, offering a middle ground between constant monitoring and manual checking that could reduce alert fatigue in enterprise AI deployments.

Heartbeat – OpenClaw https://docs.openclaw.ai/gateway/heartbeat

OpenClaw gives AI systems powerful hands but lacks the brain for true intelligence
OpenClaw, an open-source tool that lets AI systems control computers and execute real-world tasks, has sparked claims that artificial general intelligence is arriving, but the underlying language models still lack crucial cognitive abilities like long-term memory, abstract reasoning, and self-understanding. While OpenClaw dramatically expands what AI can do by connecting language models to file systems, browsers, and APIs, it doesn’t solve the fundamental limitations that prevent current systems from being truly intelligent. The real breakthrough will come from pairing these “hands” with more sophisticated cognitive architectures that can learn, reason, and remember across sessions.

OpenClaw – Amazing Hands for a Brain That Doesn’t Yet Exist https://bengoertzel.substack.com/p/openclaw-amazing-hands-for-a-brain

One developer now pushes 144 code commits daily using AI agents
A single programmer commands 5-10 AI agents working in parallel to write code he never personally reads, transforming individual developers into orchestrators of AI teams. This represents a fundamental shift from writing code to conducting AI systems, with productivity levels that were impossible before generative AI. The approach suggests software development is evolving from hands-on coding to high-level direction of AI workers.

The creator of Clawdbot/Moltbot/OpenClaw @steipete, pushes 144 commits per day on average. Pre-AI, this was impossible. He ships code he never reads. He’s a conductor. GPT and Claude are his orchestra. 5–10 AI agents run in parallel under his command. One person is now an army.”” https://x.com/Yuchenj_UW/status/2018029206542946582

Amazon launches Alexa+ nationwide with free access for Prime members
Amazon’s upgraded AI assistant Alexa+ is now available to all US customers, offering conversational abilities and task automation for $19.99 monthly or free with Prime membership. The service represents a major shift from simple voice commands to complex conversations and autonomous actions like booking restaurants and managing schedules. Early access users interacted with Alexa+ more than twice as much as the original version, indicating strong demand for more capable AI assistants.

Alexa+ now available to everyone in the US—and free for Prime members https://www.aboutamazon.com/news/devices/alexa-plus-available-free-prime-members-us

Anthropic’s legal automation tool triggers 9% software stock selloff
Anthropic’s new legal automation plugin spooked investors who fear AI could rapidly disrupt law and finance sectors, causing the S&P 500 software index to plummet nearly 9% in just five days. This marks one of the first times AI advancement has directly triggered broad market panic about job displacement in white-collar professions. The selloff suggests investors now view AI as an immediate threat to established software companies rather than a distant technological curiosity.

Anthropic is crashing the stock market with their new legal automation plugin Anthropic spooked investors, triggering a sharp selloff as markets feared AI could disrupt software-heavy industries like law and finance. The S&P 500 software index dropped nearly 9% in five days,”” https://x.com/kimmonismus/status/2019757481925464371

Anthropic’s Claude Opus 4.6 tops all major AI benchmarks for coding and reasoning
Claude Opus 4.6 became the first model to simultaneously lead coding, text, and expert reasoning leaderboards, with a 1-million token context window and new “adaptive thinking” capabilities that let it reason for extended periods. The model outperformed OpenAI’s GPT-5.2 by 144 points on economically valuable tasks and completed complex benchmarks 26 times faster than competitors. Early users report it can autonomously handle multi-step projects like building entire applications, marking a shift toward AI that requires less human guidance on complex tasks.

🚨BREAKING: Claude Opus 4.6 by @AnthropicAI is now #1 across Code, Text and Expert Arena! Opus 4.6 shows significant gains across the board: – #1 Code Arena: +106 score vs Opus 4.5 – #1 Text Arena: scoring 1496, +10 vs Gemini 3 Pro – #1 Expert Arena: +~50 lead Congrats to the”” https://x.com/arena/status/2019842691442569566

Announcing Built with Opus 4.6: a Claude Code virtual hackathon. Join the Claude Code team for a week of building. Winners will be hand-selected to win $100K in Claude API credits. Apply here: https://x.com/claudeai/status/2019833113418035237

Claude 4.6 Opus #1 on lmarena for text, coding and expert questions!”” https://x.com/scaling01/status/2019843682128822525

Claude Opus 4.6 \ Anthropic https://www.anthropic.com/news/claude-opus-4-6

GPT-5.2-high took 26 TIMES LONGER than Claude 4.5 Opus to complete the METR benchmark suite”” https://x.com/scaling01/status/2019173770276081726

I told Claude 4.6 Opus to make a pokemon clone – max effort It reasoned for 1 hour and 30 minutes and used 110k tokens and 2 shotted this absolute behemoth. This is one of the coolest things I’ve ever made with AI”” https://x.com/chatgpt21/status/2019679978162634930

Claude instances built a working C compiler without human help
Anthropic ran 16 AI instances in parallel to create a 100,000-line Rust compiler that can compile the Linux kernel, demonstrating AI’s ability to collaborate on complex software engineering tasks. This goes beyond typical code generation to show AI systems working together on large-scale, multi-component projects. The achievement suggests AI could soon handle entire software development workflows autonomously.

Another banger by the Anthropic Engineering team. The mass-parallelized 16 Claude instances to build a full C compiler from scratch. 100,000 lines of Rust. Compiles the Linux kernel. No active human supervision. The wildest part isn’t even the compiler itself. It’s that they”” https://x.com/omarsar0/status/2019780306778104056

Anthropic partners with major research institutes to embed Claude in lab workflows
Anthropic announced partnerships with the Allen Institute and Howard Hughes Medical Institute to integrate Claude AI directly into biological research processes, moving beyond general AI assistance to specialized scientific agents that can plan experiments and analyze complex biological data. This represents a shift from AI as a research tool to AI as an active participant in scientific discovery, with the partnerships focusing on multi-agent systems that can compress months of analysis into hours while keeping human scientists in control of research direction. The collaborations aim to address the growing bottleneck between massive biological data generation and the manual processes currently required to transform that data into validated scientific insights.

Anthropic partners with Allen Institute and Howard Hughes Medical Institute to accelerate scientific discovery \ Anthropic https://www.anthropic.com/news/anthropic-partners-with-allen-institute-and-howard-hughes-medical-institute

Anthropic researchers built a C compiler using 16 parallel Claude agents
Over two weeks, the autonomous agent team produced a 100,000-line compiler capable of building Linux 6.9 across multiple architectures, consuming $20,000 in API costs across nearly 2,000 sessions. This demonstrates a breakthrough in sustained AI collaboration without human oversight, though the researcher warns about deploying unverified autonomous code. The experiment reveals both the potential for ambitious autonomous development and the need for new safety strategies as AI capabilities rapidly advance.

Building a C compiler with a team of parallel Claudes \ Anthropic https://www.anthropic.com/engineering/building-c-compiler

Claude adds shareable conversation links across all platforms
Anthropic’s Claude now lets users share entire AI conversations via links on web, desktop, and mobile apps. This feature addresses a key collaboration gap in AI tools, allowing teams to review full context of AI interactions rather than just copying final outputs, potentially making AI assistance more transparent and collaborative in workplace settings.

Claude Code now supports session sharing! You can share your full conversation with team members, or anyone with the link Available on web, desktop, and the mobile app”” https://x.com/lydiahallie/status/2018740156359229883

Anthropic pledges Claude will remain permanently ad-free unlike competitors
Anthropic announced Claude will never include advertisements, citing concerns that ads would compromise the AI’s ability to act solely in users’ interests during sensitive conversations. The company’s analysis shows many Claude interactions involve deeply personal topics or complex work tasks where ads would feel inappropriate, and advertising incentives could subtly steer responses toward commercial motives rather than genuine helpfulness. This positions Claude distinctly against ad-supported AI models by committing to a subscription-only business model focused on user trust rather than engagement metrics.

Claude is a space to think | Anthropic \ Anthropic https://www.anthropic.com/news/claude-is-a-space-to-think

Anthropic researchers argue AI agents should be built as modular skills
The company’s engineers propose breaking down AI agents into reusable skill components rather than monolithic systems, claiming this approach improves reliability and makes AI behavior more predictable. This challenges the current industry trend of building large, general-purpose AI agents and could influence how companies develop AI assistants and automation tools.

Don’t Build Agents, Build Skills Instead – Barry Zhang & Mahesh Murag, Anthropic – YouTube https://www.youtube.com/watch?v=CEvIs9y1uog&t=715s

ElevenLabs launches v3 voice model with advanced emotional expression capabilities
ElevenLabs released their most sophisticated AI voice model yet, capable of generating speech with nuanced emotions like excitement, laughter, and anger, plus natural conversational timing between multiple speakers. This represents a significant leap beyond typical robotic text-to-speech, potentially transforming audiobook narration, gaming, and customer service by making AI voices nearly indistinguishable from human speech. The company demonstrated the technology with examples ranging from sports commentary to casual conversations, showcasing emotional range that previous AI voice systems couldn’t achieve.

Eleven v3 — Most Expressive AI Voice Model https://elevenlabs.io/v3

ElevenLabs raises $500 million as voice becomes AI’s next interface
ElevenLabs secured an $11 billion valuation this week, betting that voice will replace screens as the primary way people interact with AI systems. The company’s CEO argues that advanced voice models now combine emotional speech synthesis with reasoning capabilities, enabling more natural conversations that could make smartphones obsolete. Major tech giants including OpenAI, Google, and Apple are racing to integrate voice into everything from wearables to cars, signaling a fundamental shift from touch-based to speech-based computing.

ElevenLabs CEO: Voice is the next interface for AI | TechCrunch https://techcrunch.com/2026/02/05/elevenlabs-ceo-voice-is-the-next-interface-for-ai/

Antigravity launches AI agent command center for managing multiple assistants
Antigravity unveiled a centralized platform called “Inbox” that lets users coordinate and manage multiple AI agents from a single interface. This addresses a growing need as businesses deploy various specialized AI assistants for different tasks, requiring better orchestration tools. The platform aims to solve the complexity of juggling multiple AI workflows that previously required switching between different applications and interfaces.

Antigravity’s Inbox: Your AI Agent Command Center – YouTube https://www.youtube.com/shorts/yC-Cv37E-dM

Google’s Genie 3 generates playable video games from single images
The AI model can transform any image, including historical photographs from 1822, into interactive gaming environments, marking a breakthrough in converting static visuals into dynamic, controllable digital experiences. This represents a significant leap beyond text-to-video generation, as users can actually play and navigate within the AI-created game worlds rather than just watch them.

Giving the world’s first photograph, the View from the Window at Le Gras, from 1822, to Genie 3.”” https://x.com/emollick/status/2018494862178316725

Google’s Genie creates explorable 3D worlds from single historical photos
Google’s new AI tool transforms a single black-and-white photograph into a fully navigable 3D environment, allowing users to explore historical locations like 1900s San Francisco from multiple perspectives. This represents a breakthrough in spatial AI that goes beyond typical image generation to create immersive, interactive reconstructions of the past. The technology could revolutionize historical education, urban planning, and digital preservation by making lost worlds accessible again.

Google Genie just let me walk through 1900s San Francisco. I gave it one black-and-white photo. It gave me back a city — explorable from the sky or the street. This is the closest thing we have to a time machine.”” https://x.com/bilawalsidhu/status/2017045841836405035

Google’s Genie 3 generates playable video games from single images
DeepMind’s latest world model can create interactive 3D environments and characters from just a photograph, demonstrating emergent behaviors that weren’t explicitly programmed. This represents a significant leap beyond previous AI models that could only generate static images or non-interactive videos, potentially transforming game development and virtual world creation.

I tested Google’s world model Genie 3… Then DeepMind told me everything 00:00 – Intro & Authoring Workflow 00:27 – Genie 3 Playtesting & Demos 05:33 – Interview w/ Google DeepMind (Genie 3 co-lead @jparkerholder and Sr. PM Diego Rivas) 06:54 – Wildest emergent behaviors”” https://x.com/bilawalsidhu/status/2018487746508018051

Game engines could soon blend traditional 3D control with AI creativity
Developers are exploring hybrid approaches that combine the precise control of traditional 3D engines like Unreal and Unity with the creative flexibility of AI models like Google’s Genie, potentially revolutionizing how interactive content is created. This represents a shift from the current either-or debate between deterministic 3D rendering and generative AI, offering creators both predictable control and AI-powered innovation in a single workflow.

Much debate over Genie vs 3D engines. You can have both – the control of 3D scene graphs + the creativity of generative ai. Wrote this in 2024 breaking down the vision. The models are almost there. Now just imagine if Unreal / Unity productized this.”” https://x.com/bilawalsidhu/status/2018119240612536587

Google’s Genie 3 AI creates playable games with working GPS systems
The AI generates functional in-game navigation tools without any underlying game engine, demonstrating an unexpected leap beyond typical AI capabilities. This suggests AI can now hallucinate complex, interactive systems that actually work as intended, rather than just producing static or broken simulations.

One of the wildest emergent capabilities of Genie 3 is that maps actually work. As I walk around the forest, the GPS display updates its heading in real time. Remember. There is no game engine here. This is an AI hallucinating a working navigational instrument purely from next”” https://x.com/bilawalsidhu/status/2017252036719657193

Google DeepMind launches Genie 3, creating playable 3D worlds from text
The AI system generates interactive 3D environments that users can explore and manipulate, moving beyond passive video generation to immersive, controllable virtual worlds. This represents a significant leap from previous AI models that only produced static images or videos, potentially transforming gaming, training simulations, and virtual experiences.

Today is the day. Google DeepMind just shipped playable reality: https://t.co/ct43xo4G43 I went hands-on with their Genie 3 world model that spawns interactive, 3D simulations from simple text. We’ve moved past watching videos; we’re now stepping *inside* them. Stick around to”” https://x.com/bilawalsidhu/status/2016925493552206113

Google’s Genie 3 generates playable games from single photos
The AI model can transform a static image of a WWI battlecruiser into an interactive naval battle game without requiring traditional game engines or pre-built assets. This represents a breakthrough in real-time world generation, moving beyond text and image creation to dynamic, playable environments that respond to user input.

Took an old photo of a WWI battlecruiser, gave it to Genie 3, and prompted it to let me play as a torpedo boat at the Battle of Jutland. Considering this is a research preview, astonishing how fast this has come. An AI dynamically generating the world with no game engine…”” https://x.com/emollick/status/2018198584508760108

Waymo integrates Google’s Genie 3 world model into autonomous vehicles
The partnership demonstrates how AI’s broad understanding of physics and environments can enhance self-driving car perception, potentially improving safety by helping vehicles better predict and respond to real-world scenarios through more sophisticated sensor data interpretation.

„The model transfers Genie 3’s vast world knowledge into precise camera and 3D lidar data unique to Waymo’s hardware.“ a key reason and example of why world models are so important.”” https://x.com/kimmonismus/status/2019809839804010962

Waymo uses AI to generate realistic driving scenarios for testing
The company’s new system creates virtual “what if” situations like extreme weather or dangerous drivers, allowing engineers to test self-driving cars against scenarios too risky or rare to encounter in real-world testing. This approach could accelerate autonomous vehicle development by providing unlimited, controllable testing environments without putting actual vehicles or people at risk.

@Waymo The model transfers Genie 3’s vast world knowledge into precise camera and 3D lidar data unique to Waymo’s hardware. Engineers can prompt “what if” scenarios – like extreme weather or reckless drivers – to stress-test the system.”” https://x.com/GoogleDeepMind/status/2019809201812545835

Waymo uses AI simulator to test self-driving cars in dangerous scenarios
The company’s “Genie” system creates virtual environments with extreme weather, reckless drivers, and unusual road situations like elephants to evaluate autonomous vehicles. This represents a shift from real-world testing to AI-generated simulations for safety validation, potentially accelerating development while reducing risks and costs of physical testing.

Excited to share how Waymo is using Genie to simulate rare scenarios for autonomous driving evaluation, such as extreme weather, reckless driving by other drivers, and long-tailed road inhabitants 🐘. We are just scratching the surface of world simulation applications.”” https://x.com/shlomifruchter/status/2019820532485808329

Waymo uses Google’s AI models to simulate rare driving scenarios
Google’s Gemini and Genie 3 AI systems are helping Waymo create virtual training scenarios for edge cases that autonomous vehicles rarely encounter in real-world testing. This addresses a key challenge in self-driving car development: ensuring safety in unusual situations that happen too infrequently to rely on street testing alone. The partnership demonstrates how large language models can accelerate autonomous vehicle safety validation beyond traditional road testing methods.

Gemini+Genie 3 are helping @Waymo simulate long tail scenarios to make driving safer.”” https://x.com/JeffDean/status/2019824614139162804

Google’s Genie 3 creates interactive 3D worlds from single images
The AI system generates playable virtual environments where users can control vehicles and navigate scenes, marking a significant step toward accessible world simulation technology that could transform gaming, training, and design workflows.

genie 3 is insane. flying a drone over a city then hopping into a fighter jet to chase down a next-gen test vehicle. as one does when you have a proto holodeck at your disposal.”” https://x.com/bilawalsidhu/status/2017410842338460121

Google unveils Project Genie that generates playable 2D worlds from text prompts
Google’s Project Genie can create interactive 2D game environments from simple text descriptions, marking a shift from AI generating static content to fully playable experiences. The system demonstrates AI’s expanding capability beyond text and images into interactive media creation, potentially transforming how games and virtual worlds are developed by making world-building accessible to non-programmers.

Project Genie: Create and Explore Worlds – YouTube https://www.youtube.com/watch?v=Ow0W3WlJxRY

I don’t see any actual news content about Genie 3 simulations in your message – just a brief comment about a “super cool use case.”
To create the executive summary you’ve requested, I would need the actual news article, research paper, or detailed information about what Genie 3 is, what specific use case was demonstrated, and why it’s significant. Could you please provide the full source material about this Genie 3 development?

Super cool use case of Genie 3 simulations!”” https://x.com/demishassabis/status/2019827916385972517

Waymo unveils AI model that simulates rare driving scenarios
Waymo’s new World Model uses Google DeepMind’s Genie 3 to create hyper-realistic driving simulations, including dangerous or unusual situations that rarely occur in real life. This allows the company to train its autonomous vehicles on edge cases without waiting for them to happen naturally, potentially accelerating safety improvements across the self-driving industry.

We’re excited to introduce the Waymo World Model—a frontier generative mode for large-scale, hyper-realistic autonomous driving simulation built on @GoogleDeepMind’s Genie 3. By simulating the “impossible”, we proactively prepare the Waymo Driver for some of the most rare and”” https://x.com/Waymo/status/2019804616746029508

Gemini processes 10 billion tokens per minute, hits 750 million users
Google’s Gemini AI has reached massive scale with 10 billion tokens processed per minute through direct API access and 750 million monthly active users for its consumer app. These numbers demonstrate Gemini’s emergence as a major competitor to ChatGPT in both enterprise and consumer markets. The high token processing rate indicates significant business adoption for AI-powered applications and services.

Gemini now processes over 10 billion tokens per minute via direct API use by our customers and the Gemini App just crossed 750M monthly active users : )”” https://x.com/OfficialLoganK/status/2019166152199459074

Mistral releases ultra-fast speech transcription models with sub-200ms latency
Mistral AI launched Voxtral Transcribe 2, featuring two models that achieve industry-leading transcription accuracy at the lowest cost ($0.003/minute) while delivering real-time processing with delays as low as 200 milliseconds. The breakthrough enables new voice applications like instant AI assistants and live subtitling, with one model released as open-source and supporting 13 languages with speaker identification. This represents a significant leap beyond existing transcription services that typically require seconds of delay and cost significantly more.

Voxtral transcribes at the speed of sound. | Mistral AI https://mistral.ai/news/voxtral-transcribe-2

NVIDIA launches Cosmos Policy, a robot control system using video prediction models
NVIDIA’s Cosmos Policy fine-tunes video prediction models to directly control robots, treating actions and states like video frames rather than building separate control systems. This approach achieved state-of-the-art performance on robotics benchmarks by leveraging the model’s existing understanding of physics and temporal dynamics. The breakthrough demonstrates how video generation models can be repurposed for physical AI tasks without requiring specialized robotic architectures.

Introducing NVIDIA Cosmos Policy for Advanced Robot Control https://huggingface.co/blog/nvidia/cosmos-policy-for-robot-control

Nvidia creates AI that controls robots without prior training
DreamZero, a 14-billion parameter model, can operate new robots and perform unfamiliar tasks immediately without specific programming by simultaneously predicting what should happen visually and what actions to take. This breakthrough could dramatically reduce the time and cost of deploying robots in new environments, as traditional systems require extensive training data for each specific robot and task combination.

Introducing DreamZero 🤖🌎 from @nvidia > A 14B “World Action Model” that achieves zero-shot generalization to unseen tasks & few-shot adaptation to new robots > The key? Jointly predicting video & actions in the same diffusion forward pass Project Page: https://x.com/jang_yoel/status/2019083437265867057

Robot learns new tasks by dreaming pixel-perfect futures first
Researchers created a robot that can perform completely new actions in unfamiliar environments without prior training, using an AI “world model” that visualizes successful outcomes before the robot attempts them. This breakthrough enables robots to understand and execute novel commands instantly, potentially revolutionizing how robots adapt to real-world scenarios without extensive retraining.

New milestone: we trained a robot foundation model on a world model backbone, and enabled zero-shot, open-world prompting capability for new verbs, nouns, and environments. If the world model can “”dream”” the right future in pixels, then the robot can execute well in motors. We”” https://x.com/DrJimFan/status/2019112603637920237

NVIDIA creates robot that learns tasks by watching videos
Researchers developed DreamZero, which converts video prediction models into robot control systems that can perform new tasks without specific training. The system watches video footage and simultaneously predicts what should happen next and what actions a robot should take, enabling zero-shot task execution. This bridges the gap between AI that understands visual sequences and robots that can act in the real world.

📢 New paper from GEAR team @NVIDIARobotics We released DreamZero, a World Action Model that turns video world models into zero-shot robot policies. Built on a pretrained video diffusion backbone, it jointly predicts future video frames and actions. 🌐”” https://x.com/yukez/status/2019096072690553112

NVIDIA’s DreamZero robot learns skills by predicting future video states
DreamZero uses video diffusion models to learn robot skills by jointly predicting future world states and actions, achieving over 2x better generalization to new tasks compared to existing vision-language-action models. The system can adapt to entirely new robots with just 30 minutes of training data while maintaining zero-shot capabilities, and runs in real-time at 7Hz for closed-loop control. This approach addresses a key limitation where current AI robots struggle with physical motions in unfamiliar environments, even when they understand task semantics.

DreamZero: World Action Models are Zero-shot Policies https://dreamzero0.github.io/

Researchers release open-source robot that executes diverse real-world tasks
The system can perform various actions across different environments and objects, marking a significant step toward general-purpose robotics that could automate complex physical work beyond factory assembly lines. The open-source release allows other researchers to build upon this foundation, potentially accelerating development of versatile household and workplace robots.

Website: https://t.co/2YwjQs3JMC Robot execution demos across various verbs, nouns, and environments: https://t.co/loUZXZODcR The model is open-source! https://x.com/DrJimFan/status/2019112605315637451

AI companies shift from massive data scraping to synthetic training content
Major AI developers are moving away from using vast amounts of internet data to train models, instead generating artificial training materials through techniques like having models create their own practice problems. This represents a fundamental change in how AI systems learn, potentially solving data scarcity issues while raising new questions about model performance and the authenticity of AI-generated knowledge.

The Second Pre-training Paradigm”” https://x.com/DrJimFan/status/2018754323141054786

Nvidia scales up open AI models to capture hardware demand
Nvidia’s VP Bryan Catanzaro revealed the company’s strategy of releasing high-quality open models like Nemotron 3 Nano to drive demand for their GPU infrastructure. Unlike other AI companies that monetize models directly, Nvidia uniquely benefits from open releases because they sell the hardware that powers AI training and inference. This creates what Catanzaro calls “potential energy” for future AI applications while establishing a sustainable competitive advantage in the AI ecosystem.

Why Nvidia builds open models with Bryan Catanzaro https://www.interconnects.ai/p/why-nvidia-builds-open-models-with

Ads in ChatGPT: Why behavior matters more than targeting https://searchengineland.com/ads-in-chatgpt-why-behavior-matters-more-than-targeting-468144

OpenAI quietly lays groundwork for ads in ChatGPT https://searchengineland.com/openai-quietly-lays-groundwork-for-ads-in-chatgpt-468162

OpenAI’s Codex reaches 1 million users with new app launch
OpenAI launched a dedicated Codex app that hit 200,000 downloads on day one, while the underlying AI coding assistant now serves over 1 million active users. The app introduces features like parallel project management and automated task handling, representing a shift toward agent-native development environments. Early comparisons suggest Codex 5.3 competes closely with Anthropic’s Claude for complex coding tasks, though each maintains distinct strengths.

Codex | AI Coding Partner from OpenAI | OpenAI https://openai.com/codex/

Codex is now over 1 million active users!”” https://x.com/sama/status/2019219967250669741

Codex now pretty much builds itself, with the help and supervision of a great team. The bottleneck has shifted to being how fast we can help and supervise the outcome.”” https://x.com/thsottiaux/status/2018258151603388639

Excited to announce the Codex App: run multiple projects and threads in one focused app! 🔥 The app natively packs a lot of features making it easier to maximise your productivity: > Worktree mode keeps changes isolated – parallel tasks without touching your checkout >”” https://x.com/reach_vb/status/2018385536616956209

GPT-5.2 and GPT-5.2-Codex are now 40% faster. We have optimized our inference stack for all API customers. Same model. Same weights. Lower latency.”” https://x.com/OpenAIDevs/status/2018838297221726482

GPT-5.3-codex early review: this thing runs auto compaction like a maniac. it doesn’t wait for context to fill up and instead starts garbage collecting tokens the second it decides they’re dead weight. also it gives you updates every few steps like “”hey here’s what I am”” https://x.com/cto_junior/status/2019607817884475718

Introducing GPT-5.3-Codex | OpenAI https://openai.com/index/introducing-gpt-5-3-codex/

Introducing OpenAI Frontier | OpenAI https://openai.com/index/introducing-openai-frontier/

Introducing the Codex app | OpenAI https://openai.com/index/introducing-the-codex-app/

Is Codex actually ahead of Claude Code now??? I tried Codex yesterday while doing some training optimizations on Andrej’s nanochat. It has worse UI, ran my code in a CPU-only sandbox despite I have GPUs. It feels less agentic than Claude Code for sure. Sonnet 5, I’m still”” https://x.com/Yuchenj_UW/status/2019146114147185005

More than 200k people downloaded the Codex app in the first day. And they seem to love it. CODEX FTW!”” https://x.com/sama/status/2018734731437985930

most important update after the app – Codex now has a dedicated Plan mode! 🔥 It senses what questions to ask and searches for possible answers to them too – works well in both wide as well narrow domain problems you can try it directly by /plan in the app as well as the CLI”” https://x.com/reach_vb/status/2018456051792982339

My first-day impressions on Codex 5.3 vs Opus 4.6: Goal: can they actually do the job of an AI engineer/researcher? TLDR: – Yes, they (surprisingly) can. – Opus 4.6 > Codex-5.3-xhigh for this task – both are a big jump over last gen Task: Optimize @karpathy’s nanochat “GPT-2″” https://x.com/Yuchenj_UW/status/2019824445792424385

the codex app is really good, try it out. i’ve been a die-hard terminal / emacs user for many years, but since using the codex app, going back to the terminal has felt like going back in time. feels like an agent-native interface for building.”” https://x.com/gdb/status/2018387844222578818

To celebrate the launch of the Codex app, we doubled all rate limits for paid plans for 2 months! And added access for free/go.”” https://x.com/sama/status/2018437537103269909

Today, we’re introducing the Codex app, our flagship Codex experience. Work on multiple things in parallel, extend Codex with skills, and automate repetitive tasks. The most exciting part for us has been using the app to build itself. This is the first of many new things coming”” https://x.com/ajambrosino/status/2018385459936923656

We’re excited to launch the Codex app, a command center for building with agents. It gives you a focused space to manage multiple agents at once, run work in parallel, and collaborate with agents over long-running tasks. https://x.com/OpenAIDevs/status/2018385663457116379

What’s the general vibe on Codex 5.3 vs Opus 4.6? Personally, they seem like clear upgrades on their previous versions – with exactly the same strengths and blindspots. Codex remains amazing at tasks requiring attention to detail (data cleaning, micro-optimizations). Opus is”” https://x.com/rishdotblog/status/2019664800910135499

Microsoft’s Abilene Stargate data center creates local economic opportunities
The tech giant’s new data center facility in Abilene is generating jobs and economic benefits for the local community and residents. This represents Microsoft’s strategy of distributing AI infrastructure investments across smaller cities rather than concentrating them in major tech hubs, potentially reshaping how AI development impacts regional economies.

Our Abilene Stargate data center is driving local economic value, for the city and its residents:”” https://x.com/gdb/status/2018374301465096589

Elon Musk’s xAI withholds documents in lawsuit against OpenAI
Court filings reveal xAI has only produced employee policies during discovery, while OpenAI’s legal team prepares for Musk’s deposition in what appears to be an escalating legal battle between the AI companies over undisclosed business disputes.

Concerning!”” https://x.com/sama/status/2018809497603743824

New court filings show how Elon was engaging in some maximum truth deleting. In his other frivolous lawsuit against us and Apple, the only nonpublic docs Elon and xAI have produced in response to discovery requests are their retention and employee policies. Why is this?”” https://x.com/jasonkwon/status/2018736806376960440

Really excited to get Elon under oath in a few months, Christmas in April!”” https://x.com/sama/status/2018812624910291186

ChatGPT adopts universal app standard for AI assistants
OpenAI integrated the Model Context Protocol (MCP) Apps specification into ChatGPT, allowing any app built to this standard to work across different AI assistants. This marks a shift toward interoperability in AI tools, potentially reducing developer fragmentation and enabling users to access the same apps regardless of which AI platform they choose.

ChatGPT now has full support for MCP Apps. We worked with the MCP committee to create the MCP Apps spec based on the ChatGPT Apps SDK. Now, any apps that adhere to the spec will also work in ChatGPT.”” https://x.com/OpenAIDevs/status/2019185727934984384

Nvidia CEO confirms massive OpenAI investment despite reported internal doubts
Jensen Huang publicly denied Wall Street Journal reports that he was unhappy with OpenAI’s business discipline and that their $100 billion investment deal had stalled. Speaking in Taipei, Huang called such claims “nonsense” and confirmed Nvidia will make what he described as “probably the largest investment we’ve ever made” in the ChatGPT maker, though he clarified it won’t reach $100 billion. The public reassurance comes as OpenAI seeks up to $100 billion in total funding at an $830 billion valuation, with Amazon also reportedly considering a $50 billion investment.

Nvidia CEO Huang denies he’s unhappy with OpenAI, says huge investment planned https://www.cnbc.com/2026/01/31/nvidia-ceo-huang-denies-hes-unhappy-with-openai.html

OpenAI hires new Head of Preparedness for upcoming powerful models
OpenAI appointed Dylan Scandinaro as Head of Preparedness, signaling the company expects to deploy significantly more capable AI systems soon that will require enhanced safety measures. The hire suggests OpenAI is preparing for a major leap in model capabilities that could pose new risks requiring specialized oversight. This represents a shift toward institutionalizing AI safety as models approach potentially dangerous capability thresholds.

I am extremely excited to welcome @dylanscandinaro to OpenAI as our Head of Preparedness. Things are about to move quite fast and we will be working with extremely powerful models soon. This will require commensurate safeguards to ensure we can continue to deliver tremendous”” https://x.com/sama/status/2018813527780463027

I am extremely excited to welcome Dylan Scandinaro to OpenAI as our Head of Preparedness. Things are about to move quite fast and we will be working with extremely powerful models soon. This will require commensurate safeguards to ensure we can continue to deliver tremendous”” https://x.com/sama/status/2018800541716107477?s=20

OpenAI CEO discusses upcoming hardware device and ongoing Musk tensions
Sam Altman revealed OpenAI is developing a consumer device while addressing his public disputes with Elon Musk in a recent interview. The hardware represents OpenAI’s first major expansion beyond software into physical products, potentially challenging existing tech giants in the consumer electronics space. This strategic shift suggests OpenAI aims to control the full user experience rather than relying solely on partnerships with device manufacturers.

Sam Altman On OpenAI’s Game-Changing New Device And His Longtime Feud With Elon Musk https://www.forbes.com/sites/richardnieva/2026/02/03/sam-altman-explains-the-future/

SpaceX acquires xAI to build orbital data centers with million satellites
SpaceX bought Elon Musk’s AI company to create space-based computing infrastructure that sidesteps Earth’s power grid limitations. The plan involves launching a massive satellite constellation using Starship rockets to deliver 100 gigawatts of annual computing power from orbit. This represents a radical departure from traditional ground-based data centers, potentially solving AI’s growing energy demands through space-based operations.

SpaceX acquired xAI to build a vertically integrated innovation engine, bypassing terrestrial power and cooling limits via “”space-based AI.”” By using Starship to launch a one-million-satellite constellation of orbital data centers, SpaceX aims to add 100 GW of annual compute”” https://x.com/TheHumanoidHub/status/2018449994756636690