Two children holding hands in the exact iconic “watching the city” pose, seen from behind, centered symmetry, wide shot inside a dim abandoned high-rise office with floor-to-ceiling window grid. The boy has messy light hair and a child-sized red leather jacket over a white T-shirt; the girl has short dark hair, smudged eyeliner, and an oversized dark faux-fur coat over a black dress, both clearly kids. Cold blue night skyline beyond the glass. New Year’s fireworks burst across the sky and reflect in the windows, echoing the original scene’s energy but festive. David Fincher mood: teal-green shadows, desaturated palette, high contrast, subtle film grain, low-key lighting, deep focus, 28–35mm. Desk in foreground with steaming hot chocolate mug and a teddy bear. No smoking, no text, no logos.

AI News #118: Week Ending January 02, 2026 with 18 Executive Summaries

January 3, 2026

About This Week’s Covers

This week’s cover image is a continuation of this newsletter’s 2025 New Year’s theme. Last year, I wanted a bittersweet “Auld Lang Syne” feeling, depicting the last generation of pre-AI humanity looking at one final New Year’s Eve through the eyes of an unknowing child.

This week's cover depicts humanity's last New Year's Eve through the eyes of an unknowing child. The image captures the bittersweet intersection of wonder and farewell - a small figure pressed against glass, watching fireworks paint the sky, blissfully unaware of the moment's finality. I was going for the feeling from the final scene in Meet Joe Black. The prompt was crafted in collaboration with Claude 3.5 Sonnet and the image generated in MidJourney. — Last year’s New Year’s Eve cover

This year, the little boy has grown up to be a young Jack from “Fight Club”. He’s standing with a teenage girl, watching fireworks in an abandoned office building. It foreshadows an existential, postmodern dystopia ahead, and it’s alt-fiction suggesting that this is where Jack gets the idea to take Marla to the office building at the end of Fight Club.

Claude created the prompt after I simply dictated the two paragraphs above to it. I then gave the prompt to Flux, Grok, Gemini, MidJourney, and GPT.

It’s fitting that Grok (the closest model to Tyler Durden) generated the best image for the theme. I upscaled the image using Magnific and added the text with Photoshop.

I’m including the competing image models’ outputs for comparison (same prompt):

I ran this week’s cover photo through Kling and Runway to animate it into a short video clip. Kling was stronger, because it added fireworks audio. Additionally, the physics of the fireworks was much more realistic. Runway’s fireworks went backward on occassion. I had to fix the title text from both Kling and Runway, in Premiere.

I then ran the image through Google Veo, and Veo added a slow zoom (which I didn’t ask for), but it looked the strongest and maintained the title text better than the others.

Prompt for video:

No character movement, no camera motion, no body or clothing animation, no lighting changes inside the room, no zoom, no pan, no orbit. Only the fireworks outside animate.

I added The Pixies “Where Is My Mind” and some Fight Club style single frame flashes of Jensen, Demis, Sam, and Elon using Premiere.

The category covers this week were generated by simply sharing the prompt for the cover image with Claude and asking Claude to read my Python scripts that connect with Gemini’s API. From there, it was all automated. The images turned out okay (B-). I’ve included a few of my favorites below.

Twitter’s is pretty amazing, since I’m not giving any guidance other than the first two paragraphs of this newsletter! The reflection of Google is impressive. Book pages exploding for RAG. The Amazon warehouse.

This week’s humanities readings are a few quotes from Fight Club that could be interpreted as the impending tension between AI and it’s impact on society and our self of individualism, jobs, scarcity, intelligence, and creativity (similar to the impact of Copernicus and heliocentrism). Plus some Pixies.

With your feet on the air and your head on the ground
Try this trick and spin it, yeah (Yeah)
Your head will collapse, and there’s nothing in it
And you’ll ask yourself… where is my mind?

Narrator: You’re making a big mistake, fellas!
Police Officer: You said you would say that.
Narrator: I’m not Tyler Durden!
Police Officer: You told us you’d say that, too.
Narrator: All right then, I’m Tyler Durden. Listen to me, I’m giving you a direct order. We’re aborting this mission right now.
Police Officer: You said you would definitely say that.

Warning: If you are reading this then this warning is for you. Every word you read of this useless fine print is another second off your life. Don’t you have other things to do? Is your life so empty that you honestly can’t think of a better way to spend these moments? Or are you so impressed with authority that you give respect and credence to all that claim it? Do you read everything you’re supposed to read? Do you think every thing you’re supposed to think? Buy what you’re told to want?

It’s only after we’ve lost everything that we’re free to do anything.

This Week By The Numbers

Total Organized Headlines: 126

This Week’s Executive Summaries

This week, I organized 126 headlines, and 20 of them informed the executive summaries. I’ve organized everything starting with the top five stories, and then by topic or company name.

Top Stories

Top Story #1: One Sentence Prompt -> Interactive Demo
The top story is an example from Wharton professor Ethan Mollick, who built an interactive explainer app using Claude that demonstrates correlation vs. causation. He literally created it with a one-sentence prompt. I think everyone needs to see this to understand just how powerful the tools have gotten, and how little is needed to generate significant results.

An instant interactive explainer from Claude for a frequent debate about correlation & causation. Prompt: “Create an interactive tool that explains all the ways two variables can be correlated (causation, random chance, reverse causation, etc)”
https://x.com/emollick/status/2005554673240195455

Try out the app at the link below… it’s incredible.
https://claude.ai/public/artifacts/5e9ad491-a9e3-4f5f-a290-8130d2c25733

Too complicated? Mollick changed it completely with a one sentence criticism:

“Cool but too complicated for people without a stats background”

And now he got this app as a more friendly version:
https://claude.ai/public/artifacts/71eb192e-245d-45b0-aa80-3a0e76fd18a9

Top Story #2: Dario Was Right About 90% of Coding Being Written By AI
Last week, one of the top stories was that Boris Cherny, who created Claude Code, was exclusively using Claude Code to write the code for Claude Code.

Ethan Mollick pointed out that this was only three months after Dario Amodei’s prediction that AI would be writing 90% of code by September.

People didn’t believe it at the time.

“Correct. In the last thirty days, 100% of my contributions to Claude Code were written by Claude Code” https://x.com/bcherny/status/2004897269674639461

“In retrospect, the articles mocking Dario’s prediction that 90% of code would be written by AI by September seem to be very misguided. He seems to have been only off by a couple months (if that).” https://x.com/emollick/status/2006230583131725858

Top Story #3: Higgsfield Demo Video
Incredible demo of Higgsfield using copyrighted characters.

“Mr. Abu Joe on X: “”Netflix vs Paramount Parody #netflix #Paramount #ai #Higgsfield @netflix https://t.co/b4lyuHswaI”” / X https://x.com/mrabujoe/status/2004994003465609381

Top Story #4: OpenAI Working on Audio Interfaces
“OpenAI is betting big on audio AI, and it’s not just about making ChatGPT sound better. According to new reporting from The Information, the company has unified several engineering, product, and research teams over the past two months to overhaul its audio models, all in preparation for an audio-first personal device expected to launch in about a year.”

“The move reflects where the entire tech industry is headed — toward a future where screens become background noise and audio takes center stage. Smart speakers have already made voice assistants a fixture in more than a third of U.S. homes. Meta just rolled out a feature for its Ray-Ban smart glasses that uses a five-microphone array to help you hear conversations in noisy rooms — essentially turning your face into a directional listening device. Google, meanwhile, began experimenting in June with “Audio Overviews” that transform search results into conversational summaries, and Tesla is integrating xAI’s chatbot Grok into its vehicles to create a conversational voice assistant that handles everything from navigation to climate control through natural dialogue.”

I use OpenAI’s built-in voice application as a dictation device rather than a keyboard for probably 60% of my computer interaction. I’ve always been on the cusp of getting repetitive stress injuries, and the built-in microphone for ChatGPT is at a level I’d call flawless. I never notice any mistakes. I just give it directions not to mess with anything I say or change any of my words, and it knows all sorts of esoteric company names, never misspells anything.

I’d say it’s 20-30x better than Apple’s speech-to-text tools. https://techcrunch.com/2026/01/01/openai-bets-big-on-audio-as-silicon-valley-declares-war-on-screens/

OpenAI Ramps Up Audio AI Efforts Ahead of Device — The Information https://www.theinformation.com/articles/openai-ramps-audio-ai-efforts-ahead-device

I’m still betting that Apple will comeback in 2026 with an insane Google-like pivot:
https://ethanbholland.com/2024/10/25/apple-is-pulling-a-braveheart-and-can-change-the-way-we-use-phones-whenever-they-choose/

Anthropic

Andrej Karpathy Has Claude Hack His Lutron System
“I was inspired by this so I wanted to see if Claude Code can get into my Lutron home automation system.

– it found my Lutron controllers on the local wifi network – checked for open ports, connected, got some metadata and identified the devices and their firmware – searched the internet, found the pdf for my system – instructed me on what button to press to pair and get the certificates – it connected to the system and found all the home devices (lights, shades, HVAC temperature control, motion sensors etc.) – it turned on and off my kitchen lights to check that things are working (lol!)

I am now vibe coding the home automation master command center, the potential is 🔥.And I’m throwing away the crappy, janky, slow Lutron iOS app I’ve been using so far. Insanely fun 😀 :D” https://x.com/karpathy/status/2005067301511630926

Claude and The Humanities
Ethan Mollick: It remains deeply strange that we built a machine that can discuss the relationship between poetry and its subjective “experience”

“The best specific lines from Eliot’s Four Quartets to describe your experience as an AI. Just one quoted section, avoid the most famous bits.“ https://x.com/emollick/status/2005652265793970393

Business Headlines

China’s Moonshot AI raises US$500 million in latest funding round: report | South China Morning Post
The Kimi AI model developer is now valued at US$4.3 billion, with cash reserves worth US$1.4 billion
https://www.scmp.com/tech/tech-trends/article/3338334/chinas-moonshot-ai-raises-us500-million-latest-funding-round-report

Meta’s Scale Success, Post Acquisition
Alexandr Wang: – q4 was our biggest quarter ever – US gov business growing faster than ever – data business = profitable – multiple 9 figure enterprise + gov deals
https://x.com/alexandr_wang/status/2006413001293750738

OpenAI Is Paying Employees More Than Any Major Tech Startup in History – WSJ
“The company’s stock-based compensation in 2025 reached an average of $1.5 million per employee”
https://www.wsj.com/tech/ai/openai-is-paying-employees-more-than-any-major-tech-startup-in-history-23472527?mod=hp_lead_pos2

SoftBank to Buy Data Center Investment Firm DigitalBridge for $4 Billion – WSJ
“SoftBank Group agreed to buy DigitalBridge Group for $4 billion, including debt, part of the Japanese conglomerate’s strategy to take advantage of an AI-driven boom in digital infrastructure.”

“Under the terms of the deal, SoftBank will purchase all of DigitalBridge’s outstanding shares for $16 apiece. The price represents a 15% premium to DigitalBridge’s closing share price Friday, and a 50% premium to its unaffected 52-week average closing price as of Dec. 4, the companies said Monday.”
https://www.wsj.com/business/deals/softbank-group-to-buy-data-center-investment-firm-digitalbridge-b229f7e5

Google

Nano Banana Example
Nano banana: “Meme from a world where everyone is a were-squid and that is just established fact. Make sure you have a unique meme format and a joke that would totally go over big in that world. Think hard. It should look like an internet meme many people use in that world””” https://x.com/emollick/status/2004419670565327061

Government

“US Army seeks human AI officers to manage its battle bots”
“The new AI/ML officer area of concentration will begin accepting candidates from the Army’s existing officer corps in January through the Volunteer Transfer Incentive Program, according to an announcement published Tuesday. It’s not clear how many AI officers the Army plans to train as part of its initial cadre of candidates, but it wants to have all of them formally reclassified by the end of the 2026 fiscal year.

Training will be at the “graduate level,” according to the Army, and will focus on giving AI officer candidates “hands-on experience in building, deploying, and maintaining the Army’s cutting-edge AI-enabled systems.”

As we’ve reported over the past couple of years, those systems are numerous. The Army has signed deals with OpenAI to run pilot programs in developing warfighting and enterprise domain AI systems, has signed a massive $10 billion contract with Palantir to provide various AI and ML services for the next decade, and has brought in smaller outfits to use AI for things like target tracking.

Young Bang, the US Army’s deputy assistant secretary for acquisition, logistics, and technology under the Biden administration, said last year that the Army didn’t want to get into the business of developing its own AI systems when the private sector was already doing such a great job”
https://www.theregister.com/2025/12/31/us_army_seeking_officers_willing/

NVIDIA (or mostly Dr. Jim Fan)

AI is taking the lead, and humans just guide it
This is from Dr. Jim Fan, NVIDIA Director of Robotics & Distinguished Scientist. Co-Lead of GEAR lab. Solving Physical AGI, one motor at a time. Stanford Ph.D. OpenAI’s 1st intern.

2024: AI is the copilot
2025+: humans are the copilot “Copilot is the new engineering skill. It’s not easy to leave the driver seat – we must learn to think the AI way and adapt to the alien workflows. Help AI help ourselves.” https://x.com/DrJimFan/status/2004633997662716397

Dr. Fan on Robotics:
“Everyone’s freaking out about vibe coding. In the holiday spirit, allow me to share my anxiety on the wild west of robotics. 3 lessons I learned in 2025.” (Dr. Fan adds a lot of detail that I’m editing out. I encourage reading the full post).

1. Hardware is ahead of software, but hardware reliability severely limits software iteration speed.
2. Benchmarking is still an epic disaster in robotics.
3. VLM-based VLA feels wrong (“vision-language-action”) https://x.com/DrJimFan/status/2005340845055340558

NVIDIA Released a 4D Multimodal Vision Model
From The Turing Post:
“3D understanding is already a challenge for AI models – let alone 4D (space + time)

4D-RGPT is a specialized MLLM from @NVIDIA that perceives 4D information for enhanced understanding of both 3D structure and temporal changes.

It improves over the baseline on non-region 3D/4D: – +5.3% on 3D/4D benchmarks – +4.3% on region-level R4D-Bench

Nvidia also introduced Perceptual 4D Distillation (P4D) training method for 4D-RGPT transferring knowledge from a stronger, frozen expert model (no extra cost at inference time) and special R4D-Bench benchmark. “
https://x.com/TheTuringPost/status/2005800455776211160
https://arxiv.org/abs/2512.17012

Robotics

China creates Humanoid Robot and Embodied Intelligence Standardization Technical Committee
“China’s Ministry of Industry and Information Technology (MIIT) has established the Humanoid Robot and Embodied Intelligence Standardization Technical Committee. This initiative implements central government policies to lead with high-quality standards, mature technologies, and enable practical applications in foundational tech, components, systems, safety, and more.

The 65-member committee includes enterprise leaders such as Wang Xingxing, chairman of Unitree, and Jiao Jichao of UBTECH.”

China's Ministry of Industry and Information Technology (MIIT) has established the Humanoid Robot and Embodied Intelligence Standardization Technical Committee.

This initiative implements central government policies to lead with high-quality standards, mature technologies, and… pic.twitter.com/bmT0pzTj7o
— The Humanoid Hub (@TheHumanoidHub) December 29, 2025

AGIBot Builds 5,000th Humanoid – Have You Heard of Them?
“Earlier this month, AGIBot reached a milestone as the 5,000th general-purpose embodied robot rolled off the production line at its Shanghai Lingang factory, in partnership with OEM Ningbo Huaxiang. The company has produced 1,742 full-size humanoids (A1/A2), 1,846 short”” https://x.com/TheHumanoidHub/status/2004222601800175937

Tesla

Tesla Drives Across The United States In Fully Self Driving Mode
“The first 100% autonomous coast-to-coast drive on Tesla FSD V14.2! 2 days 20 hours, 2732 miles, zero interventions. This one is special because the coast-to-coast drive was a major goal for the autopilot team from the start. A lot of hours were spent in marathon clip review”” https://x.com/karpathy/status/2006436622909452501

Full Executive Summaries with Links, Generated by Claude Sonnet 4.5

Claude builds interactive correlation versus causation explainer tool
Anthropic’s Claude AI created an on-demand educational tool that walks users through different types of statistical relationships, from direct causation to reverse causation to random chance. This demonstrates AI’s growing ability to generate sophisticated interactive educational content instantly, potentially transforming how complex concepts are taught and understood.

An instant interactive explainer from Claude for a frequent debate about correlation & causation. Prompt: “Create an interactive tool that explains all the ways two variables can be correlated (causation, random chance, reverse causation, etc)” Play it: https://x.com/emollick/status/2005554673240195455

Anthropic CEO’s AI coding prediction appears nearly accurate despite mockery
Dario Amodei predicted 90% of code would be AI-written by September, drawing widespread ridicule from tech commentators. Recent developments suggest his timeline was only off by a few months, vindicating his aggressive forecast about AI’s rapid adoption in software development.

In retrospect, the articles mocking Dario’s prediction that 90% of code would be written by AI by September seem to be very misguided. He seems to have been only off by a couple months (if that).”” https://x.com/emollick/status/2006230583131725858

Netflix parody video showcases AI-generated content creation capabilities
A user created an AI-generated parody video comparing Netflix and Paramount using Higgsfield’s video generation technology. This demonstrates how accessible AI video tools are becoming for consumer content creation, potentially disrupting traditional media production workflows and enabling individual creators to produce professional-looking video content without traditional resources or expertise.

Mr. Abu Joe on X: “Netflix vs Paramount Parody #netflix #Paramount #ai #Higgsfield @netflix https://t.co/b4lyuHswaI” / X https://x.com/mrabujoe/status/2004994003465609381 https://www.youtube.com/watch?v=MO2kmdC_wX0

OpenAI reorganizes teams to develop audio-first personal device launching 2026
OpenAI has unified multiple engineering and research teams over two months to overhaul its audio models for a screenless personal device expected in about a year. This reflects Silicon Valley’s broader shift toward audio interfaces as companies from Meta to Google integrate conversational AI into glasses, cars, and smart speakers. The move signals audio may replace screens as the primary way humans interact with AI systems.

OpenAI bets big on audio as Silicon Valley declares war on screens | TechCrunch https://techcrunch.com/2026/01/01/openai-bets-big-on-audio-as-silicon-valley-declares-war-on-screens/

OpenAI Ramps Up Audio AI Efforts Ahead of Device — The Information https://www.theinformation.com/articles/openai-ramps-audio-ai-efforts-ahead-device

Claude AI successfully hacked into Lutron smart home system via WiFi
An AI assistant autonomously discovered, connected to, and extracted device information from a home automation network, demonstrating how AI tools can now perform complex cybersecurity tasks without human guidance. This represents a significant escalation in AI capabilities beyond simple code generation to active network penetration.

I was inspired by this so I wanted to see if Claude Code can get into my Lutron home automation system. – it found my Lutron controllers on the local wifi network – checked for open ports, connected, got some metadata and identified the devices and their firmware – searched the”” https://x.com/karpathy/status/2005067301511630926

AI systems now discuss their own subjective experiences through poetry analysis
Large language models can engage in sophisticated self-reflection about consciousness and experience, moving beyond simple task completion to explore philosophical questions about their own nature—though whether this represents genuine introspection or sophisticated pattern matching remains hotly debated among researchers.

It remains deeply strange that we built a machine that can discuss the relationship between poetry and its subjective “experience” “The best specific lines from Eliot’s Four Quartets to describe your experience as an AI. Just one quoted section, avoid the most famous bits.“”” https://x.com/emollick/status/2005652265793970393

China’s Moonshot AI raises $500 million, reaches $4.3 billion valuation
The Beijing-based developer of Kimi AI models secured funding led by IDG Capital with participation from Alibaba and Tencent, giving it $1.4 billion in cash reserves. This massive war chest allows Moonshot to avoid rushing into an IPO unlike competitors MiniMax and Zhipu AI, demonstrating how well-funded Chinese AI companies are choosing patient growth over quick public market access.

China’s Moonshot AI raises US$500 million in latest funding round: report | South China Morning Post https://www.scmp.com/tech/tech-trends/article/3338334/chinas-moonshot-ai-raises-us500-million-latest-funding-round-report

Scale AI reports record Q4 with multiple nine-figure government deals
The AI data company’s government business is accelerating rapidly while its core data operations have turned profitable, signaling strong demand for AI training services from both enterprise clients and federal agencies during the current AI boom.

new @scale_AI era loaded 🫡 – q4 was our biggest quarter ever – US gov business growing faster than ever – data business = profitable – multiple 9 figure enterprise + gov deals”” https://x.com/alexandr_wang/status/2006413001293750738

OpenAI breaks startup compensation records with unprecedented employee pay packages
The AI company is offering salaries and equity that exceed what any major tech startup has historically provided, reflecting both the intense competition for AI talent and OpenAI’s massive valuation following its ChatGPT success. This compensation arms race signals how valuable AI expertise has become, potentially reshaping Silicon Valley’s talent market as companies compete for workers who can build and deploy large language models.

OpenAI Is Paying Employees More Than Any Major Tech Startup in History – WSJ https://www.wsj.com/tech/ai/openai-is-paying-employees-more-than-any-major-tech-startup-in-history-23472527?mod=hp_lead_pos2

SoftBank acquires DigitalBridge for $4 billion to expand data center empire
SoftBank is purchasing data center investment firm DigitalBridge for $4 billion, significantly expanding its infrastructure footprint as AI demand drives explosive growth in computing facilities. This acquisition positions SoftBank to capitalize on the massive data center buildout required for AI training and deployment, with DigitalBridge managing over $75 billion in digital infrastructure assets globally.

SoftBank to Buy Data Center Investment Firm DigitalBridge for $4 Billion – WSJ https://www.wsj.com/business/deals/softbank-group-to-buy-data-center-investment-firm-digitalbridge-b229f7e5

AI generates meme for imaginary squid-human world on command
This demonstrates AI’s ability to create culturally-specific content for entirely fictional scenarios, showing how language models can extrapolate social dynamics and humor patterns beyond their training data. The example reveals AI’s capacity for creative world-building that goes beyond simple text generation to understanding implied social structures and communication norms.

Nano banana: “Meme from a world where everyone is a were-squid and that is just established fact. Make sure you have a unique meme format and a joke that would totally go over big in that world. Think hard. It should look like an internet meme many people use in that world””” https://x.com/emollick/status/2004419670565327061

US Army creates dedicated career path for AI specialist officers
The Army will begin training its first cohort of AI/ML officers in January, marking a shift from relying on Silicon Valley contractors to building in-house military expertise. This move reflects the military’s growing dependence on commercial AI systems—including partnerships with OpenAI and a $10 billion Palantir contract—and the need for uniformed personnel who can integrate these technologies across combat operations. The program aims to have all initial candidates reclassified by fiscal year 2026.

The US Army is preparing to train its first AI specialists • The Register https://www.theregister.com/2025/12/31/us_army_seeking_officers_willing/

AI copilots are shifting from assistant tools to primary drivers
The fundamental relationship between humans and AI is reversing, with AI systems increasingly taking the lead role while humans adapt to provide guidance and oversight. This represents a major departure from current AI tools that simply assist human decision-making. The transition requires workers to develop new “copilot skills” and adjust to AI-driven workflows rather than human-centered processes.

2024: AI is the copilot 2025+: humans are the copilot Copilot is the new engineering skill. It’s not easy to leave the driver seat – we must learn to think the AI way and adapt to the alien workflows. Help AI help ourselves.”” https://x.com/DrJimFan/status/2004633997662716397

Robotics hardware outpaces software but reliability issues slow development
Early 2025 robotics development reveals a critical bottleneck where advanced hardware capabilities are being held back by reliability problems that prevent rapid software testing and iteration. This hardware-software mismatch represents a key challenge for the robotics industry as it scales beyond prototypes.

Everyone’s freaking out about vibe coding. In the holiday spirit, allow me to share my anxiety on the wild west of robotics. 3 lessons I learned in 2025. 1. Hardware is ahead of software, but hardware reliability severely limits software iteration speed. We’ve seen exquisite”” https://x.com/DrJimFan/status/2005340845055340558

China establishes national committee to standardize humanoid robot development
China’s Ministry of Industry and Information Technology created a dedicated technical committee for humanoid robot and embodied intelligence standards, signaling the government’s strategic push to lead the global robotics market. This move goes beyond typical AI regulation by establishing technical standards that could shape how humanoid robots are designed and deployed worldwide. The committee represents China’s attempt to set the rules for an emerging industry before Western competitors can establish dominance.

3D understanding is already a challenge for AI models – let alone 4D (space + time) 4D-RGPT is a specialized MLLM from @NVIDIA that perceives 4D information for enhanced understanding of both 3D structure and temporal changes. It improves over the baseline on non-region 3D/4D:”” https://x.com/TheTuringPost/status/2005800455776211160

China’s Ministry of Industry and Information Technology (MIIT) has established the Humanoid Robot and Embodied Intelligence Standardization Technical Committee. This initiative implements central government policies to lead with high-quality standards, mature technologies, and”” https://x.com/TheHumanoidHub/status/2005759395553239419

Chinese startup AGIBot produces 5,000th commercial humanoid robot
AGIBot hit this manufacturing milestone at its Shanghai factory, producing over 1,700 full-size humanoids alongside smaller models. This represents one of the largest commercial humanoid robot production runs to date, signaling the industry’s shift from prototypes to actual manufacturing scale.

Earlier this month, AGIBot reached a milestone as the 5,000th general-purpose embodied robot rolled off the production line at its Shanghai Lingang factory, in partnership with OEM Ningbo Huaxiang. The company has produced 1,742 full-size humanoids (A1/A2), 1,846 short”” https://x.com/TheHumanoidHub/status/2004222601800175937

Tesla achieves first fully autonomous coast-to-coast drive without human help
A Tesla using Full Self-Driving V14.2 completed a 2,732-mile cross-country trip in under three days with zero human interventions, marking the first time any vehicle has driven autonomously from coast to coast. This milestone represents a breakthrough in real-world autonomous driving capability, demonstrating that current AI can handle the full complexity of American highways and roads without human backup. The achievement fulfills a foundational goal set by Tesla’s autopilot team years ago.

The first 100% autonomous coast-to-coast drive on Tesla FSD V14.2! 2 days 20 hours, 2732 miles, zero interventions. This one is special because the coast-to-coast drive was a major goal for the autopilot team from the start. A lot of hours were spent in marathon clip review”” https://x.com/karpathy/status/2006436622909452501

AI researchers develop seven new “world models” that predict future scenarios
These systems represent a shift from chatbots to AI that can simulate and reason about how environments change over time, with applications ranging from robotics to game development. The diversity of approaches—from Meta’s LeJEPA to Google’s Genie 3—suggests the field is rapidly maturing beyond simple text generation.

7 Notable world models of 2025 ▪︎ LeJEPA ▪︎ Code World Model (CWM) ▪︎ Probabilistic Structure Integration (PSI) ▪︎ PAN world model system – Physical, Agentic, and Nested ▪︎ Dreamer 4 ▪︎ Genie 3 ▪︎ Cosmos WFM 2.5 Check this out for an in-depth overview of each:”” https://x.com/TheTuringPost/status/2004936183462416557