About This Week’s Covers
This week’s cover symbolizes Amazon’s emerging AI dominance. Amazon has made a second $4B investment in Anthropic, OpenAI’s biggest competitor. Amazon is also exploring the creation of its own chips to compete with NVIDIA. Anthropic introduced a new structured standard for AIs to communicate with each other and third parties. In the race for dominance, Amazon is coming into focus as a front runner. The image prompt was generated by Claude 3.5 Sonnet, “ A sequence showing an Amazon box evolving like the famous March of Progress human evolution image, but evolving into increasingly advanced AI forms.” MidJourney, Ideogram, and Flux competed, and Flux won this time. The font is Apertura, which is the closest Adobe font to Styrene, the Anthropic brand font. The /’s in the date are reversed to face in the same direction as Anthrop\c’s logo.
The rest of the images were autogenerated by Claude 3.5 and the Ideogram API with the theme of Thanksgiving. All of the images are one shot, no second try – many turned out really creatively!
This Week’s Executive Summaries
Timeline When Major AI Leaders Predict AGI (Artificial General Intelligence)
- Sam Altman (OpenAI): 1 year
- Elon Musk (Tesla/X): 2 years
- Dario Amodei (Anthropic): 2 years
- Yann Lecun (Meta): 5-10 years
- Demis Hassabis (Google): 10 years
Amazon Doubles Down and Invests Second $4 Billion in Anthropic
Amazon is investing an additional $4 billion in Anthropic, the AI startup behind the Claude chatbot, bringing its total stake to $8 billion. While remaining a minority investor, Amazon will make AWS Anthropic’s main cloud and training partner, offering exclusive features like custom AI fine-tuning for AWS users. The deal highlights Amazon’s push to stay competitive in the booming generative AI market, projected to exceed $1 trillion in a decade. Anthropic recently debuted AI tools for complex computer tasks and business-focused products like Claude Enterprise.
Amazon to invest another $4 billion in Anthropic, OpenAI’s biggest rival
https://www.cnbc.com/2024/11/22/amazon-to-invest-another-4-billion-in-anthropic-openais-biggest-rival.html
Amazon to invest another $4B in Anthropic | Hacker News
https://news.ycombinator.com/item?id=42215126
“I know it seems like Amazon is eating Anthropic, with the funding and preferred compute partnership. However, I think the sub-title here of “deep technical collaboration” on “directly interfac[ing] with Trainium silicon” to improve the story of AWS chips could actually be immense.
https://twitter.com/andrew_n_carr/status/1860814071567925511
Amazon doubles down on AI startup Anthropic with another $4 bln | Reuters
https://www.reuters.com/technology/artificial-intelligence/anthropic-receives-4-billion-investment-amazon-makes-aws-official-cloud-provider-2024-11-22
Amazon’s Moonshot Plan: Challenging Nvidia’s AI Chip Dominance
Amazon is racing to challenge Nvidia’s control of the AI chip market – with its custom Trainium chips. These chips, developed by Amazon’s Annapurna Labs, aim to reduce the company’s reliance on Nvidia, whose GPUs dominate AI applications but remain expensive and hard to source due to overwhelming demand. Trainium2, Amazon’s latest iteration, is designed for internal use and external customers, boasting improved performance and energy efficiency. While Amazon’s chips have begun attracting clients like Databricks and Anthropic, their adoption is slowed by software compatibility challenges. By integrating AI tools into AWS and building on its legacy of custom hardware innovation, Amazon aims to position itself as a significant player in AI infrastructure. Yet, experts suggest it will take time before Amazon’s chips can compete “toe-to-toe” with Nvidia’s well-established offerings.
Bloomberg
Anthropic Unveils Model Context Protocol (MCP) to Connect AI with Data
Anthropic has launched the Model Context Protocol (MCP), an open standard that simplifies how AI systems connect to data sources. Instead of building custom integrations for each tool, developers can use MCP to create seamless, secure links between AI tools like Claude and systems like GitHub, Slack, and local databases. This addresses a major bottleneck to AI adoption: the isolation of AI models from critical data. By standardizing how AI interacts with data, MCP enables more efficient, context-aware systems that scale easily across industries. Early adopters like Block and Replit are already leveraging MCP to streamline workflows and boost AI productivity. With open-source resources and pre-built connectors available, MCP is poised to become the backbone of smarter, more connected AI applications.
Alexalbert__ | modelcontextprotocol | sourcegraph |
“Insanely cool. Anthropic just unveiled an open protocol that will connect Claude to any resource on a local computer. And in the future, it will allow Claude to be able to connect to any API or cloud server. The current state of AI models is of PhD level reasoning engines” / X
https://twitter.com/rowancheung/status/1861149923007439253
“Today @Anthropic is releasing MCP, a framework that allows Claude to run servers, giving it superpowers and effectively turning the Claude app into an API. We created some server that I think you’ll love! FileSystem: Claude can create, read, and edit files and folders locally.
https://twitter.com/skirano/status/1861081529071346161
ChatGPT Outperforms Doctors in Medical Diagnosis Study
A new study reveals that ChatGPT achieved 90% accuracy in diagnosing illnesses, surpassing human doctors (74%) and even doctors assisted by ChatGPT (76%). The small-scale research highlights the significant potential of AI in healthcare, especially for diagnostic support.
Nytimes | adcock_brett | gdb
Thomson Reuters Redefines Legal AI with OpenAI’s o1-mini Model
Thomson Reuters is testing OpenAI’s o1-mini language model in its CoCounsel legal assistant, excelling at nuanced legal tasks like spotting privileged content and subtle errors, surpassing GPT-4 in these areas. This innovation has driven a 1,400% increase in CoCounsel users, enhancing workflows such as document review, legal research, and drafting. By combining models from OpenAI, Google, and Anthropic for specialized tasks and acquiring Safe Sign Technologies for proprietary AI development, the company ensures tailored, secure solutions for law firms. Analysts view the precision-driven, multi-model strategy as a new blueprint for enterprise AI, especially in fields like the law, where accuracy is critical.
Venturebeat
H Company Launches Runner H, an AI Agent for Smarter Web Automation
H Company has introduced Runner H, an advanced AI agent designed to navigate websites by interpreting on-screen content. The tool allows users to automate complex web tasks using natural language instructions, making web automation accessible to everyone. Runner H stands out for its speed, accuracy, and ability to tackle unusually tough challenges, from precise data extraction to handling complex web interfaces. It has already shown its potential in tasks like extracting data with advanced filters or mapping bus stops on Google Maps. Aimed at both casual users and professionals, Runner H sets a new standard for real-world AI applications.
Hcompany_ai | adcock_brett
Alibaba Launches ‘Qwen with Questions’: An Open Source Model That Beats OpenAI’s o1-Preview
Alibaba has unveiled Qwen with Questions (QwQ), an open-source reasoning model that surpasses OpenAI’s o1-preview in key benchmarks. With 32 billion parameters and a 32,000-token context, QwQ outperforms o1-preview in mathematical (AIME, MATH) and scientific reasoning (GPQA), though it trails in coding tasks like LiveCodeBench. Released under the Apache 2.0 license, QwQ is open for commercial use, making it a transparent and accessible alternative to proprietary models. QwQ leverages inference-time self-review to refine its answers, reflecting a growing trend toward large reasoning models (LRMs). While it demonstrates advanced capabilities, the model has some limitations, such as mixing languages and logical loops. Its open-source release highlights a shift in AI development, with LRMs offering new paths to innovation as traditional large language model scaling reaches its limits. QwQ is available for download and exploration on Hugging Face.
Venturebeat | oodaloop | qwenlm
Video AI Company Runway Debuts Frames for Advanced Image Generation
Runway has launched Frames, a cutting-edge image generation model that offers unparalleled control over style and visual fidelity. Designed for creative professionals, Frames enables precise design of aesthetics, from the atmosphere to the finer details of a project. It excels at maintaining stylistic consistency while allowing for broad creative exploration. Currently available through the Gen-3 Alpha and Runway API, Frames supports integration into larger workflows, empowering users to easily craft visually consistent images and creative worlds.
Rowancheung | runwayml | runwayml
AI Visuals and Charts: Week Ending 11/29/2024
Pro-tip: follow anything that says segmentation or photo/video 3D transformation. I believe these developments will lead to embodied robots. I’m keeping a list and plan to publish it in a few weeks.
“holy cow, Frames will blow ur mind ✨
https://twitter.com/jeongsteph/status/1861868196954792104
“Stop watching videos, start interacting with worlds. Stoked to share CAT4D, our new method for turning videos into dynamic 3D scenes that you can move through in real-time!
“Dope 3D mapping rig: Leica’s CountryMapper running 31K-pixel array + 2MHz LiDAR with mechanical FMC and 60° FOV. Single-pass co-registered RGB and point clouds that are just ridiculous. And I’m saying this as someone who worked with Google’s street-level LiDAR. Except those
“Just read a neat AI paper called SAMURAI — it takes SAM 2 (Meta’s “segment anything” model) and makes it way better at tracking objects in videos. Basic problem is SAM 2 gets confused when things move fast or there’s a crowd of similar objects (big problem for VFX and video
SAMURAI: Adapting Segment Anything Model for Zero-Shot Visual Tracking with Motion-Aware Memory – https://yangchris11.github.io/samurai/
EchoMimicV2: Towards Striking, Simplified, and Semi-Body Human Animation
“Monocular depth estimation has gotten so good lately. It used to look like a blurry blobby flickering mess. Now it looks like a synthetic depth map you’d export from a 3d tool Like putting on a fresh pair of glasses 🤓
Top 32 Links of The Week – Organized by Category
Agents and Copilots
“We’re excited to introduce AgentAuth—the comprehensive auth solution designed for AI agents! We understand the pain every developer experiences regarding authentication—managing OAuth flows for Gmail, handling API keys for Linear, or setting up permissions across multiple
“The new @cursor_ai agent is INSANE It can run any python func in your codebase. The idea being being to check output & debug, but fuck that CRAZY SIDE EFFECT: this means you can customize the agent w/ callable tools! Like GPT Store actions or Claude MCP Cursor Agent Builder!
“Cursor Agent vs. Windsurf Agent I’ve been testing both Cursor and Windsurf agents to build AI web apps. I believe that on the agent stuff, Windsurf is one step ahead. The agent feature in Windsurf feels very native and like a first-class citizen. My full demo and test here:
“Microsoft just released a multi-agent simulation framework. Now you can get instant feedback from 100 different AI personalities instead of spending weeks on user research. 100% opensource.
“Microsoft introduced a suite of new specialized AI agents for Microsoft 365 at its annual Ignite Conference New agents include a Self-Service agent for HR / IT tasks, a SharePoint agent for document search and insights, a meeting note taker, and a few more” / X
Microsoft Copilot Customers Discover It Can Let Them Read HR Documents, CEO Emails – Slashdot
Rabbit
Teach mode, Rabbit’s tool for automating R1 tasks, is now available to all users
Replit
“You can now clone any app just by pasting the link into Replit Agent. – Paste the link – Hover over it – Click Take screenshot Real game-changer for building interfaces without coding.
“While developing @Replit Agent, we constantly face the challenge of how to interact reliably with a large number of external services. Similarly to what Web standards enabled in the past 30 years, we need a lingua franca between AI agents and the Web ecosystem. MCP has” / X
“I spent 10 minutes googling for a market cap chart and couldn’t find a decent one, so I made one with Replit Agent in two minutes. Still can’t believe you can make software faster than you can google for it.
Augmented and Virtual Reality (AR/VR)
“Introducing 𝐒𝐀𝐑𝟑𝐃, which tokenizes 3D objects into multiscale tokens and generates 3D objects by autoregressive next-scale prediction. 𝐒𝐀𝐑𝟑𝐃 enables fast 3D generation and comprehensive 3D understanding. arXiv:
Autonomous Vehicles
“BREAKING: xAI has discussed a deal with Tesla, xAI would get some Tesla revenue in exchange for providing the carmaker with access to its technology and resources” / X
Business and Enterprise
2024: The State of Generative AI in the Enterprise – Menlo Ventures
Ethics/Legal/Security
Trump eyes AI czar to partner with Elon Musk for U.S. edge
Russia plotting to use AI to enhance cyber-attacks against UK, minister will warn | Russia | The Guardian
AI grandma fights back against scammers – CBS News
O2 unveils Daisy, the AI granny wasting scammers’ time – Virgin Media O2
“🤖 93% of Gen Z workers use AI tools weekly, but nearly half of all workers aren’t comfortable admitting it. The tech adoption gap isn’t about usage—it’s about openness. Why are we still treating AI tools like a workplace secret? 🤔” / X
International
OpenAI faces critical test as Chinese models close the gap in AI leadership | VentureBeat
OpenAI to present plans for U.S. AI strategy and an alliance to compete with China
Open Source
OLMo 2: The best fully open language model to date | Ai2
Tülu 3 is a leading instruction following model family, offering fully open-source data, code, and recipes designed to serve as a comprehensive guide for modern post-training techniques.
“New study shows LLMs outperform neuroscience experts at predicting experimental results in advance of experiments (86% vs 63% accuracy). They use a fine-tuned Mistral 7B but other models worked too. Suggests LLMs can integrate scientific knowledge at scale to support research
Robotics and Embodiment
“How far are we from embodied intelligence? We first ask whether we can build a model to locate semantic concepts in 3D. Meet Find3D🔍: a model to locate any part of any 3D object based on any text query! 👉
Science and Medicine
“Augmented reality x-ray vision to “see through” concrete. Your infrastructure won’t just be scanned — it’ll be anchored to reality. Demo: Pix4D reality capture with precise geospatial localization.
“WaPo Editorial: Be thankful for the applications of AI in medicine. More accurate detection of cancers (breast cancer, prostate, skin, brain…), faster diagnosis of strokes, sepsis, and heart attacks, faster/better/cheaper MRIs, affordable full-body MRIs in 40 minutes. Much,” / X
“A new Science study presents “Evo”—a machine learning model capable of decoding and designing DNA, RNA, and protein sequences, from molecular to genome scale, with unparalleled accuracy. Evo’s ability to predict, generate, and engineer entire genomic sequences could change the
Hundreds More Nazca Lines Emerge in Peru’s Desert – The New York Times
Large language models surpass human experts in predicting neuroscience results | Nature Human Behaviour
“Thrilled to announce Boltz-1, the first open-source and commercially available model to achieve AlphaFold3-level accuracy on biomolecular structure prediction! An exciting collaboration with @jeremyWohlwend, @pas_saro and an amazing team at MIT and Genesis Therapeutics. A thread!
Video News
Luma expands Dream Machine AI video into platform, mobile app | VentureBeat
Twitter/X/Grok
“BREAKING: xAI has discussed a deal with Tesla, xAI would get some Tesla revenue in exchange for providing the carmaker with access to its technology and resources” / Xhttps://x.com/teslaownersSV/status/1861849343126384807
Leave a Reply