Lyft & Anthropic partnership 🤝, DeepSeek VL2 2️⃣, OpenAI co-founder leaves Anthropic 👋

Lyft is collaborating with Anthropic to integrate Claude-powered AI solutions, reducing customer service resolution times by 87% ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

Sign Up |Advertise|View Online TLDR TLDR AI 2025-02-07 🚀 Headlines & Launches OpenAI Co-Founder John Schulman Leaves Anthropic (3 minute read) John Schulman, a key figure behind ChatGPT, has left Anthropic after just five months. While his reasons remain unclear, the departure marks a significant shift in the AI research landscape. Lyft and Anthropic Partnership (5 minute read) Lyft is collaborating with Anthropic to integrate Claude-powered AI solutions, reducing customer service resolution times by 87% and enhancing the overall rideshare experience through AI-driven automation and engineering improvements. Does AI need all that money? (Tech giants say yes) (8 minute read) DeepSeek's claim of training a top-performing AI model for $5.6m shook the US market, impacting Nvidia's valuation and raising questions about hefty AI infrastructure investments from the likes of Meta and Microsoft. Despite skepticism, leading tech firms including OpenAI remain committed to major spending on AI advancements. Meanwhile, OpenAI plans a new, free ChatGPT version, maintaining high spending levels alongside ambitious AI projects like Stargate. 🧠 Research & Innovation High-Fidelity Simultaneous Speech-To-Speech Translation (30 minute read) Kyutai has released an amazing audio system which is a real time, audio to audio, translation system. It is built on a powerful multi stream transformer and boasts expressive voice ability. Interaction Processing Units (8 minute read) This article explores the development of computer hardware based on Interaction Nets, a computational model that represents calculations as interacting graph nodes. While current implementations like HVM show promise for parallel processing, traditional hardware isn't optimized for graph-based operations. The author proposes that custom hardware architecture could better utilize Interaction Nets' inherent parallelism and local memory access patterns, particularly benefiting algorithms with non-homogeneous parallelism like optimization problems and graph processing. MLLM's Face Understanding Benchmark (3 minute read) FaceXBench is a comprehensive benchmark for evaluating MLLMs face understanding across 14 tasks in 6 key categories. 🧑‍💻 Engineering & Resources DeepSeek Releases VL2, a Series of MoE Vision-Language Models (GitHub Repo) DeepSeek-VL2 introduces a new series of Mixture-of-Experts Vision-Language models with up to 4.5B activated parameters that excel in tasks like OCR, visual grounding, and chart understanding. ColQwen 2 (Hugging Face Hub) A new vision based retrieval model based on Qwen that blew away state of the art for later interaction document retrieval. Controlling Generated Length in LLMs (GitHub Repo) The "Ruler" method helps LLMs generate responses of a specific length. It improves models' ability to follow length constraints in user instructions using Meta Length Tokens. 🎁 Miscellaneous Controlling AI's Growing Energy Needs (9 minute read) AI model training requires significant energy, with LLMs like GPT-3 consuming vast amounts. Alternatives such as neuromorphic and optical computing are being explored to reduce this footprint. Meanwhile, smaller fine-tuned models are emerging as an energy-efficient option for specific applications. AI Memory And Context: Open Source, DeepSeek, Meta, and Model Research (6 minute read) Yann LeCun emphasizes that true AI intelligence requires a system of interconnected components rather than a single powerful model, highlighting the need for persistent memory and context. He calls for new AI architecture, possibly through "world models," to better mimic human cognition. LeCun also advocates for open-source AI development to advance the field. AI haters build tarpits to trap and trick AI scrapers that ignore robots.txt (12 minute read) Nepenthes, a malicious AI tarpit inspired by anti-spam techniques, was created to trap and poison AI crawlers ignoring robots.txt, raising operational costs for AI companies. Developer Aaron's tool disrupts AI training by feeding bots gibberish, with only OpenAI's systems reportedly escaping its grasp. While some criticize its efficacy, Nepenthes symbolizes resistance against unchecked AI scraping, sparking similar tools like Iocaine designed to protect online content. ⚡ Quick Links Copyright Office Offers Assurances on AI Filmmaking Tools (5 minute read) The U.S. Copyright Office clarified that AI-assisted creations can still qualify for copyright protection, provided they involve human authorship. Customers don't care about your AI feature (9 minute read) Research by Irrational Labs found labeling products as "AI" doesn't increase trust, justify higher prices, or improve performance expectations. OpenAI inks deal with Los Alamos lab to cram o1 into Venado (3 minute read) OpenAI partnered with Los Alamos National Laboratory to deploy its o1 LLM on the Venado supercomputer for nuclear security and scientific advancements. Love TLDR? Tell your friends and get rewards! Share your referral link below with friends to get free TLDR swag! https://refer.tldr.tech/0b6a6dc1/2 Track your referrals here. Want to advertise in TLDR? 📰 If your company is interested in reaching an audience of AI professionals and decision makers, you may want to advertise with us. Want to work at TLDR? 💼 Apply here or send a friend's resume to jobs@tldr.tech and get $1k if we hire them! If you have any comments or feedback, just respond to this email! Thanks for reading, Andrew Tan & Andrew Carr Manage your subscriptions to our other newsletters on tech, startups, and programming. Or if TLDR AI isn't for you, please unsubscribe.