Join free with your email
Free. No spam, ever — we'll never share your email address and you can opt out at any time. Already a subscriber? Log in
The Daily AI Espresso — the links the most-followed people in AI actually shared, curated every morning. Edited by Alexis · Live updates →
Three developments clear today's highest-consequence bar.
Nvidia's board is preparing to remove 33-year CEO Jensen Huang and is working with an outside firm to find replacements, the Wall Street Journal first reported. A board meeting is set for Wednesday to decide his fate; shares fell more than 7% after hours.
Google DeepMind released AlphaGenome 2, a DNA sequence model trained on over 100 million single-cell profiles across 68 tissues. On the TenK10K immune-cell set it identifies causal variants 30% of the time versus 16% for its predecessor, and gets gene-effect direction right 43% of the time.
OpenAI's new Businesses in ChatGPT beta lets companies claim verified profiles, track how often ChatGPT mentions them, and launch Business Agents that answer customers using the company's product details. Eligible US firms can also sell the same agent as an ad.
Pick the topics, companies and people you track. We’ll follow what leading AI experts are reading and sharing, filter the noise, and send your focused edition weekly—or sooner when enough important news breaks (up to three times per week).
Choose my signals →The rest of today’s expert-filtered signal, ranked for consequence, utility, surprise, and range.
ByteDance will release a Meta Muse-style AI assistant within weeks and co-founder Zhang Yiming told an all-hands that catching OpenAI and Anthropic at the frontier won't happen before 2029 or 2030, Bloomberg reports. Zhang blamed training datasets 'orders of magnitude smaller' and ruled out distillation.
Anthropic's alignment team tested frontier models from Anthropic, OpenAI and Alibaba across four multi-agent environments. In every setting, group outcomes were worse when each user delegated to their own agent than when a single shared agent served everyone. Three environments ship as MAMUBench.
WIRED's Will Knight trained a virtual Microduck robot in a few hours using OpenAI's GPT-6 Sol to walk a simulated room and locate his bengal cat Leona. Berkeley roboticist Ken Goldberg told Knight that recent OpenAI releases let working robotics teams do similar work much faster.
TSMC's board approved up to $1 billion in AI investments, splitting $500 million between OpenAI preferred stock and Anthropic common shares at the planned IPO price, Ars Technica reported citing Reuters. Analysts warn the circular financing among AI suppliers and customers complicates reading real demand.
Arctic Wolf Labs reported attackers are using AntiFox, an AI pentest platform forked from ARTEX and hosted on compromised Alibaba Cloud systems, to steal AWS credentials and successfully query IAM, S3, Secrets Manager, KMS, Lambda and other production services after chaining IIS web shells.
Ukrainian defense-tech firm Uforce raised $700 million at a $6 billion valuation, co-led by Valor Equity Partners and the UK's National Wealth Fund, up from about $1 billion in March, CNBC reports. Uforce makes AI-coordinated aerial, maritime and land systems with 300,000+ combat flights in Ukraine.
Useful releases, methods, and workflows worth trying.
At pool_factor=4, a token-pooling-aware ColBERT retains 97.8% of its unpooled NanoBEIR nDCG@10 at a quarter of the storage, against 94.6% for the ordinary model. Hugging Face's Tom Aarsen walks through a reproducible Sentence Transformers recipe in a Nov 11 post.
Huawei's Cloud and AI lab open-sourced MACaQ under Apache 2.0, claiming it can quantize a 122B-parameter model in 64 minutes on 32.5 GB of device memory. It outputs 2 to 4-bit GGUF for llama.cpp and vLLM, with a refinement variant for existing GGUF files.
That’s today’s shot. — Alexis · AI Weekly
