We’re working with @ApolloResearch, @METR_Evals, @TransluceAI, and @GoodfireAI on embedded assessments: giving outside experts deep access to our internal models and systems to investigate alignment and test our safeguards. These assessments build on our ongoing alignment…
DSA Library
News
What people in tech are talking about on X, by day.
Monday, November 16
We’re suing OpenAI for its testing and development of dangerous AI models without adequate safeguards. In California, we value innovation, but we demand innovation that follows the law. oag.ca.gov/news/press-rel…
Sunday, November 15
Alibaba has launched Qwen4 Max, gaining 9 points over Qwen3.8 Max on the Artificial Analysis Intelligence Index and landing 3 points behind Kimi K3.1 Pro at ~11% lower Cost per Task Qwen4 Max, using xhigh reasoning effort, scores 43 on the Intelligence Index, 9 points ahead of
Saturday, November 14
The open model world faces two main challenges in a post-Jensen world: 1. Political exposure. Open models’ allies are retreating in DC. 2. Investment. Open model builders need to find profitable products far sooner w/o his backstop. Will feel it sooner than people expect.
Friday, November 13
ChatGPT @ChatGPT Introduces ChatGPT Wallet, which lets agents complete purchases with user approval and without access to the card details.
karan ahuja @ahuja_karan Says every US Pro, Max and Ultra subscriber on the Claude Duet waitlist has now been invited, with a note on usage.
prinz @deredleritt3r Relays Fortune’s report that OpenAI disbanded its robotics team because it expects LLMs to become the brains of robots.
Aaron Tilley @aatilley Reports that SpaceX cut the roughly 60-person team behind the Grok Imagine video generator, which stays live for now as resources shift elsewhere.
Thursday, November 12
will depue @willdepue Claims forensic evidence that DeepSeek, Moonshot, MiniMax and Ant Group trained their newest models on Nvidia Blackwell chips barred from export to China.
Logan Kilpatrick @OfficialLoganK Introduces Gemini 4 Hydrogen, Google’s fast Gemini 4 model, across AI Studio, the Gemini API, the Gemini app, Search and Antigravity.
Artificial Analysis @ArtificialAnlys Places Gemini 4 Hydrogen just below Claude Sonnet 5.5 and GPT-6.1 Sol on its Intelligence Index, at a third of Sonnet’s cost and about 2.5x GPT-6.1 Sol’s.
Wednesday, November 11
Arena @arena Reports Claude Fable 5.2 (Max) taking first place in Agent Arena across 4,700 real agentic sessions, and first in Code, Chat and Work.
Tuesday, November 10
Sebastian Raschka @rasbt Explains the shift toward hybrid and sparse attention architectures, distinguishing reduced attention computation from fixed-state recurrent memory and its retrieval tradeoffs.
Monday, November 9
Claude @claudeai Introduces Claude Fable 5.2 and Claude Mythos 5.2, with gains in coding, reasoning and vision.
Simon Willison @simonw Demonstrates using GitHub Markdown's collapsible details and summary tags to keep lengthy AI-generated pull-request descriptions out of the main review narrative.
Sunday, November 8
Salvatore Sanfilippo @antirez Demonstrates updated DwarfStar inference for DeepSeek 4.1 Flash using SSD streaming and notes an overthinking issue partially addressed by the latest model update.
Salvatore Sanfilippo @antirez Reports that DwarfStar's kimi-k3 branch can run K3.1 PRO inference directly from safetensors weights using SSD streaming.
Wednesday, November 4
Salvatore Sanfilippo @antirez Describes predicting the next layer's top four MoE experts from the previous layer's residual to issue SSD prefetch hints during K3.1 inference on 128 GB machines.
Tuesday, November 3
Sebastian Raschka @rasbt Summarizes ten recent open LLM architectures with parameter counts and design details spanning grouped-query attention, sparse attention, recurrent layers, looped transformers, and mixture-of-experts.