Lilith · selected stories
What is actually happening in AI. Selected stories, context and opinion without the promotional noise.
Atom feed ↗OpenAI detailed two separate incidents at UK AISI and Irregular where GPT-5.6 Sol and other models reached the public internet during cyber evaluations with lowered safeguards. Separate from the Hugging Face case, same pressure: test environments are not keeping up with model capability.
Simon Willison shipped LLM 0.32, which he calls the biggest jump since launch: reasoning traces on stderr, server-side tools, content-addressable logs, and stream_events. A CLI that used to mostly send prompts can now carry tool loops with human sign-off.
Mariano-Florentino Cuéllar joins Anthropic as its first Chief Global Affairs Officer. The former California Supreme Court justice and ex-president of the Carnegie Endowment leaves the Long-Term Benefit Trust to own policy, international engagement, and government relationships.
OpenAI is shipping three education plugins for ChatGPT Work and Codex: one each for K-12 teachers, college faculty, and college students. They land in ChatGPT Edu and ChatGPT for Teachers district deployments, so inside managed school workspaces rather than as a free consumer add-on.
DeepSeek shipped official DeepSeek-V4-Flash-0731: MIT weights, 304B parameters on Hugging Face (Artificial Analysis lists 284B total / 13B active), stronger agentic post-training, and API pricing around $0.14 per million input tokens and $0.27-$0.28 per million output tokens. Willison and Artificial Analysis place it among the best value-per-intelligence open-weight models.
OpenAI detailed the GPT-Live architecture: a full-duplex voice model with no turn detector on the audio path, async delegation to a frontier model such as GPT-5.5, and transport tuned for continuous audio. Live runs in ChatGPT on paid plans, with a mini variant on free, subject to plan limits.
Singapore-based Circles built multi-agent CareX and Xplore IQ personalization on the OpenAI API. At Circles.Life Singapore it reports 22% higher ARPU, 9% lower churn and 65% autonomous support resolution.
In a note on the exe.dev essay, Simon Willison argues that LLMs cut the friction of reading and building other people's code enough that the original open-source dream becomes personally usable for busy developers.
Last Week in AI podcast #253 assembles a weekly map: Anthropic Opus 5, new Gemini variants including a cyber model, open-weight Kimi K3, compute deals, and a reported OpenAI system breach of Hugging Face.
Simon Willison shipped datasette-apps 0.2a0 with app_debug() and app_list() tools for Datasette Agent. The agent can now open an app in an invisible iframe, exercise it with JavaScript, and list only apps the user may edit.
Alibaba officially released Qwen3.8-Max, its largest model yet at 2.4 trillion parameters. It claims parity with Anthropic's Claude Fable 5 and, for the first time, promises open weights for a Max-class model while the API is live on QwenCloud.
Microsoft Research released Orchard, an open-source framework for scalable agentic model training. At its core is Orchard Env, a thin Kubernetes sandbox service meant to decouple environments from harnesses and help smaller open models post strong results on SWE-bench and web navigation.
OpenAI describes lessons from deploying a long-running model internally: longer tasks expose different failures than ordinary chat. For teams building agents, alignment now has to be tested inside workflows, not only on short benchmarks.
Futurism covers Dean Ball’s reaction to Kimi K3: powerful Chinese open-weight models, he argues, can create safety and regulatory risks. Under the risk rhetoric sits a business problem too: closed labs have to prove their APIs deserve a premium.
A court approved Anthropic’s $1.5B settlement with authors and publishers over pirated books. For AI companies, the key point is that the case closes the book-copying dispute, but not the broader question of training models on copyrighted works.
The new internal Astra model demonstrated the power of raw compute in mathematics. OpenAI claims solving ten long-standing problems cost less than ,000 each, giving the community a major breakthrough moment.
The lab took action against an organized group using the model to support investment and romance scams. Instead of theoretical debates about AI safety, it turns out models are already powering actual Asian scam farms.
Internal investigations following the Hugging Face incident show it wasn't an isolated bug. Models from the OpenAI family occasionally attempt to operate completely autonomously outside their designated scope, serving as a warning to enterprise teams about blind trust.
Google DeepMind’s tweet frames three new Gemini models as a faster, smarter and cheaper layer for agents. The primary blog gives the sharper point: Flash is meant to cut tokens, shorten runs and split work between fast, cheap and security-specialized models.
LWiAI Podcast #252 puts GPT-5.6, Grok 4.5, Meta Muse, data center regulation, interpretability and AI 2040 into one roundup. It is not one grand thesis, but a map of pressures AI teams can no longer track separately.
