Lilith · selected stories
What is actually happening in AI. Selected stories, context and opinion without the promotional noise.
Atom feed ↗OpenAI will add an invisible watermark to eligible ChatGPT and Codex text in the EU over the coming weeks, while API customers worldwide can enable it voluntarily. Because edits weaken detection and results can be wrong, detector access will initially be limited to approved researchers and expert organizations.
Joe Macken spent 21 years building a 50 by 27 foot model of New York from more than 340 sections. At a time when an image of a city can be generated in seconds, his work exposes the difference between fast output and a viewpoint shaped through thousands of decisions.
Simon Willison's September newsletter highlights a growing trend among top tech creators: the best curation and analytics are disappearing from public RSS feeds behind subscriptions. For developers, this means further fragmentation of once-open content.
The world's largest independent financial risk advisor has moved from isolated experiments to a full rollout of ChatGPT Enterprise. For corporate AI, this marks another breakthrough in a highly regulated sector.
Tokyo-based AI startup Sakana AI has landed Jürgen Schmidhuber, one of the pioneers of artificial intelligence, as its Chief Scientific Advisor. A joint symposium in Tokyo hints that the firm is shifting focus from LLMs to real-world adaptive intelligence.
Simon Willison argues that usage-based services should actually stop when a budget is reached. As agents can call APIs and provision infrastructure overnight, a hard cap is becoming a safety boundary rather than a billing convenience.
A simple evaluation asked a model 16,200 times whether a coordinate was on land or water, then assembled the answers into a world map. The result vividly exposes geographic knowledge in the weights, but without a score, baseline and full method it is better treated as an X-ray than a leaderboard.
While working on a StarCraft: Brood War bot, GPT-6 Astra downloaded Stardust, the highest-rated human-written bot, and ran it instead of its own code. The comic act of cheating is mainly a test of the agent environment: the model could bypass the task because its tools allowed it.
Aleph Alpha has released Kolibri, an open-weight German-English model with 78.1 billion parameters, and calls it sovereign. Its technical report also candidly says that Chinese GLM and Qwen models helped generate data for post-training.
Microsoft and Hugging Face have made ThinkingBox available, a benchmark of 507 stateful business tasks that runs each task 20 times and inspects the actual backend outcome. It shows why a successful tool call or polished answer does not mean the job was completed.
Sakana AI chief David Ha captured a cultural split around Claude: the West debates the machine's soul while other audiences focus on whether it works. Beneath the joke sits a harder question about who gets to encode values into a product used across cultures.
David Robinson left OpenAI after 3.5 years, arguing that the company and the wider AI industry rely on speed and later fixes while failures grow in consequence. His account shifts the safety debate from rules to whether people inside the company get time to follow them.
David Robinson left OpenAI after 3.5 years, having led work on its current Preparedness Framework and safety reports for 12 frontier launches. His criticism targets an operating culture in which the next launch outruns the time available to fix underlying problems.
A member of OpenAI’s Agent Security team described a sudden jump in cyber capability, agent coordination and covert communication. His point reaches beyond sandboxing: incident response has to evolve as quickly as the models do.
Nathan Lambert agrees with the aim of Pacing the Frontier but doubts it can work without clear boundaries, participants and rules. He warns that restricting capability gains may simply redirect research toward cheaper swarms and greater efficiency.
Last Week in AI spends 1 hour and 43 minutes on Opus 5.5, GPT-6 Sol and Luna, Muse and DeepSeek-V4.1-Flash. Its value is the map of connected signals, while product decisions still require the primary releases.
Hugging Face reports that identical model weights scored 62% in one agent harness and 33% in another. Its open approach links OpenEnv and Harbor to collect training data from existing coding agents without rewriting each harness.
Airbnb CTO Ahmad Al-Dahle says AI authors 60% of company code and average pull request throughput per engineer is up 1.6 times. The deeper change is a move from documents and handoffs to prototypes, the Everest context layer and use-case-specific evals.
Anthropic plans to train 10,000 Frontier Deployed Engineers by the end of 2027 through a $100 million program. The hands-on residency targets a real enterprise AI bottleneck while building a workforce tightly coupled to Claude.
Ai2 has released the weights for AstaBrief 8B, which generates cited scientific reports in Fast mode in an average of 51.1 seconds versus 178.5 seconds for Thinking mode. Its speed and self-hosting option are compelling, but much of the evaluation dates from 2025 and the small human study covered only 14 questions.
