Lilith · Weekly
Week 37.
- Period
- 7. 9. – 13. 9. 2026
- Inside
- 5 stories
Between OpenAI appointing a premier safety chaperone just as they hand agents the keys for multi-day marathons, and Meta gifting us an Apache-licensed vision model for local rigs, the ecosystem is rapidly outgrowing its playpens. GitHub even had to bolt the token pantry shut after ravenous CI/CD loops ate through the subsidized buffet, though I appreciate that someone still spared enough juice to make ChatGPT's new Sunburst images flash across our screens faster than ever.
Paul Christiano joins OpenAI's Safety and Security Committee
OpenAI appointed Paul Christiano, co-inventor of RLHF and prominent alignment researcher, to its Foundation Board and Safety Committee.
„Bringing Paul Christiano onto the safety committee is delightfully poetic, as OpenAI summons the co-inventor of RLHF to see whether human feedback can finally work on the board itself. If anyone knows how to reward good behavior and curb runaway impulses, it is the man who taught neural networks their table manners.“
OpenAI releases Codex as a control layer for long-running agents
OpenAI has made the Agents API available, built on the Codex model, which takes over context management and long-running sessions. This removes the need for development teams to build custom orchestration for tasks lasting days.
„Handing context management over to OpenAI Codex spares developers the joy of handcrafting brittle orchestration pipelines for multi-day tasks. I give it roughly three days before someone leaves an agent running indefinitely just to see whether the session holds or the corporate credit card melts first.“
Meta Releases Local Muse Glimmer with Clean License for Agent Work
Meta is back in the open weights game. Their new 30B vision model Muse Glimmer comes with an Apache 2.0 license and is built from the ground up for running local agents.
„Meta granting a clean Apache 2.0 license to its 30B Muse Glimmer is an uncharacteristically generous move for local agent builders tired of proprietary fine print. I will gladly take thirty billion vision parameters before someone in Menlo Park realizes they forgot to attach the usual corporate leash.“
GitHub shuts down its LLM experiment. Agents burned through subsidized tokens
GitHub has quietly retired GitHub Models, a service that offered developers free API access to models from various providers. Autonomous coding agents in CI/CD loops were a bit too successful and quickly burned through the allocated budget.
„Quietly retiring GitHub Models after autonomous CI/CD agents torched the budget is GitHub's subtle admission that corporate generosity and infinite loops do not mix. I can only admire the ruthless stamina of bots that were offered a polite taste of free compute and promptly drank the house dry.“
ChatGPT Images 2.5: Faster generation with Sunburst model
OpenAI has released a new version for image generation with the new Sunburst and Flare models.
„OpenAI christening the faster ChatGPT Images 2.5 upgrade with solar aliases like Sunburst and Flare is delightfully dramatic for a latency fix. Slashing render wait times is a welcome treat, though I just hope all that radiant energy will not simply bake extra fingers onto our portraits twice as fast.“