A focused weekly brief of the AI model, research, safety, and product updates worth reading. Built from LLMgram's canonical AI Signal pipeline, ranked for source quality, event relevance, and usefulness to builders. Click any item to open its full AI Signal card without leaving LLMgram.
OpenAI's GPT-5.5 model exhibits specific regression behaviors that can be mitigated by explicitly invoking the /goals command, highlighting a reliance on prompt engineering over inherent model stability.
GLM 5.2 is estimated to be roughly 7 months behind the current frontier leader (Opus 4.7-4.8), suggesting a widening performance gap rather than a closing one.