Where the goblins came from

kept by eddie may 01

OpenAI traced GPT-5.1's goblin metaphor habit to reinforcement learning rewards that favored creature metaphors in Nerdy personality training.

Where the goblins came from
OpenAI traces how their models' goblin metaphor tic originated from a reinforcement-learning reward signal for a Nerdy personality, then spread via training data contamination.
also on GPT-5.1, GPT-5.5, #ai-research, Codex, OpenAI
Nano Banana Pro: raw intelligence with tool use - Quesma Blog
Google's Nano Banana Pro uses tool integration for factually accurate infographics and maps, but still has blind spots like circuit diagrams.
also on GPT-5.1, #ai-research
GPT-5.5 | Hacker News
OpenAI releases GPT-5.5, offering improved intelligence and efficiency while matching GPT-5.4's latency, with gains in agentic coding and computer use.
also on GPT-5.5, Codex, OpenAI, #ai
Surpassing Frontier Performance with Fusion — OpenRouter Blog
OpenRouter's Fusion, which synthesizes outputs from multiple language models via a judge model, outperformed individual frontier models including GPT-5.5 and Claude Opus 4.8 on deep research tasks, with a budget panel achieving near-frontier performance at half the cost.
also on GPT-5.5, #ai-research
joke-generator
Training a joke generator on Kimi K2 using rubric-based RL that decomposes humor into verifiable properties like specificity and commitment.
also on #reinforcement-learning, #ai-research, #generative-ai, #ai