An inspection of Muse's runtime uncovered a model ID azure/muse-special used in one agent session amid a majority of Meta’s Avocado sessions. Logs contain a gpt_responses_v1 signature with an encrypted payload starting gAAAAA (an OpenAI marker) and tool-call IDs formatted like OpenAI's Responses API. The shipped model catalogue includes many Avocado variants plus Claude (Anthropic), Sonnet/Haiku, GPT-5.5/5.6 variants via OpenAI/Azure/Codex and other Meta-hosted routes. Repository files include an Anthropic client, prompt conversion and SSE parsers, API key files restricted to an inference-proxy, and an environment reverse-proxy kill-switch.
Those traces show Muse routes some requests to frontier providers rather than running solely on Avocado, enabling selective use, A/B testing, or behind-the-scenes routing for tasks where external models perform better. Encrypted reasoning blobs produced by muse-special are forwarded to Azure and are unreadable to Meta’s RL server, while Avocado sessions write unencrypted chain-of-thought text into transcripts that can be used for training unless users opt out. The runtime therefore supports server-side provider selection and telemetry; the observed evidence shows muse-special is an OpenAI model served through Azure, with only a single outlier session observed.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Tesla workers are resisting training the Optimus humanoid robots as replacements, citing difficulties with complex robot hands and job security concerns. Tesla aims to produce 1,000 Optimus robots weekly by the end of 2026 despite these challenges. (arstechnica.com)
OpenAI is experiencing issues with Codex, with root cause identification underway. Most systems, including chat completions and fine-tuning, maintain near 100% uptime, but some components are affected. (status.openai.com)
Opus 5.5 is seen as a significant update, sparking excitement and concern among developers and users. The rapid advancements in AI and LLMs raise questions about the pace of technological change. (news.ycombinator.com)
Large Language Models (LLMs) are often oversold as transformative AI, but their actual capabilities are still limited and misunderstood. Hype around LLMs has led to polarized debates and unrealistic expectations, overshadowing a more nuanced understanding of their practical uses. (jenson.org)
Recurse enables developers to quickly create and deploy specialized AI agents using a serverless harness and customizable manifests. It supports building agents for tasks like coding, design, and verification, with a focus on clear requirements and measurable results. (recurse.run)
AI is transforming computing by blurring the lines between programmers and users, enabling individuals to create and customize applications easily. This shift raises questions about the future of software distribution and the nature of applications in a world where many programs serve only one or two users. (sockpuppet.org)
Today's best Hacker News stories, summarized and screenshotted, one email a day.