GPT 5.5

1. Something about the model auto-routing changed (allegedly) (www.reddit.com via reddit)

   8w grokgpt-5cursor+1

   Don't get me wrong, I love Cursor and will continue using it no matter what! However, there are a couple things happening that I find "strange" to say the least.

2. Autonomous disproofs of the sum-product conjecture over $\mathbb R$ with GPT-5.5 Pro (arxiv.org)

   8w gpt-5openai

   OpenAI's recent disproof of the Erdős unit distance conjecture marked a milestone for AI in mathematics. It also inspired another breakthrough: a human disproof of the Erdős--Szemerédi sum-product conjecture over $\mathbb R$.

3. We probed a pinned GPT-5.5 endpoint: every request carried ~1,447 hidden tokens (tosea.ai via hn)

   +2 8w gpt-5claude-code

   We Fed Claude Code's System Prompt to GPT — Its Fingerprint Drifted Like a Different Model Hidden system prompts can push an LLM's fingerprint to JSD 0.46 — the 'different model' band. 5,660 controlled requests show what drifts, which prob…

4. Relay-Bench: Evaluating LLMs on Multi-Domain Reasoning Chains (arxiv.org) discussed ↗

   8w gpt-5

   Introducing Relay-Bench, an unsaturated, holistic, text-only benchmark that measures LLMs' ability to complete an assortment of tasks from distinct domains in a single prompt. The leading model, GPT-5.5 (xHigh), scores 43.3%.

5. Save GPT-5.5 (save-gpt-5-5.fyi via hn)

   +21 8w gpt-5

   Save-gpt5-5.ai I love this model in ways so personal that I don't have the heart to share. Yes, I even paid for the ridiculous upcharge for the ".ai" TLD [nope].

← all threads