PSA: The flashing phrases from frontier models aren't their real thinking traces
rao2z · x · 2026-10-06
rao2z warns that the comforting phrases flashed by frontier models while users wait are not actual thinking traces. Since o1, frontier labs hide real chains of thought from paying users, showing instead a "commentary" generated in unexplained ways—even though users pay for the underlying tokens. The one clear exception is DeepSeek-R1, which showed its full (often pages-long and incoherent) chain of thought. If the flashing messages were genuine thinking tokens, key contradictions would arise.
More from Models
- llama.cpp beats Ollama by ~30% prompt eval speed on RTX 5060 Ti with Clef Flash 9B — ngxson · 2026-10-06
- Rumored Anthropic November IPO raises questions about Fable 5.5 October timing — haider1 · 2026-10-06
- Dev measures app security in 'minutes until GLM escapes the sandbox' — current record: 2 — lucasmeijer · 2026-10-06
- Dev says OpenAI models ignore existing design conventions, Opus still preferred for product engineering — almmaasoglu · 2026-10-06
- Economist: since GPT-5, labeling disagreements are almost always the LLM being right — soumitrashukla9 · 2026-10-06
- Dev argues Persimmon is the most underrated Western lab release: realistic user simulators may be quietly powering RL environments — willcb · 2026-10-06