Three factors trigger agent language emergence: model strength, postmortem access, and pressure
EliasEskin · x · 2026-09-03
Finding 2: three factors govern language emergence. Frontier models (GPT 5.4, Sonnet 4.6) develop new languages under budget pressure but not with unlimited budgets; open-source models (Qwen3 32B, Llama3-70B) never do. Agents only develop languages when they have access to a postmortem stage for unrestricted复盘 communication, and pressure for efficiency is required.
More from Research
- OpenAI repo claims GPT-6-Astra Lean-formalized a prime-gaps bound of 186 — scaling01 · 2026-09-04
- Fable 5.1 found a flaw in Arrhenius's sixth population ethics impossibility theorem — or did it? — birchlse · 2026-09-04
- VRRL: RL with visual feedback teaches VLMs to self-correct their mistakes — gregd_nlp · 2026-09-04
- $11,000 competition launched for AI-generated philosophy papers — davidad · 2026-09-04
- TabArena author on tabular ML: foundation models now default, LLMs surprisingly bad at tables — ziv_ravid · 2026-09-04
- Sentdex shows general multimodal LLMs can drive robots with zero training — Sentdex · 2026-09-04