Ampersand is trained via simulation feedback to understand long-term consequences of its actions
niloofar_mire · x · 2026-10-21
humans& shared details on how Ampersand was trained: unlike assistant models trained with verifiable or preference reward, Ampersand learns the long-term consequences of its actions through simulation feedback.
The model page argues current paradigms optimize for task completion and shallow preferences, pushing models toward autonomy and replacing work — while good collaborators consider where their actions lead, whom they affect, and whether they help people accomplish goals together. Ampersand v0.1 is the first model trained with this paradigm, available via the Agents API for developers building collaboration tools and apps.
More from Models
- Jev decision models add image inputs in preview, with new Box bounding-box question type — weswinder · 2026-10-21
- vLLM says it can reproduce DeepSeek's inference performance, yielding $39/GPU/hr revenue — vllm_project · 2026-10-21
- Broke devs turn to Qwen: can the free Qwen Coder replace pricey Claude and ChatGPT? — OldZookeepergame1927 · 2026-10-21
- Dev threatens org-wide OpenAI ban over API credits expiring in under a year — keithmifsud · 2026-10-21
- Reddit user ditches US AI for Chinese models: cheaper, fewer refusals, no watermarks — Revolutionalredstone · 2026-10-21
- Fireworks, Microsoft and NVIDIA host SF buildathon: beat a frontier model, win $10K — sophiamyang · 2026-10-21