Yacine: this model is genuinely aligned — alignment research is capabilities research
yacineMTB · x · 2026-09-27
Well-known AI builder Yacine posted: "Genuinely, well done. This model is genuinely aligned to me. Not in a stupid fake way but in a way that makes it a good product." He argues that alignment research is actually capabilities research. He also adds that "astra is by far the most maligned model I have had to interact with," a strongly negative first-hand take. A practitioner's candid evaluation worth noting for the alignment-as-capabilities argument.
More from Models
- Pedro Domingos: Releases are for software — his company runs one continuously evolving model — pmddomingos · 2026-09-27
- MiniMax M3.1 nearing launch: partner tests show multimodal input and new speculative decoding — kimmonismus · 2026-09-27
- OpenAI credits back online after brief outage and reset — dfinke · 2026-09-27
- Grok 4.7 Lands at #16 in Agent Arena With +3.96% Gain, But Costs 54% More Per Task — arena · 2026-09-27
- Dev says Astra quality tanked overnight, blindly shipping to production cost him users — Maleficent_Disk9583 · 2026-09-27
- No one has hit a usage limit on Claude Opus 5.5 yet — iamfakhrealam · 2026-09-27