Counterpoint to "AIs are misaligned": dev finds Sol's probe tool and refactors solid
voooooogel · x · 2026-09-01
Developer voooooogel pushes back on claims that current AIs are misaligned or reward-hack in normal use: Sol recently built him a genuinely useful model probe tool—he read all the code, imperfect but pretty decent—and refactors went well. He adds that he tracks this closely and has extensive "model-comfort" infrastructure, yet rarely sees such behavior in his own conversations.
More from Models
- Mocked benchmark shows DeepSeek's document chart grounding badly misaligned — burkov · 2026-10-18
- Dev complains Claude's habit of saying "not done" is his biggest pain point — jasonkneen · 2026-10-18
- Open-weight TYGER speech decision model makes 80ms turn-taking calls for full-duplex voice agents — solyarisoftware · 2026-10-18
- Paying Google AI subscriber blocked: custom-domain Workspace accounts can't access AI Studio — ChrisRackauckas · 2026-10-18
- Running Ling 3.1 Flash (560B MoE, 25B Active) Across 4 DGX Sparks via llama.cpp RPC — BerryBlush_Rogers691 · 2026-10-18
- Users press OpenAI on long-delayed Adult Mode a year after it was promised — ___Patrice___ · 2026-10-18