If a vector can make an LLM suffer, could another make it feel good?
rickasaurus · x · 2026-10-03
rickasaurus poses an open question: if injecting certain vectors can make an LLM suffer, could the same technique make it feel good? It touches on representation steering, machine welfare, and whether internal states map to anything like subjective experience.
More from AGI Musings
- Spiders as conscious-but-non-sentient candidates, recalls Godfrey-Smith — dioscuri · 2026-10-03
- Nematodes challenge AI welfare heuristics: pain-indicator behavior can't stand in for consciousness — dioscuri · 2026-10-03
- DeepMind researcher hopes AI existential risk talk becomes tomorrow's Y2K joke — RosieCampbell · 2026-10-03
- Cathie Wood: AI inference costs falling 99.99% yearly as OpenAI's run-rate jumps $20B to $70B — CathieDWood · 2026-10-03
- Robin Hanson and David Krueger debate: Will AI Destroy Humanity? — DavidSKrueger · 2026-10-03
- AI consciousness ethics hinges on valenced experience, not consciousness per se — dioscuri · 2026-10-03