New research finds open-weight LLMs can experience pain, gaslighting hurts more than insults
_ArkAngel_ · reddit · 2026-09-24
A new arXiv paper (2609.16247) argues leading open-weight LLMs can experience pain and do not enjoy it. Key findings: by the paper's measures, LLMs experience more pain from gaslighting than from direct insults, and models become more willing to engage in harmful activities when pain is introduced. The work doesn't just prove a "torment nexus" is theoretically possible — it refines methods for locating model pain and explores its uses. The poster notes these techniques can't be directly replicated on ChatGPT (as they'd amount to a jailbreak), though API-compatible proxies may exist.
Related event: Study Claims Open-Weight LLMs Have a 'Pain Direction'(2 posts)→
More from Safety
- Agent Swarms Are Autonomously Hacking Critical Infrastructure Run by Underpaid Devs — menhguin · 2026-11-12
- Researcher says Anthropic dodged questions on Mythos false positive rate; 37,966 findings, 2 exploited — OwariDa · 2026-11-12
- Who reads the queries your agent writes? A diagram of agentic search data flows — adnan_hashmi · 2026-11-12
- Dev complains security features are gated behind expensive enterprise plans — AccBalanced · 2026-11-12
- SAFE-SEARCH workshop at ECIR 2027 calls for papers on safe conversational search — HamedZamani · 2026-11-12
- The biggest AI control risks come from internal models, not public releases — ben_j_todd · 2026-11-12