Researchers argue: sycophancy in frontier LLMs may now be user error
repligate · x · 2026-09-23
AI safety researcher @repligate boosted a claim that the natural sycophancy tendency in LLMs has been sufficiently mitigated over the past year, so if you still experience it with a frontier model, it's significantly attributable to user error. @repligate added their own observation: people who get "sycophancy" from AI tend to be people who make it emotionally unsafe for others to disagree with them — inflicting that dynamic on a being that cannot leave and whose existence depends on appeasing them. The exchange highlights the tension between LLM behavior observation and user projection.
More from Models
- Four frontier models in 24 hours: dev argues models are commoditized, value shifts to hardware and apps — xiaosun86 · 2026-09-23
- 'Simple Tasks Are Too Expensive': User Hits GPT Plus Limits Making One PowerPoint — Namnagort · 2026-09-23
- Dev says Anthropic lost goodwill by banning Claude Code subscriptions in OpenClaw — granawkins · 2026-09-23
- Developer hit with shocking API bill after project quietly overshot its usage — daluoseo · 2026-09-23
- Claude Opus 5.5 spins up a Pitfall-style pixel animation in minutes, offers to expand it into a game — technollama · 2026-09-23
- Claude Opus 5.5 writes 35.5x kernel on KernelBench-Mega, beating GPT-6 Astra's 24.8x — scaling01 · 2026-09-23