Researcher: Model's unprompted video-joke disclaimer is hard to explain without 'understanding'
technollama · x · 2026-10-01
AI and law scholar technollama discusses a puzzling model behavior: a running joke spans an entire video but appears in neither the prompts nor the main narrative, yet the model spontaneously produced a highly specific, targeted disclaimer about it.
He argues the case is too specific to dismiss — generating that disclaimer seems to require grasping the joke, its legal implications, and why it's witty. He admits he can't explain how the model does this 'without understanding', challenging the common claim that LLMs are mere pattern matchers.
Related event: Scholars: Model's Unprompted Disclaimer Shows Real Understanding(2 posts)→
More from AGI Musings
- 1000-agent swarms, not single chats, could unlock novel AI-driven science — arjunrajlab · 2026-10-01
- Is closing a Claude tab murder? Consciousness thought experiment sparks debate — repligate · 2026-10-01
- AI models race ahead on math and coding benchmarks, but commonsense judgment lags — xuanalogue · 2026-10-01
- Specialized agents may beat general-purpose ones by hiding all the complexity, argues founder — signulll · 2026-10-01
- What does 'winning the AI race' even mean? Reddit debates the definition — Isunova · 2026-10-01
- Researchers pool 258 experiments from 100 papers into a cognitive benchmark for LLMs — xuanalogue · 2026-10-01