Interp is full of unverifiable slop, but divine questions may be answered sooner than expected
ericjmichaud_ · x · 2026-08-27
Continuing his interpretability thread, Eric Michaud stresses skepticism given the field's unverifiability—there is a lot of slop out there. But he is curious what we'll discover if and when verifiability changes; perhaps we won't have to wait 100+ years for the 'divine questions' to be answered.
Related event: Interpretability research may take a century, researcher warns(4 posts)→
More from AGI Musings
- Expression matters more than tool choice in AI creation — sujingshen · 2026-08-27
- AI makes average content free, making human taste expensive — aigleeson · 2026-08-27
- Opinion: Human Intelligence Will Explode Before ASI Arrives — yeastsplainer · 2026-08-27
- Nick Land: The path to thinking moves to 'planetary technosentience' — SydSteyerhart · 2026-08-27
- Domain-specific ASI before AGI? Reddit debates where the line falls — Youknowwhyimherexxx · 2026-08-27
- Researcher: LLM-speak is invasive and will reshape human language symbiotically — anderssandberg · 2026-08-27