Jagged frontier persists: models still struggle to abstract novel concepts into working patterns
Dan_Jeffries1 · x · 2026-09-23
Dan Jeffries warns that fast improvements on known tasks can fool you into thinking the jagged frontier is gone. Skip hardcoded demo classifiers or repurposed game state — try extracting genuinely novel, generalizable uses from the latest general-purpose models (which he calls amazing and powerful). You'll find they still have a very jagged frontier and struggle with abstracting new concepts into working patterns.
More from AGI Musings
- Yarin Gal's deadpan answer on LLM-written papers: "any text written by an LLM" — yaringal · 2026-09-23
- Few People Have Internalized That Longevity Escape Velocity Is Near — LesaunH · 2026-09-23
- Quanta Explains How Pricing Algorithms Can Drive Up Prices Without Collusion — burny_tech · 2026-09-23
- Model release fatigue: devs say gains are marginal, but some argue new models clearly outpace old ones — Rasmic · 2026-09-23
- Autonomous driving's data moats will fall to world models and open source within a decade — DominiqueCAPaul · 2026-09-23
- AI commentator doubles down: model progress 'all true', citing Terence Tao's o1 graduate-student benchmark — RichmanRonald · 2026-09-23