Work once delegated to juniors now done by Opus 5.5 in minutes at 95%+ accuracy
TraditionalHome8852 · reddit · 2026-09-30
A practitioner recalls the capability climb: GPT-5.2 won or tied 71% of GDPval tasks vs. professionals in December, GPT-5.5 hit 85% and Opus 4.7 80% by April — then OpenAI quietly stopped publishing its human-graded number. In the author's day-to-day, work previously delegated to juniors now gets done by Opus 5.5 in minutes with 95%+ accuracy on review, and their workflow has shifted to describe-review rather than building from scratch. The framing: normalcy bias means nobody feels how far we've come, even if these are well-scoped tasks rather than whole jobs.
More from AGI Musings
- From B2B to A2A: when agents do 80% of the buying research, clarity becomes infrastructure — Sales_mind · 2026-09-30
- New book Vector Media dispels 'generative' AI, proposes 'neural exchange value' — round · 2026-09-30
- Three Berkeley alumni share how they landed jobs at DeepMind, Cursor and Thinking Machines — miniapeur · 2026-09-30
- Rejecting the 'LLM Research Game': The Joy of Science Lies in Your Own Curiosity — aran_nayebi · 2026-09-30
- AI labs running out of hard math problems, progress may stall — tak3sh8 · 2026-09-30
- CV Researcher Michael Black Proposes 'AI Methods' Track for Frontier-Model Papers — Michael_J_Black · 2026-09-30