AI trainer's pivot: models now do 'most things', so the bottleneck is the idea
beglen · x · 2026-11-21
A reflection from an AI training practitioner as ChatGPT turns four. His old course opened with "language models are good at four things: thinking, writing, synthesising, teaching," then had learners hunt for matching tasks — a framework that worked when tools were brilliant at some things and useless at others.
No longer: ask what a model can help with today and the answer is most things, so the list helps nobody. AI used to be the hard part; now the blocker is the scale of the idea and the willingness to pursue it, so he's redesigning his training from a new starting point. Directly useful for anyone doing AI education or enablement.
More from AGI Musings
- Reading old books matters: without them you're a hostage to the present — 0xkarasy · 2026-11-21
- Two years after mocking 'AI proved RH' claims, skeptic updates: now we do — burny_tech · 2026-11-21
- 'AI just solved math's hardest problem, but surely my mid-level accounting job is safe' — birchlse · 2026-11-21
- Coping with AI math results: calling millennium-problem solutions 'brute force' misses the point — burny_tech · 2026-11-21
- 'Culture chomper for the cold utilitarian machine god': X users debate OpenAI vs Anthropic vibes — akbirthko · 2026-11-21
- '70 Days Until All Math Is Solved': Viral Millennium Problems Chart Pokes Fun at Claude — burny_tech · 2026-11-21