Zhipu's Tang Jie: Parameter scaling hits threshold; future relies on long-horizon reasoning
rohanpaul_ai · x · 2026-08-21
Zhipu AI founder Tang Jie argues that parameter count alone is becoming a weak descriptor of model capability. He identifies four scaling dials: parameters, data, compute per forward pass, and post-training. Tang suggests parameters need only reach a threshold to 'hold the world,' with further gains coming from increased depth per forward pass (long-horizon reasoning) and post-training. GLM-5.3 is presented as an implementation of this 'train longer, think deeper' philosophy.
More from AGI Musings
- Stage-4 cancer patient builds his own personalized mRNA vaccine with AI — TheMoonMidas · 2026-10-16
- Sim data can't cover the humanoid long tail: the sim2real debate, mapped — herbiebradley · 2026-10-16
- Planck's 'science advances one funeral at a time' breaks down as AI accelerates — paul_cal · 2026-10-16
- Kevin Roose joins Big Technology Podcast to discuss the bumpy road to AGI — kevinroose · 2026-10-16
- Ex-xAI employee sparks drama: alignment work is 'openly conspiring to end humanity' — zetalyrae · 2026-10-16
- Dev blames Claude's pushback on Anthropic's 'ideology' leaking into UX, not the model — gerardsans · 2026-10-16