'Grad students would be so mad if they knew what actually works in training SOTA LLMs'
code_star · x · 2026-10-22
- A popular quip claims the methods that actually work in training SOTA LLMs are wildly disconnected from academic understanding—grad students would be furious if they knew.
- The author adds the same holds for many people in industry, where practitioners equally don't know what really works.
More from AGI Musings
- signull: Tech now evolves in weeks while institutions and labor markets take years to adapt — signulll · 2026-10-22
- Scientific bottlenecks: reinvent assays, don't just automate them — owl_posting · 2026-10-22
- Higgsfield CEO: 50%+ of social content will be AI-generated by end of 2027 — SimplyAnnisa · 2026-10-22
- AI academic writing's vague-language trap: it reads as consistent to you but confuses everyone else — anshulkundaje · 2026-10-22
- AI safety researchers blast chip 'red queen race' and bot-vs-bot cybersecurity — dhadfieldmenell · 2026-10-22
- 'Agentic Inequality' paper published in Nature Machine Intelligence: who captures AI's surplus agency — IasonGabriel · 2026-10-22