Only New Algorithmic Paradigms Can Drive Qualitative Leaps
willccbb · x · 2026-07-16
The author argues that if improvements like architecture search and kernel grinding are already priced in by the market, the real breakthroughs will come from new algorithmic paradigms that deviate from the traditional "pre-training → RL" pipeline.
A reply adds that many new tricks are essentially just "compute multipliers"—effective but with limited room for further exploration. Currently, one of the biggest "data multipliers" observed is still an RL algorithm from the 90s, and data scaling remains far messier and harder to achieve than compute scaling.
Related event: Researchers cool expectations around 1000x AI R&D speedups(8 posts)→
More from AGI Musings
- The Evolution of LLM Business Models: Selling Outcomes Over Tokens — yacineMTB · 2026-07-22
- Bindu Reddy says GPT-6 is coming soon, with Alibaba, DeepSeek and Kimi close behind — bindureddy · 2026-07-22
- Bindu Reddy says the industry still lacks a way to train 20T models and scale post-training RL — bindureddy · 2026-07-22
- Advanced AI Models Are Becoming Impossible to Plug and Play — emollick · 2026-07-22
- AI suggested a better composition, and that made one user uneasy — Sydde · 2026-07-22
- The Thimble and the Waterfall: AI's Data Bottleneck and Feedback Loops — dyamins · 2026-07-22