Ex-Google Brain researcher: key algorithmic wins were found under 1e20 FLOPs, then scaled to 1e25
_arohan_ · x · 2026-09-07
Former Google Brain member arohan notes that many innovations from his Google Brain era that made it into frontier models — now credited as 'algorithmic progress' — were discovered with less than 1e20 FLOPs of compute and only later scaled to 1e25 FLOPs. The observation implies major algorithmic breakthroughs don't require massive compute; small-scale experiments can yield frontier-shaping methods.
More from Infra
- DeepSpeed shows AutoEP/AutoTP/AutoSP parallelism gains, hosts office hours — PyTorch · 2026-10-27
- Cloudflare Durable Objects validated as top architecture for AI agent WebSocket servers — tristanbob · 2026-10-27
- Gary Marcus: With Q3 data added, Anthropic's compute-efficiency growth shows deceleration — GaryMarcus · 2026-10-27
- Inference is just Kubernetes with a multi-trillion-dollar market, and the communities overlap — matt_slotnick · 2026-10-27
- vLLM eyes Rust migration as inference engines race to rewrite the stack — AccBalanced · 2026-10-27
- Resin lets an agent build standalone LLM inference engines; vLLM and SGLang optimize DeepSeek kernels differently — AccBalanced · 2026-10-27