Karpathy Thinks 1B Will Be Enough: TwIL-8B, Checked on a Mac Studio
From Frontier Watch · 9:53
Show by Guerin Green
A viral post quotes Andrej Karpathy saying today's models waste capacity remembering things and that a one-billion-parameter model can beat most of them. Most of that is his, from the No Priors podcast in September 2024; the 90% and the comparison with today's models are not. webAI's TwIL-8B is held up as proof: its 93/92/90 percent-of-frontier scores are webAI's own, its licence is non-commercial research and education only, and in our spot check on a Mac Studio it passed 164 of 168 checks but took twice as long per answer as DeepSeek V4 Flash. Full write-up: https://cogcore.novcog.us.com/ · Video: https://youtu.be/8wW9LZ1UfuE
More episodes of Frontier Watch
More episodes
- A New Website Gets Bots in Seconds: My Research Paper in Plain English
- They Watched the Second Lap: My AI Research Paper in Plain English
- Does AI Training Teach New Reasoning? A New Paper Says No. I Re-Ran It
- InferSSD on a 256 GB Mac Studio, Tested: Plain llama.cpp Wrote 3x Faster
- WiCi One, Checked: A 748B AI Off an SSD vs the Mac Studio We Already Run
- AI Agent Costs, Checked: Same Price Per Token, 8x the Bill
- Semantic Caching, Checked: What It Saves, Where It Breaks, and the Threshold That Decides
- Kimi K3.1 Pro: #4 in the World, Called Open, and You Cannot Download It Yet
- Is GEO Dead? I Checked Every Study Used to Bury It
- Linking Two Macs for AI: What EXO and RDMA Actually Do
PodSearch is a free, independent podcast search engine. Read about PodSearch, the privacy policy or contact us.