🇨🇦 Heliox: Where Evidence Meets Empathy — "Folding Giant AI Brains Into Your Pocket"
The AI industry has spent years telling us bigger is the only way forward. This episode traces the quiet counter-story: a proven mathematical theorem that lets engineers shrink massive models onto ordinary phones and laptops, without guessing and without dumbing them down.
We go from emergent "outlier" numbers that broke early compression attempts, through GPTQ and the Hessian matrix, to the discovery that engineers had secretly reinvented a 40-year-old lattice geometry algorithm — all the way to the linearity theorem itself.
🍎 Apple
https://podcasts.apple.com/ca/podcast/folding-giant-ai-brains-into-your-pocket/id1769969487?i=1000788142123
🎵 Spotify
https://open.spotify.com/episode/656VznGMPbuOndtbbZ3E5t?si=93697f693c8d4fb8
▶️ YouTube
https://youtu.be/3i0PdNMebbw
🎧 Listen
https://www.buzzsprout.com/2405788/episodes/19739570
📖 Read
https://helioxpodcast.substack.com/p/folding-giant-ai-brains-into-your?r=52sqs7&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
📻 Available for Broadcast on PRX
https://exchange.prx.org/p/633381
#AI #MachineLearning #Quantization #LLM #TechEthics #Science #OpenSource #EdgeAI
#quantization
2 posts · Last used Sep 07
Replying to
@Lydie@tech.lgbt i would say words per minute is not as important as model accuracy - you will have to test, great project #quantization
You've seen all posts