670: LLaMA: GPT-3 performance, 10x smaller

episode

Super Data Science: ML & AI Podcast with Jon Krohn not yet transcribed

▲ 0

Transcript

This episode hasn't been transcribed yet Upvotes decide what gets transcribed next — 0 so far. Sign in to upvote

Related

recent transcribed audio

Transrapid, Flugtaxis, Hyperloop – Warum es neuartige Verkehrstechnik so schwer hat 28 min · 22 Sep 2026 · Das Wissen | SWR

Las 10 de Erazno, el Chokolatazo, Parodias, Nacadas y mas 1h 25m · 22 Sep 2026 · Erazno y La Chokolata El Podcast

Cross-code immorality, Lachie Neale's strike, crowd behaviour - Andrew Voss | Across the Divide! (22.09.26) 17 min · 22 Sep 2026 · Whateley

S13 EP19: It's Exhausting Having an Opinion 37 min · 22 Sep 2026 · Parenting Hell with Rob Beckett and Josh Widdicombe

Fall in Love With the Life You Have 51 min · 22 Sep 2026 · The Candace Cameron Bure Podcast

NPR News: 09-21-2026 9PM EDT 5 min · 22 Sep 2026 · NPR News Now

Report any issue

Source

Recorded14 Apr 2023

Publisherfeeds.megaphone.fm

Handlerec:6202716

Show notes

How does Meta AI's natural language model, LLaMa compare to the rest? Based on the Chinchilla scaling laws, LLaMa is designed to be smaller but more performant. But how exactly does it achieve this feat? It's all done by training a small model for a longer period of time. Discover how LLaMa compares to its competition, including GPT-3, in this week's episode. Additional materials: www.superdatascience.com/670Interested in sponsoring a SuperDataScience Podcast episode? Visit JonKrohn.com/podcast for sponsorship information.

Ask this episode

“What did the council decide on item 42?” — answered with the speaker and the moment.

Sign in