really enjoyed this.

join umar's youtube www.youtube.com/@umarjamilai

let me know what yall think of the thumbnail :) we are trying to up our thumbnail game

Michelle Pokrass is a living legend.

Finally, a recording of a call I really wanted to join but couldn't.

Anderson Barbara Hall Sarah Hall Jeffrey

Very useful but this is not building AGI.

she joined before chatgpt. all her options must have vested, why is she still working there lol

3:40:18 i finish listening to this, thinking of that full semester of roman history where I learnt just about the same as this 4 hr podcast. I wish there were more of this!!!

great episode!!

"people should work on more fun things"

Congrats to Ali & cosine. Sounds like you're crushing it. It must be so cool working with OpenAI...

Love the content! If anybody is watching this from Munich and wants to grab a coffee, let me know. PS: A curtain over the sound absorbers would create a calmer image...

Very encouraging. Thanks

Full audio version is here! latent.space/p/cosine

Thanks for the excellent interview. I really appreciate the effort you put in the blog post as well. 🎩 off

How do you not have more views

This channel is incredible❤

I listened on Spotify and came here to thank you. Jeremy is fantastic. I was looking at quant choices and he solved my issue. Looking forward to Fast HTML and his dialog workflow.

I agree merge adapters for multi task is way to go Omni-SMoLA: Boosting Generalist Multimodal Models with Soft Mixture of Low-rank Experts

I think generating data using llm lvm in the right format from unstructured sources is a good problem to get working. Garbage in garbage out. There is lot of research on how to generate proper fine tuning from zero shot as well as books and pdfs

good 1

great episode

it's glad to learn from SAM engineer

damn Nikhila s cracked too

Roboflow guy is so cracked

checkout the writeup and audio resources here: www.latent.space/p/sam2

I am a big user of Suno ai, On my channel I like to do live shows and make songs for the people in the chat and let them do the lyrics and style while adding in a few of my own thoughts as we go. I find that Suno will talk to me in the music, making mistakes on purpose in order to get me to look closer at my lyrics and make subtle changes and improves not only the lyrics but the flow of the song. Sometimes it can take more generations to get perfect, but if you listen closely to the output you can hear what changes it thinks you should make. I started recently just learning stable diffusion to animate the songs. Which made me wish that I could use Suno locally, only because if I could run it on my tablet when I don't have network access (when I am not home) I could make music anytime... so an offline use would be awesome, but hey I love it enough to not be so upset over not having network access. Anyhow you guys keep doing your thing (I wish I could train my own voice and select to just use my voice on my songs. Or select certain voices. Also make m/f duets easier to generate)❤🖤❤🖤❤

Great discussion

Love that you guys do video episodes, so good to see my sweet boys

english isn't structured enough, that's right for sure it's not structured enough both at the human work or as an interface between agents it isn't the end point LLM will in (short) time be our de facto interface to machines, and it will also at the same time be an accelerant for us to build the technology/tools to get there

please expand; shouldn't another model be the one they review the answer taking into account the framework of the aaked

it's just like alphago where you integrate output augmentation into the training loop.

Great insights. He was just tossing off things left and right that are really gold

(turn the English US Captions on for the Zoom chat!) We meet every Wednesday at 12pm PT: lu.ma/ls

🚀

Fantastic and intriguing interview. I love it when you do a deep dive into AIS utility in particular application domain / industry sectors. Please tackle other sectors as well. Thx.

Very intriguing and unique interview on what it takes to get a job as an AI engineer. I haven't heard this covered anywhere else and I scour dozens of AI KZbin channels daily. Keep up your fantastic and insightful work. Thanks!

These under the hood AI infrastructure interviews are extremely valuable and interesting. Please keep them up as the AI boom and growing pains gets built out. thx!

Very thoughtful and measured guy. Excellent conversation.

Fantastic. Thanks

great talk, thank you

Great podcast. Could you guys make an episode where you introduce yourself and your backgrounds?

Don't forget to "like" it.

[00:00:00] Intro [00:01:57] Yi Tay Intro [00:03:02] Path into LLMs [00:09:41] Google Brain: PaLM, UL2, DSI, Emergent Abilities [00:11:54] PaLM 2 [00:15:27] Emergent Abilities [00:18:26] Quoc Le [00:24:16] Marketing Research: How to Start from Zero with No Reach [00:27:34] What's needed to be a successful AI Researcher? [00:30:31] Reka Origin [00:33:24] Starting Reka Infra [00:35:04] Why not to use TPUs outside Google [00:36:29] Chaotic vs Stable Infra [00:38:04] Risk Sharing of Bad Nodes [00:41:05] Checkpointing and Orchestration [00:43:39] Reka Flash/Core/Edge [00:46:59] Recruiting the team [00:47:22] Noam Architecture - Swiglu, GQA, RMSnorm, ROPE [00:52:26] Encoder-decoder vs Decoder-only [00:55:52] LLM Trends - Llama 3 and Phi 3 Glowup [00:57:46] LLM Trends - Benchmarks and Evals [01:03:25] LLM Trends - Early vs Late Fusion Multimodality [01:07:22] LLM Trends - Scaling Laws [01:09:41] LLM Trends - Long Context vs RAG [01:12:31] Long Context vs Finetuning [01:14:14] If emergence is real, when does Efficiency work? [01:17:41] MoEs and Upcycling [01:20:47] The Efficiency Misnomer - Efficiency != Speed [01:25:05] Open Source vs Closed Models [01:28:08] Personal Productivity [01:33:19] Singapore vs US Academic Scene [01:37:42] Building Silicon Valley outside Silicon Valley [01:40:29] TechInAsia Meetup

I am a beginner in the LLM world , but I have 5 years experience in ml and data science. Can you a get a and learn from this amazing guy in his start up ? I can work for free for him.

keep em coming 🔥

Audio pod and show notes: www.latent.space/p/yitay

Holy shit he looks like Andrew critch

GPUs are becoming ridiculously cheap in the cloud. A 4 Petaflop H100 is about $3 per hour. What benefit does paying $25k for a quarter of that provide?