Julia Turc's banner

Julia Turc

@juliarturc • 25,234 subscribers

Explaining AI on YouTube • YC S24 Founder • Ex-Google Research • Eastern-European nihilist & American optimist

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

This is what happens when you plug LLMs into voice assistants, instead of a decade of handwritten rules. This video dissects Voxtral (a family of OSS speech models) and the foundational work behind it (audio tokenization, semantic/acoustic disaggregation, etc). Thank you Mistral AI for your collaboration and for your detailed technical reports in an increasingly opaque industry! 00:00 Intro 01:03 Modular vs end-to-end speech models 03:30 Speech-to-Text 06:07 Delayed Streams Modeling (DSM) 09:41 Whisper Streaming 10:33 Voxtral Realtime 13:07 Voxtral Text-to-Speech 14:28 Throwback: WaveNet 15:24 Audio tokenization 20:39 The Voxtral Codec 21:49 Back to Voxtral TTS 25:30 Outro

This is what happens when you plug LLMs into voice assistants, instead of a decade of handwritten rules. This video dissects Voxtral (a family of OSS speech models) and the foundational work behind it (audio tokenization, semantic/acoustic disaggregation, etc). Thank you Mistral AI for your collaboration and for your detailed technical reports in an increasingly opaque industry! 00:00 Intro 01:03 Modular vs end-to-end speech models 03:30 Speech-to-Text 06:07 Delayed Streams Modeling (DSM) 09:41 Whisper Streaming 10:33 Voxtral Realtime 13:07 Voxtral Text-to-Speech 14:28 Throwback: WaveNet 15:24 Audio tokenization 20:39 The Voxtral Codec 21:49 Back to Voxtral TTS 25:30 Outro

298,327 次观看 • 3 个月前

"World models" is one of the buzziest yet ambiguous terms in AI right now. I started this video with many questions: - How are they different from video generation? - Can they do more than AI slop? - Can LeCun be trusted given that he wears knee-high white socks? Many thanks to TJ Galda and NVIDIA AI for helping me answer (most) of these questions!

"World models" is one of the buzziest yet ambiguous terms in AI right now. I started this video with many questions: - How are they different from video generation? - Can they do more than AI slop? - Can LeCun be trusted given that he wears knee-high white socks? Many thanks to TJ Galda and NVIDIA AI for helping me answer (most) of these questions!

248,493 次观看 • 3 个月前

How far are we from "Her"? I couldn't help but wonder, given the rapid fire of GPT-Live and Thinking Machines. Turns out, full duplex speech-to-speech is not particularly new. Moshi has been around for more than 2 years. Many thanks to Neil Zeghidour for bringing so much nuance. Check out kyutai who are doing god's work and open-sourcing it (not sponsored!). 00:00 Intro 01:24 ChatGPT Standard & LLM-based cascades 04:08 ChatGPT Advanced & end-to-end speech models 05:35 ChatGPT Live & Her 07:38 Thinking Machines & multi-stream modeling 09:13 Moshi & full duplex assistants 13:24 Moshi’s inner monologue 15:38 Why aren’t full duplex models everywhere? 17:08 Tool calling in full duplex models 18:00 Moshi RAG 19:06 Is the future full duplex? 20:24 How far are we from Her?

How far are we from "Her"? I couldn't help but wonder, given the rapid fire of GPT-Live and Thinking Machines. Turns out, full duplex speech-to-speech is not particularly new. Moshi has been around for more than 2 years. Many thanks to Neil Zeghidour for bringing so much nuance. Check out kyutai who are doing god's work and open-sourcing it (not sponsored!). 00:00 Intro 01:24 ChatGPT Standard & LLM-based cascades 04:08 ChatGPT Advanced & end-to-end speech models 05:35 ChatGPT Live & Her 07:38 Thinking Machines & multi-stream modeling 09:13 Moshi & full duplex assistants 13:24 Moshi’s inner monologue 15:38 Why aren’t full duplex models everywhere? 17:08 Tool calling in full duplex models 18:00 Moshi RAG 19:06 Is the future full duplex? 20:24 How far are we from Her?

73,293 次观看 • 1 个月前

For a meager $250/mo, you too can be a paper plane flying above a cascade while burning a small village's worth of electricity. Really impressive though, it's literally real-time in near-720p.

For a meager $250/mo, you too can be a paper plane flying above a cascade while burning a small village's worth of electricity. Really impressive though, it's literally real-time in near-720p.

136,847 次观看 • 4 个月前

This is why we need to gate-keep science. Two pseudo-intellectuals thinking they discovered something deep, conflating >social attention >transformers attention >quantum physics observer (attention) These have nothing in common, other than the ambiguity of English language. Naked ladies on Instagram have nothing to do with a weighted average followed by softmax. But they're both so mind-blown by their discovery. Dunning–Kruger will only get amplified by AI sycophancy. Please call me out if you see me going beyond my own DK threshold.

This is why we need to gate-keep science. Two pseudo-intellectuals thinking they discovered something deep, conflating >social attention >transformers attention >quantum physics observer (attention) These have nothing in common, other than the ambiguity of English language. Naked ladies on Instagram have nothing to do with a weighted average followed by softmax. But they're both so mind-blown by their discovery. Dunning–Kruger will only get amplified by AI sycophancy. Please call me out if you see me going beyond my own DK threshold.

119,734 次观看 • 4 个月前

Early on, Jonathan Ross foresaw the unprecedented demand for compute of the 2020s. After pioneering Google's TPUs, he founded Groq, a "Language" Processing Unit dedicated for inference. After joining NVIDIA, he is betting on the GPU+LPU combo to power the agentic chains of AI calling AI calling AI. This interview is part of my larger series on dedicated AI chips, so drop any suggestions/questions below. Thank you NVIDIA for hosting us at your HQ! 00:00 Intro 01:00 The Google TPU & Groq origin story 02:41 How is Groq different from a GPU? 04:43 Static scheduling makes Groq faster 05:47 Does Groq work with Mixture-of-Experts? 09:27 Are LPUs limited to text models? 11:03 Diffusion models 13:41 NVIDIA Vera Rubin: the GPU+LPU combo 15:26 Will Groq still be sold as a standalone chip? 16:49 How does agentic AI impact inference economics? 19:21 Will AI replace CUDA kernel engineers? 21:08 Will AI democratize hardware design? 26:11 Jevon's paradox: An endless demand for compute 29:04 What should kids learn in the AI age?

Early on, Jonathan Ross foresaw the unprecedented demand for compute of the 2020s. After pioneering Google's TPUs, he founded Groq, a "Language" Processing Unit dedicated for inference. After joining NVIDIA, he is betting on the GPU+LPU combo to power the agentic chains of AI calling AI calling AI. This interview is part of my larger series on dedicated AI chips, so drop any suggestions/questions below. Thank you NVIDIA for hosting us at your HQ! 00:00 Intro 01:00 The Google TPU & Groq origin story 02:41 How is Groq different from a GPU? 04:43 Static scheduling makes Groq faster 05:47 Does Groq work with Mixture-of-Experts? 09:27 Are LPUs limited to text models? 11:03 Diffusion models 13:41 NVIDIA Vera Rubin: the GPU+LPU combo 15:26 Will Groq still be sold as a standalone chip? 16:49 How does agentic AI impact inference economics? 19:21 Will AI replace CUDA kernel engineers? 21:08 Will AI democratize hardware design? 26:11 Jevon's paradox: An endless demand for compute 29:04 What should kids learn in the AI age?

46,967 次观看 • 3 个月前

After a few weeks of Flow Matching, I still think it's bonkers. But also so so clever. I wish more of ML was like this, instead of the typical "let's add an extra layer here and there and see what happens". Link below.

After a few weeks of Flow Matching, I still think it's bonkers. But also so so clever. I wish more of ML was like this, instead of the typical "let's add an extra layer here and there and see what happens". Link below.

48,204 次观看 • 5 个月前

Why are Transformers replacing CNNs? I wanted to understand if the hype around Transformers is the main driver behind this shift. But it turns out there are strong theoretical reasons for it. Cynicism loses this time.

Why are Transformers replacing CNNs? I wanted to understand if the hype around Transformers is the main driver behind this shift. But it turns out there are strong theoretical reasons for it. Cynicism loses this time.

53,056 次观看 • 9 个月前

Diffusion models clicked for me when I started seeing them through the lens of particle motion. I built this interactive playground where you too can clickety-clack to understand how drift, noise, and other hyperparams control diffusion. I hereby submit this as penance for the sin of YouTube edu-tainment 😇 Link in the first comment.

Diffusion models clicked for me when I started seeing them through the lens of particle motion. I built this interactive playground where you too can clickety-clack to understand how drift, noise, and other hyperparams control diffusion. I hereby submit this as penance for the sin of YouTube edu-tainment 😇 Link in the first comment.

32,207 次观看 • 6 个月前

A lot of people think that diffusion LLMs == BERT. But masked dLLMs are just one form of text diffusion, popularized by Subham Sahoo and his group at Cornell. There's much more to explore in text diffusion land! Check out our conversation (link below)!

A lot of people think that diffusion LLMs == BERT. But masked dLLMs are just one form of text diffusion, popularized by Subham Sahoo and his group at Cornell. There's much more to explore in text diffusion land! Check out our conversation (link below)!

26,124 次观看 • 5 个月前

An H100 GPU can run 10^15 FLOPs/s. But somehow, a model like Gemini Flash-Lite is stuck at 200 tokens/s. These numbers show just how much of a bottleneck memory bandwidth is. In this video, we look at how LLMs are evolving to work around this limitation. Thanks Inception for sharing your insights!

An H100 GPU can run 10^15 FLOPs/s. But somehow, a model like Gemini Flash-Lite is stuck at 200 tokens/s. These numbers show just how much of a bottleneck memory bandwidth is. In this video, we look at how LLMs are evolving to work around this limitation. Thanks Inception for sharing your insights!

26,110 次观看 • 6 个月前

Diffusion clicked for me when I read about score-based models, a line of work pioneered by Stefano Ermon (et al.) at Stanford. So it was a full-circle moment to collab with him and Inception on a video about training & sampling techniques for making diffusion LLMs faster.

Diffusion clicked for me when I read about score-based models, a line of work pioneered by Stefano Ermon (et al.) at Stanford. So it was a full-circle moment to collab with him and Inception on a video about training & sampling techniques for making diffusion LLMs faster.

27,186 次观看 • 7 个月前

When reading diffusion papers, my most common reaction is "mkay the math works out but WHY WOULD A SANE PERSON CHOOSE TO DO THIS". This included the (foundational) DDPM formulas. The saving grace is that you can visualize it as 2D particle motion and get a solid intuition.

When reading diffusion papers, my most common reaction is "mkay the math works out but WHY WOULD A SANE PERSON CHOOSE TO DO THIS". This included the (foundational) DDPM formulas. The saving grace is that you can visualize it as 2D particle motion and get a solid intuition.

20,528 次观看 • 6 个月前

Just published a new Flow Matching tutorial (link below). It's a little unorthodox, so I want to hear your feedback before turning it into a video. On the surface, the FM algorithm looks totally arbitrary: 🫤Pick two arbitrary points 🫤(Arbitrarily) decide to draw a straight line 🫤(Arbitrarily) model particle moving at constant speed 🫤Pick an arbitrary point in time 🫤Evaluate the velocity and use it as training label 🤯Mathematically totally valid Hopefully this tutorial will turn all of these arbitrary-looking things into intuition.

Just published a new Flow Matching tutorial (link below). It's a little unorthodox, so I want to hear your feedback before turning it into a video. On the surface, the FM algorithm looks totally arbitrary: 🫤Pick two arbitrary points 🫤(Arbitrarily) decide to draw a straight line 🫤(Arbitrarily) model particle moving at constant speed 🫤Pick an arbitrary point in time 🫤Evaluate the velocity and use it as training label 🤯Mathematically totally valid Hopefully this tutorial will turn all of these arbitrary-looking things into intuition.

17,970 次观看 • 5 个月前

没有更多内容可加载