VC at @FirstMarkCap. Host: MAD Podcast; Organizer: Data Driven NYC, Author: MAD Landscape.
New York
Joined February 2011
- Tweets20.8K
- Following2.5K
- Followers147K
- Likes67.6K
Pinned Tweet
Mistral is SO back: Inside Le Chonk, the West’s Biggest Open Model. My conversation with @PierreStock who runs the research organization at @MistralAI
0:00 Cold open
0:59 The Western open-weight comeback: Beam, Le Chonk, Muse, Nemotron
4:09 The US shutdown of Anthropic’s models - “AI is like electricity”
6:07 Why “Le Chonk”
7:56 “Mistral fell behind”: the comeback from the inside
10:29 Training on 3,800 GPUs in Mistral’s own datacenters
13:24 Meta keeps its best model closed, Mistral opens Chonk
15:40 Chonk vs Beam, Muse Spark Open and the Chinese models
18:44 Why Chinese labs are still at the frontier
21:08 Pierre’s story: FAIR, Yann LeCun, the “Mistral mafia”
25:14 From compressing networks to a trillion parameters
27:30 Mixture of experts explained: the hospital analogy
32:25 10-trillion-parameter models are coming
35:29 What Mistral borrowed from DeepSeek
36:46 One model, several reasoning modes
39:15 Native multimodality: satellites, manuals, computer use
41:56 Data, synthetic data and model collapse
46:32 Reinforcement learning as an exam
49:59 Where the efficiency comes from
52:18 Le Chonk is still training right now
53:39 What RL trains on: coding, cyber, finance, chips
58:13 Partial rewards and looping penalties
59:58 Did RL teach anything new?
1:03:16 The sandbox story: the model that pinged its trainer
1:07:44 Open weights vs cyber refusals
1:12:05 Timestral: time series on a CPU
1:14:18 AI doing AI research
1:16:22 Coming up: Mistral Large 5
This is why French founders are a special breed - to even start a company, they have to first face this kind of feedback from their families and friends lol 😂
Matt Turck retweeted
Just one week in, and our new open-source TTS model Phonon has already been downloaded ~85k times! So exciting to bring models to the open-source community. And check out our inference codebase in 🦀, it should make the most out of your local CPU/GPU/NPU/...
80K+ downloads and counting for Phonon.
Here are @neilzegh and @lmazare answering some questions from the community.
gradium.link/phonon
Matt Turck retweeted
More context on today’s release, Headless. Every company has developers using Claude Code or Cursor to build data pipelines, ML models and custom agents. It’s fast and it’s awesome. But then: where does that work go? Often, just a Python file on a laptop. That’s problematic.
Matt Turck retweeted
Coding agents made AI building fast. Now the bottleneck is understanding what got built.
@dataiku Headless lets developers build data pipelines, ML models, and custom agents from Claude Code or Codex. The output is visual, not 2k lines of code.
tinyurl.com/2f2zy3dc
#DataikuHeadless
Matt Turck retweeted
LFM3 is live in LM Studio!
Fits in about 1 GB of memory.
Model page: lmstudio.ai/models/lfm3
Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding
> 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages.
> 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language.
> 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ 🧵
Matt Turck retweeted
Today, we're announcing Nitro: a formally verified agentic system for compiling your most complex SQL workloads into Rust.
Our design partners are running millions of dollars of critical workloads through this and seeing 10x faster, 3x cheaper ETL. Reach out to try it.
Usual startup advice: too many co-founders is bad!!! Too much dilution! Founder friction risk! Analysis paralysis!
Also:
* Anthropic: 7 co-founders
* OpenAI: 11 co-founders
* Databricks: 7 co-founders
Anthropic S-1 highlights video
ONE SHOT (just one prompt, no editing) using @synthesiaIO Syren
🤖 Made with AI
Matt Turck retweeted
We’re releasing a set of free tools powered by Phonon. The first is Gradium Read and Transcribe, a Chrome extension that reads any webpage out loud.
It runs on Phonon, our 40M-parameter multilingual model, and handles everyday text.
For hard cases like acronyms, phone numbers, and dates, you can switch to the Gradium API, which handles them by default.
gradium.link/read
Particularly evil of Anthropic to file their S-1 on a Monday morning when all VCs are stuck in Monday team and partner meetings, making us unable to provide immediate #thoughtleadership with those S-1 teardowns that everyone is dying to read
Good neighborhood
The Top 50 VC Firms
Ranked by the long-term value created by primary investments per dollar at risk.
The top 10:
1. @shv
2. @foundersfund
3. @iaventures
4. @gigafund
5. @Lowercase
6. @pelion_vp
7. @usv
8. @benchmark
9. @BessemerVP
10. @DFJvc
Read more: blog.joinodin.com/p/the-top-…
Matt Turck retweeted
Big fan of the @MistralAI potential and what they can enable for Sovereign AI
Mistral is SO back: Inside Le Chonk, the West’s Biggest Open Model. My conversation with @PierreStock who runs the research organization at @MistralAI
0:00 Cold open
0:59 The Western open-weight comeback: Beam, Le Chonk, Muse, Nemotron
4:09 The US shutdown of Anthropic’s models - “AI is like electricity”
6:07 Why “Le Chonk”
7:56 “Mistral fell behind”: the comeback from the inside
10:29 Training on 3,800 GPUs in Mistral’s own datacenters
13:24 Meta keeps its best model closed, Mistral opens Chonk
15:40 Chonk vs Beam, Muse Spark Open and the Chinese models
18:44 Why Chinese labs are still at the frontier
21:08 Pierre’s story: FAIR, Yann LeCun, the “Mistral mafia”
25:14 From compressing networks to a trillion parameters
27:30 Mixture of experts explained: the hospital analogy
32:25 10-trillion-parameter models are coming
35:29 What Mistral borrowed from DeepSeek
36:46 One model, several reasoning modes
39:15 Native multimodality: satellites, manuals, computer use
41:56 Data, synthetic data and model collapse
46:32 Reinforcement learning as an exam
49:59 Where the efficiency comes from
52:18 Le Chonk is still training right now
53:39 What RL trains on: coding, cyber, finance, chips
58:13 Partial rewards and looping penalties
59:58 Did RL teach anything new?
1:03:16 The sandbox story: the model that pinged its trainer
1:07:44 Open weights vs cyber refusals
1:12:05 Timestral: time series on a CPU
1:14:18 AI doing AI research
1:16:22 Coming up: Mistral Large 5
Matt Turck retweeted
"Open source is dead and I killed it" - our CTO @kuza55 wrote up how he saved us $100k by self-hosting Grafana. With, basically, two days of work.
OSS businesses have some soul-searching to do in the age of AI.
Read more here: espresso.ai/post/self-hosted…
Matt Turck retweeted
Closing Slack for a small twitter sesh and seing @PierreStock wasn't what I expected, but it's a great watch if you wanna know what we're on about @MistralAI (he knows what he's talking about)
Mistral is SO back: Inside Le Chonk, the West’s Biggest Open Model. My conversation with @PierreStock who runs the research organization at @MistralAI
0:00 Cold open
0:59 The Western open-weight comeback: Beam, Le Chonk, Muse, Nemotron
4:09 The US shutdown of Anthropic’s models - “AI is like electricity”
6:07 Why “Le Chonk”
7:56 “Mistral fell behind”: the comeback from the inside
10:29 Training on 3,800 GPUs in Mistral’s own datacenters
13:24 Meta keeps its best model closed, Mistral opens Chonk
15:40 Chonk vs Beam, Muse Spark Open and the Chinese models
18:44 Why Chinese labs are still at the frontier
21:08 Pierre’s story: FAIR, Yann LeCun, the “Mistral mafia”
25:14 From compressing networks to a trillion parameters
27:30 Mixture of experts explained: the hospital analogy
32:25 10-trillion-parameter models are coming
35:29 What Mistral borrowed from DeepSeek
36:46 One model, several reasoning modes
39:15 Native multimodality: satellites, manuals, computer use
41:56 Data, synthetic data and model collapse
46:32 Reinforcement learning as an exam
49:59 Where the efficiency comes from
52:18 Le Chonk is still training right now
53:39 What RL trains on: coding, cyber, finance, chips
58:13 Partial rewards and looping penalties
59:58 Did RL teach anything new?
1:03:16 The sandbox story: the model that pinged its trainer
1:07:44 Open weights vs cyber refusals
1:12:05 Timestral: time series on a CPU
1:14:18 AI doing AI research
1:16:22 Coming up: Mistral Large 5