Product, AI Research @ Meta - 3x founder CEO + exits.

San Francisco, CA
Joined May 2010
AI — Metrics and Learnings from a real-world Power User by @newtechmitch blog.presence.ai/ai-metrics-…
4
Michel Meyer retweeted
Official ProgramBench leaderboard update: Our first open weight model we evaluated, GLM 5.2, scores an impressive 8.5% almost resolved, achieving 3rd place overall.
5
6
1
28
3,719
Michel Meyer retweeted
ProgramBench is now accepting submissions to the official leaderboard! We also open sourced *all* of the 2000+ agent trajectories and codebases from our paper’s main results.
4
12
1
53
4,786
Michel Meyer retweeted
This year was huge for the LeRobot community! 💛🤗
3
21
3
214
28,862
👉 Relentless it is. Yes. This is our 3rd launch in 6 weeks at Yuma, and once again we’re pushing right to the edge of what AI can do for ecommerce. Today’s release is a big one: 🤩 Deep Search 🤩 For years, ecommerce brands have sat on a goldmine of insights hidden inside tens of thousands of support tickets. The real Customer Voice lives there: what customers think, what’s breaking, what’s confusing, what’s blocking purchases… everything. But no team can afford to manually read thousands of tickets. Dashboards only scratch the surface. Critical issues stay buried until it’s too late. Before the AI era, turning support conversations into actionable insights simply wasn’t possible. ⚡️Now it is. Deep Search is a ChatGPT-like interface on top of your entire support ticket cloud. You ask a question in plain language, like: • “Why are customers abandoning checkout this week?” • “Which products are causing defects?” • “Are refunds spiking for a particular SKU?” • “What issues are trending that we haven’t spotted yet?” Deep Search scans all support conversations in seconds, groups and ranks the main themes, and gives you clear insights backed by real ticket snippets. You can ask follow-ups naturally. You can save any question as a recurring insight and get it emailed to you automatically. Insights that used to take days of manual work now take minutes. 🕐 Brands can finally spot product defects earlier, understand sentiment shifts, diagnose friction in the purchase journey, and improve support ROI across the board. This is the first time Customer Voice becomes accessible at scale. Proud of the team for pushing so hard this quarter. Another step toward our mission: AI that not only executes work, but explains what is happening and why. Let's go @Yuma_dot_AI team! 📈
6
4
26
3,686
Michel Meyer retweeted
So excited we finally get to share 🚀🔥Meta Code World Model! 🔥🚀 ✅ 32B dense open-weights LLM 💪 Strong coding skills 🧪 CWM enables a TON of cool research on world models for code generation! We would really like you to check it out, have some fun, and get inspired! 😊
(🧵) Today, we release Meta Code World Model (CWM), a 32-billion-parameter dense LLM that enables novel research on improving code generation through agentic reasoning and planning with world models. ai.meta.com/research/publica…
1
6
29
2,012
👋 Friends! Today is a big day for Yuma. It's the start of a 6-week marathon where we'll be announcing 3 major milestones. Each one confirming even more Yuma's edge in AI for ecommerce. To get started, I'm thrilled to officially announce that Yuma is launching its 2nd product line: Sales AI. This isn't just "conversational AI", it's something more profound. 90% of shoppers don't want to chat; they simply want the right information, at the right time, to tip them over the edge and convert. And this is exactly what Sales AI delivers! ➡️ The Eye-Catching Quote One of our early partners, Parachute Home, credits Sales AI with improving both customer experience and revenue. "Our shoppers love the interactive FAQ widget. They get answers fast, and on desktop, in one A/B test, we saw our Revenue Per Visitor increase +18% and our Average Order Value +4%," says Mehdi Ait Oufkir, CEO of Parachute Home. ➡️ The Story Yuma AI's customer service AI orchestration platform has automated over 5 million customer inquiries for leading brands since 2023. For 2 years, we watched a pattern repeat. A lot of "support" messages were really pre-sale questions that should be answered right on the product page, not through support channels. So we built Sales AI for shoppers to get answers to their burning questions right on the PDP, boosting Revenue per Visitor and AOV by up to 18% and 4% respectively in controlled A/B tests. ➡️ What it is In its simplest form, It’s a Q&A widget for product pages that answers relevant questions in real time based on shoppers’ behavior. Sales AI works with any Shopify and Shopify Plus store. In short: - Increase your Shopify store's Conversion Rate (CR) by up to 6%; boost AOV and Revenue per visitor. - AI-powered product FAQ widget tailored to your brand; that auto-improves based on performance data. - Clear path to customer conversion by answering their most burning questions. - Discover brand-new shopper insights, like what's preventing conversions. - 2 clicks to install, no setup time required. Works with any A/B test tools. That's it for today! At @Yuma_dot_AI , we are forging ahead with an incredible team, transforming the AI + Retail landscape, and this is just the beginning. Stay tuned for our next 2 big announcements, and if you're looking for the next industry disruptor, we're hiring! 📈
8
7
2
45
37,649
Michel Meyer retweeted
check out cwm!! open weights and inference code, many fun details to ponder from the report, and most importantly a lot of new research directions opened :)
(🧵) Today, we release Meta Code World Model (CWM), a 32-billion-parameter dense LLM that enables novel research on improving code generation through agentic reasoning and planning with world models. ai.meta.com/research/publica…
3
44
6,130
Michel Meyer retweeted
Code World Model: producing code by imagining the effect of executing instructions and planning instructions that produce the desired effect.
(🧵) Today, we release Meta Code World Model (CWM), a 32-billion-parameter dense LLM that enables novel research on improving code generation through agentic reasoning and planning with world models. ai.meta.com/research/publica…
67
166
14
1,777
241,906
Michel Meyer retweeted
We’re excited to share our preparedness report on Code World Model (CWM), FAIR’s latest open-weight model for code generation and reasoning. This report was developed by the SEAL team and the AI Security team, marking our first external publication since part of SEAL joined Meta just 1.5 months ago. It reflects our ongoing commitment to safety and alignment as we look ahead to future open-source and frontier models. We’re actively working to broaden our evaluation coverage and improve our processes, and we welcome collaboration and feedback from the broader research community. For more details, please read the full report: ai.meta.com/research/publica…
5
19
2
140
55,093
new research from Meta FAIR: Code World Model (CWM), a 32B research model we encourage the research community to research this open-weight model! pass@1 evals, for the curious: 65.8 % on SWE-bench Verified 68.6 % on LiveCodeBench 96.6 % on Math-500 76.0 % on AIME 2024 🧵
94
147
37
1,399
869,971
Michel Meyer retweeted
LLMs for code should do much better if they can iterate on tests -- but they don't. Our new work (RLEF) addresses this with execution feedback at RL *training time* to use execution feedback at *inference time*. arxiv.org/abs/2410.02089 is just out! 1/6
6
87
13
333
139,418
Michel Meyer retweeted
Since joining @a16z, I've seen a lot of otherwise 👍 decks with 👎 metrics practices, including: - vanity metrics - non-standard metrics definitions - data visualization tricks - etc. Here are the biggest mistakes I see alongside helpful entertainment-themed companion graphs👇
34
149
25
1,140
Michel Meyer retweeted
From one of our CEOs - an explainer of why change management often goes sideways.
60
638
70
5,233
Michel Meyer retweeted
Since we're back to 2019 valuations, may as well bring back the #SaaS Funding Napkin via @chrija of @PointNineCap #vc
37
305
24
2,555
Michel Meyer retweeted
Elon’s 10-year Tesla options were about to expire, so he had to sell them. The Twitter bid did allow him to do that without his facing questions about why he was selling. And he sold at an excellent price!
entire thing was clever ruse to SELL + LIQUIDATE $8.5 BILLION of TESLA STOCK (w/plausible excuse for doing it) 80% odds Elon pays $1B breakup fee + walks away with $7.5B liquidated 20% spends $100M fighting litigation honestly think he can “land rockets” but can’t fix ‘bots’?
44
186
50
872
Michel Meyer retweeted
I wish I had read this 2013 note from Zuck when we were acquired by FB/Messenger one year after. To this day nobody has pulled off the "Messenger Agent Platform" vision he described in his point #4. (@acheyer got close with Viv, but they were acquired by Samsuns too early...)
Mark Zuckerberg on Messenger December 9, 2013
2
1
13
Michel Meyer retweeted
Investing as LPs into small funds is the new angel investing
37
47
10
679
At one point during his emotional presentation, Zelensky paused to ask one of the American lawmakers to mute themselves. “Senator Rick Scott, please mute your mic,” he said.
1,072
5,006
1,499
32,968
Michel Meyer retweeted
1/ last quarter, i met with 114 engineers, made 86 introductions, & closed 6 candidates for portfolio companies. while each conversation is unique in its own way, there's one topic comes up without fail: the mysterious role of a “founding engineer” 🔮
16
40
10
404
Michel Meyer retweeted
Exclusive: inside Kharkiv’s devastated Freedom Square
243
5,260
513
15,330