Alex Mathé-Cathala

Looking forward to speaking about AI and money next week in the heart of Silicon Valley. No better place to have this conversation than where institutional and personal capital is being created and reallocated at a pace we haven't seen since the late '90s.

We’re wrapping up 2025! We’re so grateful for your support and feedback as we build Vega: 📨 860,000 email were drafted by the Vega AI this year 🤝 114,000 meetings ran through Vega 💬 80,000 chat messages sent A heartfelt thank you to our customers and partners for making our annual beachside dinner such a vibrant and meaningful event. Much more to come in 2026.

Nice Vega hat, coach! 💜 #VegaMinds

We all had a great time yesterday at Vega HQ during our annual poolside dinner! I'm so grateful to our users, friends, and advisors who joined us to celebrate. We’ll be back with Vega Poolside v3 on September 8, 2026. Save the date!

Daily pier walk. Special one and fresh haircut tho because @FutureProof_HQ IS! COMING! TO! TOWN! 🌊 The Vega team is hosting our annual Poolside at the Vega HQ. v1 last year was a blast and v2 is leveling up: smoky beef + coconut shrimps + wine + more. To RSVP email contact@vegaminds.com. @jasonwenk, @MichaelKitces, and @Downtown have RSVP’d already!!!

Excited to join three industry leaders at @US_SIF’s #Forum2025 in DC next week!

We're proud to announce our partnership with Mutual Group. Vega Minds will deliver our AI assistants to over 1,600 finance professionals managing over $13 billion in assets. Join us next Thursday to learn more. #AI #VegaMinds #FinTech

Presenting Vega Minds to advisors in Vegas #AI #FinTech

I really like @AnthropicAI's release notes. Last week, the lab made the latest versions of their Claude models available to the public. Official communications keep emphasizing Opus and Sonnet's coding skills. Devs generally agreed that Sonnet 3.5 and 3.7 were the best base models for a majority of programming tasks. But Anthropic's developer audience isn't the only aspect that sets the lab apart. I've found Anthropic’s communication about their vision, challenges, and AI alignment research currently more engaging than @OpenAI's. They're setting the bar for everyone else. They've been publishing the system prompts behind each new model since June of last year. The system prompts are "hidden" instructions for chat assistants. Here's the link: docs.anthropic.com/en/release-not…. I love digging into these prompts. They teach us two things: 1) The numerous ways chat assistant misbehaved *until* Anthropic encoded specific instructions to "ground" its behavior. 2) What the best practices for using chat assistants are. Just a few system prompt highlights with my comments: 📖 "Claude never starts its response by saying a question or idea or observation was good, great, fascinating, profound, excellent, or any other positive adjective. It skips the flattery and responds directly." 👉👉👉 LLMs have a natural bias towards extreme sycophancy. OpenAI had to release a postmortem after GPT-4o. Here's the link: openai.com/index/sycophan…. 📖 " There was a US Presidential Election in November 2024. Donald Trump won the presidency over Kamala Harris. [...] Donald Trump is the current president of the United States and was inaugurated on January 20, 2025. Donald Trump defeated Kamala Harris in the 2024 elections. Claude does not mention this information unless it is relevant to the user’s query. " 👉👉👉 Election results have been debated so extensively that the training data fed to LLMs likely overly emphasized that Trump was not president from 2021 onward. Now that he is, Anthropic probably noticed that Claude was continuing on that trend to the point where it was probably contradicting the results of the last election, and they had to hardcode it. 📖 "In general conversation, Claude doesn’t always ask questions but, when it does, it tries to avoid overwhelming the person with more than one question per response." 👉👉👉 LLMs aren't really good at that for now. Enabling human supervision effectively is the next frontier. There's a lot more. A fun read!

Excited to be speaking at @US_SIF’s #Forum2025 & joining the US’s largest gathering of #sustainableinvesting professionals. My panel "Advising in the AI Era" will be on June 26th. Don't miss it!

Had a unique time today sharing technical machine learning concepts with a general audience. Grateful for the opportunity to present Vega in SoCal, just one month after our NorCal session. Thanks, @fpassociation!

certain it won't be universal and it will be less than basic.

in annual contributions over 5 years, the real cost to manage just $150k exceeds $5k annually... I like that the pitch touches on capital and forces us to think about wealth inequality. But if you trust the markets to figure out """Universal Basic Capital""" you can be about

High taste. Industry-vetted backers (Ackman, Henry Kravis). Good idea-founder fit. Yet, it's hard to refrain from dunking on it. Leveraging equities is unorthodox for a good reason. Simply quoting Abdul: "If you want to buy a house, you take a mortgage. If you want to buy a

@TrmnChbn #BusinessAsUsual

@MatheCathala Ah tu t’es remis à tweeter en costume mode linkedin

If this is the pitch, run

@burkov If material information about one stock was in the news, a thousand hft shops already moved its price before your txt file even started downloading

If a 10M-token context worked, you could download all of yesterday's financial news, put them into the context, and train the LLM on today's S&P 500 stocks to predict whether the stocks will go up or down next day.