Clara Cornu @clastiche

bringing the berlin club queuing to nyc

Prior Labs @prior_labs

2 days ago

🌌 Yes, we already told you about our agentic night in New York. No, we’re not over it. More nights like this are coming, stay tuned!

0 0 5 448 0

first tweet, 28 lucky followers, here we go gtc week, lots of conversations and I keep coming back to this piece. Europe needs AI labs, not just the applicative layer. Very inspiring to see more and more of them: @elevenlabsio, @bfl_ml, @PriorLabs (💙) more of these please!

Arthur Mensch @arthurmensch

2 weeks ago

Should Europe leave frontier AI to the Americans and the Chinese? The idea has gained ground in Europe. It is a fundamental mistake. Build or surrender: for now, the choice is still ours. Full article below. x.com/i/article/2110…

268 766 4K 439K 1K

The difference between Jev and TFMs boil down to zero-shot classification vs grounded predictions. But because the two share a lot of keywords in descriptions, we get a lot of questions about how the two are related! So I've been sneaking it into my talks over the last few weeks! 𝐉𝐞𝐯: zero-shot classifications for light-weight and cheap decision routing and (possibly) low stakes classifications. How? unstructured in -> structure out. 𝐓𝐚𝐛𝐮𝐥𝐚𝐫 𝐟𝐨𝐮𝐧𝐝𝐚𝐭𝐢𝐨𝐧 𝐦𝐨𝐝𝐞𝐥𝐬 𝐥𝐢𝐤𝐞 𝐓𝐚𝐛𝐏𝐅𝐍: grounded predictions (classify labels and regress values) based on ✨ learned data ✨ over tables. How? structure (tabular data, and a lot of it) in -> grounded, explainable predictions out. Definitely not zero-shot There's a world where these two work together for predictive agents

Clara and Lilly on what drew them to Prior Labs and and why they still talk about TabPFN after work (sorry, friends). Want to work in the founders' office, run the projects that grow the company behind the next generation of tabular foundation models? We're hiring across teams in Freiburg, Berlin and New York, including Founders Associates and a new cohort of Visiting Associates for students. Open positions in the comments.

@nihipap it did scale further! priorlabs.ai/technical-repo…

@paul_cal @burkov Yes that’s available now: priorlabs.ai/tabpfn/large-d…

This paper really is groundbreaking. It solves a long-standing embarrassment in machine learning: despite all the hype around deep learning, traditional tree-based methods (XGBoost, CatBoost, random forests, etc) have dominated tabular data—the most common data format in real-world applications—for two decades. Deep learning conquered images, text, and games, but spreadsheets remained stubbornly resistant. This paper's (published in Nature by the way) main contribution is a foundation model that finally beats tree-based methods convincingly on small-to-medium datasets, and does so very fast. TabPFN in 2.8 seconds outperforms CatBoost tuned for 4 hours—a 5,000× speedup. That's not incremental; it's a different regime entirely. The training approach is also fundamentally different. GPT trains on internet text; CLIP trains on image-caption pairs. TabPFN trains on entirely synthetic data—over 100 million artificial datasets generated from causal graphs. TabPFN generates training data by randomly constructing directed acyclic graphs where each edge applies a random transformation (using neural networks, decision trees, discretization, or noise), then pushes random noise through the root nodes and lets it propagate through the graph—the intermediate values at various nodes become features, one becomes the target, and post-processing adds realistic messiness like missing values and outliers. By training on millions of these synthetic datasets with very different structures, the model learns general prediction strategies without ever seeing real data. The inference mechanism is also unusual. Rather than finetuning or prompting, TabPFN performs both "training" and prediction in a single forward pass. You feed it your labeled training data and unlabeled test points together, and it outputs predictions immediately. There's no gradient descent at inference time—the model has learned how to learn from examples during pretraining. The architecture respects tabular structure with two-way attention (across features within a row, then across samples within a column), unlike standard transformers that treat everything as a flat sequence. So, the transformer has basically learned to do supervised learning. Talk to the paper on ChapterPal: chapterpal.com/s/a1899430/acc… Download the PDF: nature.com/articles/s4158…

How do you build a succesful System of Agents? Position at the data source: The most powerful Systems of Agents sit where information is born. They don't wait for data entry into CRMs or ERPs. By capturing data at its origin, they earn the right to orchestrate every downstream action.

Zizheng was one of our interns at NVIDIA back in summer 2023. Later, when we were considering to make him a FT offer, he chose to join DeepSeek without much hesitance. Back then, the DeepSeek multimodal team only has 3 people. I am still very much impressed by Zizheng’s decision at that time. He has been an important contributor of several important works at DeepSeek, including DeepSeek-VL2, DeepSeek-V3, and DeepSeek-R1. I am personally very happy for his decision and the great achievements. Zizheng’s case is a very typical example of what I have witnessed in recent years. Many of our best talents come from China, and these talents don’t have to succeed only in a US company. Instead, we learn a lot from them. The same Sputnik Moment has already happened in AV back in 2022, and it will continue to happen in Robotics and LLM industry as well. I love NVIDIA and want to see her as a continued major contributor to the path of AGI and general autonomy. But if we keep cooking up geo-political agendas and creating hostile opinions to Chinese researchers, we will shoot ourselves in the foot and lose even more competitiveness. We need more talent density, professionalism, learnings, creativity and stronger execution. We don’t need political narratives and clowns like Alexandr Wang.

Zizheng Pan @zizhpan

2 years ago

This moment is absolutely phenomenal to me.

803 2K 33K 4.5M 3K

Michael Scott as Emmanuel Macron: a thread

keyboard_arrow_left Previous keyboard_arrow_right Next

Looking to read up on generative AI over the holidays? Our team at @a16z has been publishing some 🔥 content across verticals. A recap ⬇️

When TechCrunch interviews a French CEO who just sold their startup to a US BigTech

As promised: here is our @sequoia Gen AI market map V2! Very thankful for the hundreds of folks who wrote in additions and edits. AI people are amazing 🥹

“What’s it like living in the UK?”

Using GPT-3 to gain new insight into historical figures

keyboard_arrow_left Previous keyboard_arrow_right Next

Here’s a look at what DALL·E 2 can do. 👀🧵👇 Want to see more? Follow along on Instagram: instagram.com/openaidalle/

As you know, our #ICO is coming very soon, on January 27th precisely! ⏰ In our latest Medium article, we explain everything you need to know to participate 👇 You have more questions? Join our Discord : discord.gg/C9am9GE85q cardashift.medium.com/participate-in…

🔥 #ICO ANNOUNCEMENT 🔥 ⚡️- Flash sale: from January 27th to January 28th 🪙 - Public sale: from January 28th to February 1st The KYC process will last from January 21st to January 26th - Mark your calendars! 📆 #Cardanocommunity #Cardano

“Startup valuations aren’t that high!” Startup Valuations: