The frontier lab building Large Tabular Models for enterprise prediction
San Francisco
Joined January 2025
- Tweets97
- Following3
- Followers322
- Likes28
Fundamental retweeted
proud to see Fundamental named to the #disruptors60
great recognition of the team and the work we're doing
@TechCrunch
Fundamental retweeted
New arena for us! Literally, figuratively … @AEGworldwide is bringing NEXUS to the @LAGalaxy and @LALIVE, where every seat, campaign and sponsorship starts as a prediction.
fundamental.tech/news/fundam…
Fundamental retweeted
your data shouldn’t have to leave your control to benefit from AI
@LTMpredict 🤝 @VAST_Data
Fundamental retweeted
If you’re in SF for Dreamforce this week, you might spot a few Large Tabular Models (LTM) around town 👀🇺🇸
@LTMpredict @Dreamforce
Fundamental retweeted
Recently, I had a great chat with my friend and former colleague @GarneloMarta, CSO at Fundamental (@LTMpredict).
We talked about diffusion models (duh 🤭), music generation, typicality, blogging, @kaggle and more (also restricted Boltzmann machines 🫣)
fundamental.tech/news/unders…
Fundamental retweeted
somewhere right now an exploration and production leadership team is staring at a dashboard, deciding how much capital to commit to a field based on subsurface models that leave half the picture in the dark. they are guessing. and they are good at it. but they are guessing. multiply that one guess across every operator, every basin, every hour, and you've found one of the largest hidden costs in the world economy.
Fundamental retweeted
yep, people think 'AI for tables' means an LLM reading your spreadsheet. It doesn't. LLMs are great at summarizing a table. Predicting the next row is a completely different problem - and most tools quietly skip it
@LTMpredict #NEXUS
Fundamental retweeted
Large language model #chatbots can struggle to deal with tabulated data. Many companies rely on large, connected #spreadsheets for their data, so they need a more specialized program: large tabular models. buff.ly/GKvUns8
Fundamental retweeted
Kind of wild … our model called Argentina v Spain on June 11 on TBPN when betting markets were predicting France to be the winner
@tbpn @jordihays @johncoogan
Happy Thursday. On today's show:
- @arielcoco (Navan)
- @mikewior (Allen Control Systems)
- @jtatarchuk (TensorWave)
- @fraenkelj (Fundamental)
- David Reger (NEURA Robotics)
- Matthew Joseph (Minerva)
- Ade Ajao (Base10)
See you on the stream.
Fundamental retweeted
LLMs are trained on language.
Ask one to predict the next word in a sentence - incredible.
Ask it to find the pattern in 2M rows of enterprise sales data -> it falls apart. That's not what it
was built for.
Rows and columns don't behave like language. The relationships that matter are between
columns, not words.
@LTMpredict #fundamentaltech
Fundamental retweeted
Banks don't compete on who writes the best email.
They compete on who makes the best decisions.
That's why we built NEXUS.
Excited to see MUFG, one of the world's largest financial institutions, deploy it. 🤝
fundamental.tech/news/why-ja… @LTMpredict
Fundamental retweeted
Many advances in AI are really part of the same story: finding ways to make deeper models work
@sedielem discusses this in his latest conversation with @GarneloMarta
Full interview coming soon
Fundamental retweeted
It's crazy that the best predictive model for many billion dollar enterprise decisions is still XGBoost
> a model released in 2016!
Imagine if software engineering still relied on technology from 9 years ago
Fundamental retweeted
One of the biggest questions in AI today is surprisingly simple:
Why did we apply modern AI breakthroughs to text and images before applying them to the data that powers nearly every enterprise?
Our founding advisor @GaelVaroquaux explains how tabular data became the forgotten corner of ML, despite being at the heart of how organizations operate.
That's exactly why we're so excited about the future of LTMs.
Fundamental retweeted
NEXUS predicted that the most likely scoreline in the Brazil vs Morocco game would be 1-1, giving it the highest percentage likelihood (12%).
Our model also predicted that the most likely result in the Haiti vs Scotland game would be 0-1 (13%).
NEXUS was correct in both cases🤯
soccer.fundamental.tech/
Fundamental retweeted
LLMs cracked language. But what about the other half of your brain?
We’ve only solved half of AGI - the creative one that understands language. The data and math side? Still unsolved. BUT according to Jeremy Fraenkel, we are very close to changing that…
cc @fraenkelj