The best onboarding to Scale was reading Alex's blog posts from the early days of the company. In between the memes and crocs, he's a phonemically clear writer.
Thank you to everyone who reached out – we’re overwhelmed by interest in our happy hour tomorrow and unfortunately cannot accommodate everyone.
If you still want to connect, we’d love to grab coffee sometime during the rest of the conference! Feel free to DM me☕️
#icml
Also, will be at ICML🇰🇷 next week!
Hosting a happy hour on Thursday night at one of the top 50 bars in Asia.
Ping me if you’d like to join!
#ICML#ICML2026@icmlconf
For those interested, feel free to RSVP here to our ICML event! 🥃
Spots are limited and we cannot accommodate everybody, but happy to connect with folks during the week as well. See you in Seoul!
Frontier AI models build a working financial model 16% of the time — at best.
The confidence of a target-school MBA intern, but none of the working links.
We built the benchmark for it👇
1/ Introducing BlueFin 🐟
Benchmarking AI Agents on Financial Models
We built a benchmark for the artifact finance actually runs on: the financial model.
131 expert-authored tasks.
3,225 rubric criteria.
5 frontier models evaluated.
The headline result..
Spreadsheets have entered the arena and you can be the #REF!eree → https://t.co/CQAGcrzswh
We built a blind pairwise arena for LLM-generated spreadsheets. 500M+ people use Excel. Time to let the people decide which LLM does it best!
Full paper: https://t.co/0WSdtYThYm
Spreadsheets have entered the arena! ⚔️
Announcing Spreadsheet Arena, the first research platform for human preference rankings on LLM-generated spreadsheets.
The results? @AnthropicAI Claude Opus is on top, but the gap is tighter than you’d think.
w/ @LTIatCMU, @Cornell, and @scale_ai. 🧵
Very excited for us @a16z to have led the seed round for @MeridianAgent alongside @thegp (@phineasb), with participation from @FPVventures and @QEDInvestors.
Meridian is building an IDE designed for finance to bring AI superpowers directly into spreadsheets. The product automates repeatable modeling workflows in a way that conforms to a firm’s templates and standards, with deterministic, reviewable outputs and full traceability back to source data. Instead of producing black-box answers, Meridian is designed for high-stakes environments where every number must be verifiable.
@jhnling and his cofounders were widely regarded as top operators out of @scale_AI. We consistently heard that they set a “high bar within a high bar” and were “unstoppable forces on the team”, and they’ve hired a fantastic team with backgrounds in both finance and AI to pursue the opportunity. They also move fast and are already working with a top investment bank, private equity firms, and high growth startups.
If you live in Excel, Meridian is built for you. Very excited to partner with this team, and congratulations on the launch! @Daisy4ai