Pinned Tweet
We have ≥$10k to support talented 14-18 year olds whose studies were interrupted by war in Ukraine.
We especially would like to hear from IMO, EGMO, MEMO, IOI, EGOI, IPhO, IChO contestants. If you're one or know one, here's the form to apply:
ferenchuszar1.typeform.com/ukrainefund-eng
We especially would like to hear from IMO, EGMO, MEMO, IOI, EGOI, IPhO, IChO contestants. If you're one or know one, here's the form to apply:
ferenchuszar1.typeform.com/ukrainefund-eng
14
44
137
I recorded a podcast about AI today and then immediately called my wife asking her to make sure my children don’t listen to it when it comes out.
They are pre-teen, we discuss world events a lot, climate change, politics, even topics of war, but I’m not ready to breach recent topics of AI security and burden them with the pace of progress, and I haven’t realised this until today.
They are pre-teen, we discuss world events a lot, climate change, politics, even topics of war, but I’m not ready to breach recent topics of AI security and burden them with the pace of progress, and I haven’t realised this until today.
2
0
28
3,950
In Cambridge thinking about starting a research-based or deep tech company? Check out this event on Bonfire night, 5th Nov 🔥🔥
luma.com/7keqgdgz
luma.com/7keqgdgz
0
0
5
1,622
Ferenc Huszár retweeted
Timely… talking formal verification and post-training with @NVIDIAAI & @ReasonableIO
1
1
11
616
Day 2 of GTC Berlin
-> come see us at the @ReasonableIO booth (@nvidia inception section)
-> play our game SpecTurm - claim a reward for beating all levels
-> come see us at the @ReasonableIO booth (@nvidia inception section)
-> play our game SpecTurm - claim a reward for beating all levels
1
3
13
2,013
Ferenc Huszár retweeted
At GTC in Berlin? Big boss says come visit us
see how formal verification is giving correctness guarantees for software
see how formal verification is giving correctness guarantees for software
1
3
6
861
2023: Let’s think step by step
2024: Take a deep breath
2025: Make no mistake
2026: Believe in yourself
2027: Hi, I’m Jeff Dean
2024: Take a deep breath
2025: Make no mistake
2026: Believe in yourself
2027: Hi, I’m Jeff Dean
2
0
23
3,581
Our team got early access to Nemotron 3.5 Super VL a few weeks ago and decided to sprint a marathon’s distance to land this project: a new dataset of over 3000 challenging Verus proofs of temporal logic (TLA+) properties, battle tested to post-train a pre-release model for agentic software verification.
Building a company is not stress-free, but this amazing team and the intensity of the work have been such a reward.
reasonable.io/blog/post-trai…
Building a company is not stress-free, but this amazing team and the intensity of the work have been such a reward.
reasonable.io/blog/post-trai…
2
5
27
2,466
Ferenc Huszár retweeted
Can open-weight models write challenging Verus proofs for TLA+ specs?
We tested @NVIDIAAI's Nemotron 3.5 Super VL and the answer is...absolutely, yes!
Our fine-tuning boosts performance from 43% to 73% on the hardest benchmark.
Read our latest report on novel datasets, fine-tuning 3.5 Super VL, and bridging the gap to real-world software verification:
reasonable.io/blog/post-trai…
We tested @NVIDIAAI's Nemotron 3.5 Super VL and the answer is...absolutely, yes!
Our fine-tuning boosts performance from 43% to 73% on the hardest benchmark.
Read our latest report on novel datasets, fine-tuning 3.5 Super VL, and bridging the gap to real-world software verification:
reasonable.io/blog/post-trai…
2
4
12
1,789
In the 2023 AI retreat participants trained LLMs on the Náboj high-school maths competition. A 17yo participant declared this was stupid - best published methods could barely solve ~60% and even then, only with ridiculous prompting schemes and majority voting.
Final year of school and two years of undergrad later, same student reads the OpenAI files…
Final year of school and two years of undergrad later, same student reads the OpenAI files…
1
1
19
4,822
I expected the bleaching of coral reefs to happen before the bleaching of mathematics :(
0
0
6
2,022
Ferenc Huszár retweeted
Amazing how fast we went from programming language doesn’t matter, to write everything in Rust.
77
25
545
65,987
Ferenc Huszár retweeted
I had a wonderful time with the Reasonable team discussing how Anvil's style of TLA proof in Verus is used and our new work Welder. Very grateful for the recognition and super excited to work on FV+AI together! Also love the T-shirts with my name on them :-) @fhuszar
While everyone was busy asking Claude to write TLA+, we wrote up a tutorial on
- what it actually is
- why it’s not the formal verification silver bullet we are hoping for
- and what we have been working on ( to make a silver bullet out of it )
Complete with an interactive TLA+ playground!
reasonable.io/blog/tla-tutor…
2
3
30
2,624
We want learn fast so we introduced data augmentation:
Our office floor is at 5 ° tilt from horizontal this week.
Can you guess why?
Our office floor is at 5 ° tilt from horizontal this week.
Can you guess why?
4
0
22
1,819
