
DeepProve proves which AI model answered you. Cited by @AnthropicAI. Halo keeps robots and drones in sync when GPS and radio go down. @LagrangeFndn $LA
Instagram @lagrange.dev
Joined May 2022
- Tweets3.9K
- Following393
- Followers93.3K
- Likes2.8K
Pinned Tweet
The Frontier Red Team at @AnthropicAI just made the case that you should be able to check which AI model answered you, without ever seeing its weights. Their research names DeepProve “the current peer-reviewed state of the art.”
LAGRANGE retweeted
six robots ran a shanghai convenience store for 24 hours. one hit a full trash bin, left the store and didn’t come back. i can relate.
Two claimed proofs of the Generalized Riemann Hypothesis in one day. One was found by @__alpoge__ with @claudeai and comes as Lean code a computer can check. The other is a human draft a mathematician has been checking by hand for months.
Which one would you trust first?
Source: math216.wordpress.com/2026/1…
2
266
LAGRANGE retweeted
if you’re paying for a model through a reseller, you have no idea what model you’re actually getting.
prime intellect just proved it. someone swapped the deepseek weights and the only reason anyone knows is @OpenRouter told them.
Looks like @PrimeIntellect was compromised in a very strange way. No public comms from them yet, but disclosed by OpenRouter to affected folks (such as myself)
In October, our LA newsletter covered Google’s first AI chips in orbit: they could only compute in bursts, because nothing up there carries the heat away.
Today @blueorigin signed on to launch Project Suncatcher for @Google. What gets solved first, the heat or the links between satellites?
Blue Origin and Google have entered into a launch services agreement for Project Suncatcher, Google’s research moonshot exploring how an interconnected network of solar-powered satellites equipped with its AI chips, Tensor Processing Units (TPUs), could scale machine learning compute in space.
bit.ly/4CdBOk1
We send the LA Letter every two weeks: industry news, unique breakthroughs, robots, AI and the infrastructure under both.
Sign up:
566f013b.sibforms.com/serve/…
3
226
Two truths and a hallucination. An AI made one up:
1. An audit of AI resellers found 7 of 28 didn't act like the model on the label.
2. A pet store chatbot told a customer their goldfish had burnout.
3. In NYC, hiring AI needs a yearly bias audit.
Which is fake?
Shockingly, it was 2. An AI wrote that one for this post.
-1 is a real audit from this year: arxiv.org/abs/2605.29524
-3 is New York City’s Local Law 144.
LAGRANGE retweeted
a model wrote a new proof today and the best part is nobody has to believe it. lean checks every step.
Yesterday we wrote that the US banned @deepseek_ai from government software without saying how to tell which software has it.
Today, Washington is weighing a rule that makes US companies disclose when they use Chinese AI models.
How would you check?
@natolambert asked this morning what even counts as a Chinese model: a finetune of @Kimi_Moonshot , a model trained on DeepSeek outputs. If the company can’t answer that, a disclosure on its website won’t either. teapaw.org/natolambert/status/212…
This is stupid. What counts as a chinese model?
What about Thinky's model distilled from Kimi?
What about composer, continued pretraining?
What about Nemotron, using tons of DeepSeek synthetic data?
What about that silly nous finetune for "american" values?
228
LAGRANGE retweeted
ai now builds your portfolio, writes the order, and asks you to approve it. week one you read every single line. by week three are you still reading or just your own compliance department that signs whatever’s in front of you.
The US banned @deepseek_ai from government software in an order signed in July and published in October. The deadline to ask for more time passed a month before anyone could read it. If a vendor had DeepSeek inside its software, how would you know?
It reaches other companies' software too, and nobody has said how to check. lagrange.dev/blog/the-us-ban…
3
410
For anyone not reading the paper before Wednesday: it proves every token of a model’s full answer, and checking the proof takes 1 to 3.7 seconds.
Replying to @chbpap
@chbpap presents our deepprove paper at ACM CCS in the Hague on wednesday, the one @AnthropicAI called the current peer-reviewed state of the art.
If you’re there, go find him. what do you want to ask him?
AI made images, video and text in Europe now have to carry a label saying so. The label gets added by whoever made them. It can't prove which AI made them, or that it's true. Would a label change whether you believe what you're seeing?













