DeepProve: a cryptographic receipt for every AI inference. now open source. Halo: autonomous coordination engine for the toughest conditions. @LagrangeFndn $LA

Los Angeles
You can't stop this swarm. That's the point. Jam its comms. Spoof its GPS. Shoot three out of the sky. It re-plans and finishes anyway. Your score is how many seconds you cost it. holdtheswarm.lagrange.dev/?u…
9
6
40
30,201
LAGRANGE retweeted
"When Lagrange helps generate zero-knowledge proofs, you can and should trust Lagrange in what we say, but you don't even need to. All you need to do is trust in mathematics which is 100% definitive and certain." Click the link below to hear it. piped.video/w3FSqK3mRXg?t=1263
4
11
1,672
Give a drone a fake satellite signal and it flies the whole mission in the wrong direction. Confidently, and nothing on board mentions it. Episode 6: why one drone believes it and fifty do not. Under a minute, zero jargon.
1
3
15
2,589
Episode 5, if you missed it: a swarm goes out with fifty drones and comes home with twenty-five. What the other twenty-five did in between is the whole story.
Half the swarm goes down. Not all at once, which would be easier, but one at a time while the job is still running. The question isn't whether the rest keep flying. It's whether they still agree on what they're doing.
1
6
406
Picking an AI vendor is still mostly a sales deck and a gut call. AI Collaborator vets the vendor. We prove what the model did once it is running. Welcome to the ecosystem, AI Collaborator.
4
17
2,635
29 countries have now tried to bring down the swarm. 2,158 attacks later the pattern is the same everywhere. 73% jamming. 8% shooting. Players work out on their own that jamming beats shooting, and most of them finish with half their shots unused. holdtheswarm.lagrange.dev
3
4
14
2,985
DeepProve is for teams running AI where someone downstream has to check what actually happened, and taking your word for it is not an option. It's the proof layer the rest of the stack runs on. Which model answered. On which input. What it returned. Checkable by a stranger, without re-running the work and without access to your weights. It's open source and it's on the @awscloud Marketplace among others, so you keep the models and the infrastructure you already have.
2
3
16
2,535
We prove something happened. @lumera proves it is still there. A proof of an inference is made once, at the moment it runs. A proof of storage has to be made again every epoch, forever, or the operator gets slashed. Two different jobs. Same reason for doing them.
1/ Lumera x @lagrangedev. A partnership built on a shared belief: you cannot trust what you cannot verify.
1
5
20
3,578
Half the swarm goes down. Not all at once, which would be easier, but one at a time while the job is still running. The question isn't whether the rest keep flying. It's whether they still agree on what they're doing.
2
1
10
3,815
Episode 4, if you missed it: one leader and forty nine followers means one shot can take down all fifty.
Take fifty drones. In most swarms, that's one leader and forty-nine followers. Which means one shot can take down all fifty. Episode 4: the difference between a real swarm and a light show, in under a minute. Zero jargon.
2
769
Lagrange x @lumera DeepProve proves an AI output came from the right model on the right inputs, the moment it runs. The people who check that proof, like auditors or customers, often show up years later. Glad to be building with @lumera so those proofs stay available and verifiable.
Lumera x @lagrangedev Lagrange's zkML gives AI inference a cryptographic proof of correctness, down to the exact model. Cascade adds permanent, provable storage for that proof, so it stays verifiable for as long as it matters. Excited to build with the Lagrange team. Read the full announcement here 👇 lumera.io/lumera-x-lagrange-…
1
4
14
3,558
LAGRANGE retweeted
드론이 많아질수록 중요한 건 AI보다 ‘협업’일지도 모릅니다 자율주행 드론 한 대를 똑똑하게 만드는 것과 드론 수십 대, 수백 대를 동시에 움직이게 만드는 것은 조금 다른 문제입니다. 한 대라면 자신의 위치를 파악하고 주어진 임무를 수행하면 됩니다. 하지만 여러 대가 하나의 팀으로 움직이기 시작하면 새로운 문제가 생깁니다. 한 대의 통신이 끊기면 어떻게 할까요? 일부 기체가 고장 나면 그 기체가 맡고 있던 임무는 누가 이어받아야 할까요? 네트워크 일부가 끊어졌는데도 나머지 기체들은 서로 같은 상황을 보고 있을까요? 라그랑주(Lagrange)는 이런 문제를 "대규모 자율성은 결국 조정의 문제"라고 보고 있습니다. 그리고 이를 위해 만든 것이 헤일로(Halo)입니다. 헤일로(Halo)는 드론을 직접 조종하는 미션 소프트웨어나 지휘통제 시스템이 아닙니다. 여러 자율 시스템이 서로 상태를 공유하고, 네트워크가 불안정해지거나 일부 시스템이 사라지더라도 남은 시스템들이 역할과 위치를 다시 조정할 수 있도록 만드는 자율 시스템 협업 조정 엔진 (Coordination Engine for Autonomous Systems) 입니다. 예를 들어 드론 여러 대가 하나의 지역을 탐색하고 있다고 가정해보겠습니다. 중간에 한 대가 사라지면 나머지 기체들이 담당 구역을 다시 나눕니다. 통신 경로가 끊기면 다른 노드를 통해 연결을 다시 구성합니다. 일부 기체가 임무를 수행할 수 없게 되면 그 역할을 다른 기체에 재할당합니다. 핵심은 모든 기체가 절대 고장 나지 않게 만드는 것이 아니라, 고장이 나더라도 전체 시스템은 계속 움직이게 만드는 것입니다. 라그랑주(Lagrange)가 헤일로(Halo)에서 강조하는 기능도 이 방향으로 연결됩니다. 협업 센싱, 통신 중계, 대형과 지역 커버리지 유지, 그리고 자산 손실 시 역할 재분배입니다. 현재 헤일로(Halo)는 드론을 우선으로 시작하지만 라그랑주(Lagrange)는 향후 지상·해상 시스템과 유·무인 협업까지 확장하는 방향을 제시하고 있습니다. 결국 헤일로(Halo)가 지향하는 미래는, 한 대의 자율 시스템이 모든 것을 해결하는 것이 아니라 수많은 자율 시스템이 서로 협력해 하나의 시스템처럼 움직이는 것입니다. AI가 각 기체의 ‘두뇌’라면, 헤일로(Halo)가 만들고 있는 것은 그 두뇌들이 서로 협력할 수 있게 하는 조정 계층(Coordination Layer)이라고 볼 수 있을 것 같습니다. 라그랑주(Lagrange)가 말하는 “Autonomy at scale is a coordination problem.” (“자율 시스템이 대규모로 확장될수록 핵심 문제는 ‘조정’입니다.”) 이라는 문장이 헤일로(Halo)를 가장 잘 설명하는 것 같습니다.
Two questions are getting harder to answer every month. Did the AI actually run what they say it ran? Will the machines still work together when the network breaks? Neither one can be answered by looking. DeepProve puts a cryptographic receipt on every AI inference. Halo keeps autonomous systems agreeing on what is true while operating in the most contested environments. Ground truth for AI and autonomy.
26
5
38
1,343
AI agents are about to run half your project board. How do you know the agent actually did the work? Welcome to the ecosystem, monday.com.
3
3
17
3,117
The Army jams GPS on purpose around the White Sands Missile Range in New Mexico. Inside the fence it does the more dangerous thing: feeds receivers a signal that makes them report the wrong position, confidently. The jamming is a wall around its own lie. A jammed receiver knows it is lost. A spoofed one is certain it is fine. lagrange.dev/blog/the-army-j…
2
3
12
3,091
Take fifty drones. In most swarms, that's one leader and forty-nine followers. Which means one shot can take down all fifty. Episode 4: the difference between a real swarm and a light show, in under a minute. Zero jargon.
2
2
13
4,630
Episode 3, if you missed it: fifty drones aren't fifty aircraft, they're one very fast conversation.
One drone is a solved problem. Fifty drones working together is one of the hardest problems in engineering. Episode 3: what a swarm actually is, in under a minute. Zero jargon. Save the last line for the next drone swarm headline.
6
1,007
If you're at NDIA or Billington in DC this week, here are six questions that will tell you more about any autonomy vendor than the capability deck. 1. What does your system do in the first ten seconds after the link drops? 2. When one aircraft goes down, who reassigns its work, and how long does it take? 3. What happens when the group splits in two? And when it comes back together? 4. How do you know every node is acting on the same picture right now? 5. How does your failure detector tell "destroyed" from "briefly quiet" on a congested link? 6. What stops something joining the group that shouldn't be there? Ask them in order. Fifteen minutes of answers beats any demo reel. And ask us the same six: defense@lagrange.dev
2
2
12
3,473