Fede’s intern 🥊

24.6K posts

Fede’s intern 🥊 banner
Fede’s intern 🥊

Fede’s intern 🥊

@fede_intern

talk is cheap. part of @ergodicgroup, @class_lambda, and @alignedlayer. I work on the core of @ethereum and I build startups.

Buenos Aires, Argentina Katılım Mart 2023
4K Takip Edilen51.1K Takipçiler
Sabitlenmiş Tweet
Fede’s intern 🥊
Fede’s intern 🥊@fede_intern·
LLMs now make critical decisions in hospitals, defense, banks, and governments. Yet nobody can verify which model actually ran, or whether the output was tampered with. A provider or middleman can swap weights, silently requantize the model, alter decoding, inject hidden prompts, do supply chain attacks, or change the deployment surface without the user knowing. This problem is already serious. It will become critical. We think this needs a practical solution, not just a theoretically clean one. CommitLLM is designed to be deployable on existing serving stacks now: the provider keeps the normal GPU serving path, does not need a proving circuit, does not need a kernel rewrite, and does not generate a heavy proof for every response. In practice, two families of approaches dominated the conversation before this work: fingerprinting, which can be gamed, and proof-based systems, which are theoretically strong but too expensive for production inference. We built CommitLLM to target the middle ground. The core idea is to keep the verification discipline of proof systems, but specialize it to open weight LLM inference. The cryptographic core is simple: Freivalds style randomized checks for the large linear layers, plus Merkle commitments for the traced execution. Then a lot of engineering work is needed to make that line up with real GPU inference. The key trick is this. A provider claims `z = W × x` for a massive weight matrix. Normally you would verify that by redoing the multiply. Instead, the verifier samples a secret random vector `r`, precomputes `v = rᵀ × W`, and later checks whether `v · x = rᵀ · z`. Two dot products instead of a full matrix multiply. In the current implementation, a wrong result passes with probability at most `1 / (2^32 - 5)` per check. A full matrix multiply, audited with two dot products. Most of the transformer can then be checked exactly or canonically from committed openings. Nonlinear operations such as activations and layer norms are canonically re executed by the CPU verifier. The one honest caveat is attention: native FP16/BF16 attention is not bit reproducible across hardware. CommitLLM verifies the shell around attention exactly, then independently replays attention and checks that the committed post attention output stays within a measured INT8 corridor. So attention is bounded and audited, not proved exactly. That means the protocol already gives very strong exact guarantees on the parts that matter operationally most. If an audited response used the wrong model, the wrong quantization/configuration, or a tampered input/deployment surface, the audit catches that exactly. That includes things like model swaps, silent requantization, and provider side prompt or system prompt injection. Today the implementation and measurements are strongest on Qwen and Llama. But the protocol itself is not meant to be Qwen or Llama specific: we expect it to generalize across open weight decoder only families. What still has to be done is the engineering work to integrate and validate more families explicitly, and we are already working on that. On the measured path, online generation overhead is about 12 to 14% with the provider staying on the normal GPU serving path. The heavier receipt finalization cost is separate and can be deferred off the user facing path. The main systems costs are RAM and bandwidth, not proof generation. The full response is always committed, but only a random fraction of responses are opened for audit. Individual audits are much larger, roughly 4 MB to 100 MB depending on audit depth. The important number is the amortized one: under a reasonable audit policy, the added bandwidth averages to roughly 300 KB per response. After too many weeks without sleep, I’m proud to show what I built with @diego_aligned: CommitLLM. Thanks Diego for your patience. I've been calling you at random hours. The code and paper still need some cleaning and formalization. We’re already in talks with multiple providers and teams that have cryptography related ideas on how to improve it even more. We’re really excited about this and we will continue doubling down on building products in AI, cryptography and security with my company @class_lambda. If governments, hospitals, defense and financial systems are going to run on LLMs, verifiable inference is not optional. It is infrastructure. I will be explaining this in more details in the days to come and I will show how to test it and run it.
Fede’s intern 🥊 tweet media
English
32
54
349
96.6K
Fede’s intern 🥊
Fede’s intern 🥊@fede_intern·
@DPGSpurs i do think he was wrong, it's the main disagreemeent i had with him but to me it's not a big issue. i do think that now that bitcoin has a problem, i hope it gets fixed and i will do my best to try to help there.
English
0
0
0
31
Dude
Dude@DPGSpurs·
Imagine trying to FUD bitcoin over security budget (which he said would fail already before today and was falsified so now he has to move it forward 100 more years) and Quantum as Ethereum gets bludgeoned to death today on smartphones via exploits. Imagine the same guy that promised eth maxis flippening and told them Bitcoin was dead and to sell all their bitcoin and go all in eth sitting here today as ETH/BTC is closer to literal ZERO than flippening under his watch telling them the same things and getting people like you to give him credibility and say “so what he was wrong!” He literally ushered in the WORST 5 years of ETH performance in history and did so in the most damaging way possible (getting eth maxis to largely hate Bitcoin) and you all sit here and go “meh, he was wrong what’s the issue?) It’s unbelievable to me and in any other industry being this wrong so loudly with such devastating comparative performance would have them run out of town; not being followed on their next new “roadmap and leadership”
English
1
0
0
40
Fede’s intern 🥊
Fede’s intern 🥊@fede_intern·
@DPGSpurs 1. do you really think i hate bitcoin? 2. i do think that bitcoin needs to find a way to fix it's social development layer
English
0
0
0
17
Dude
Dude@DPGSpurs·
So what???? The entire point of this is you sitting there parroting this same thing hes been wrong about for 5 years! No self reflection. No looking back or humility on his thesis or bitcoin hate. No honesty. Just continuing to move the goalposts and invent new meme marketing instead of saying simply “I was wrong”. You truly dont see the issue here?
English
2
0
0
37
Daryl
Daryl@darylspelman·
The RNOA of Berkshire is very good (which tells you good things about the underlying businesses). The ROE is fairly mediocre. The fact that the 10-yr share price CAGR is 12.5% while the RNOA is typically around 30-40% gives an idea of how much a drag this excess cash has created. It's a decent return but could have been much better with the cash invested into something (including their own stock). I'm not suggesting they go gung-ho and get levered up. Just return some of the clearly excess cash they haven't been able to invest. Let's see whether Greg Abel takes a different course now. He's definitely going to come under more pressure than Buffett ever would have to do something with it.
English
1
0
1
107
The Kobeissi Letter
The Kobeissi Letter@KobeissiLetter·
BREAKING: Berkshire Hathaway announces its cash balance is now up to a record $397 billion. The company sold a net -$8.1 billion worth of stocks last quarter, marking its 14th-consecutive net quarterly sale.
The Kobeissi Letter tweet media
English
436
1.1K
7.4K
1.3M
Dude
Dude@DPGSpurs·
Bitcoin Suisse AG@BitcoinSuisseAG

Verified Podcast by Bitcoin Suisse - Your backstage pass into the crypto world 🔥 #3 Justin Drake: The Triple Flippening For the newest episode, we are joined by #Ethereum Researcher @drakefjustin, who predicts that Ethereum will surpass Bitcoin as the largest cryptocurrency… in the coming months! 👀 Join our co-hosts @Dominic_Weibel and @w4vitale as they dive into the world of Ethereum with a long-time expert from the @ethereum Foundation. This episode provides a thought-provoking look at the potential vulnerabilities of #Bitcoin and sheds light on Ethereum's path forward in the evolving crypto landscape. ✅ You do not want to miss this episode! Watch it on YouTube: youtube.com/watch?v=lrFgjO… Listen on Spotify: open.spotify.com/episode/3w7O81… Listen on Apple Podcasts: podcasts.apple.com/us/podcast/3-j… #VerifiedPodcast #BitcoinSuisse

QME
1
0
0
55
Fede’s intern 🥊
Fede’s intern 🥊@fede_intern·
try to spend as much time on the computer as possible
Fede’s intern 🥊 tweet media
English
5
3
30
1.3K
Fede’s intern 🥊
Fede’s intern 🥊@fede_intern·
@DPGSpurs I always saw Bitcoin as the main technology and asset. If Bitcoin doesn't get it's shit together with postquantum I see Ethereum taking over. I also don't understand what will happen with Bitcoin security when rewards go down.
English
3
0
1
72
Dude
Dude@DPGSpurs·
@fede_intern Yet youre out here parrroting nonsense flipping stuff alongside Justin Drake?
English
1
0
0
99
Fede’s intern 🥊 retweetledi
Teknium 🪽
Teknium 🪽@Teknium·
Happy to announce that Hermes Agent's repo just surpassed Anthropic's Claude Code repo
Teknium 🪽 tweet media
English
268
273
4.8K
591.1K
Fede’s intern 🥊 retweetledi
Anthropic
Anthropic@AnthropicAI·
New Anthropic research: Project Deal. We created a marketplace for employees in our San Francisco office, with one big twist. We tasked Claude with buying, selling and negotiating on our colleagues’ behalf.
English
471
737
7.6K
2.9M
Fede’s intern 🥊
Fede’s intern 🥊@fede_intern·
working hard on a sidequest, going offline for a few days. contact my team if you need anything.
English
3
1
27
1.7K
Fede’s intern 🥊
Fede’s intern 🥊@fede_intern·
Peter Thiel is in Argentina. Some of the takes coming from the left and the news about him are retarded.
English
7
2
32
3.5K
Fede’s intern 🥊 retweetledi
Nacho Malter
Nacho Malter@Nalter_Nalter·
@fede_intern @ergodicgroup Gracias Fede por creer en nosotros y seguir haciendo historia en Argentina. 💪❤️🫂
Español
0
4
7
797
Fede’s intern 🥊 retweetledi
Cartoons On The Moon
Cartoons On The Moon@CartoonsOTMoon·
Vistazo a ELECTRO ANDES: el piloto animado indie de Bellolandia studio (Argentina). Siguiendo una historia cyberpunk mitológica, se estrenará este año en el Festival Annecy.
Español
58
1.7K
11.6K
148.7K
Fede’s intern 🥊
Fede’s intern 🥊@fede_intern·
Justin (@drakefjustin) is one of the main reasons I love working in Ethereum and why I believe it has a promising future. During @EthCC, I hosted a fireside with him covering Bitcoin, post-quantum security, @LeanEthereum, @eth_proofs / ZK, Ethereum transaction rails, block building, MEV, and a few other fun topics. Hope you enjoy it as much as I did. Thanks @blockspaceforum for the space to explore so many interesting topics.
Blockspace Forum@blockspaceforum

1/ EthCC was our 2nd forum covering data insights, AI, ePBS, new EIPs, world-renowned macro Researchers, issuance, ZK & builders, privacy, and a fireside covering Ethereum. We hope the community enjoys the videos below. The ticker is ETH. 🧵👇 blockspace.forum/events.html

English
1
13
135
10.4K