Rami Khalil (@ramikhalil) header
Rami Khalil (@ramikhalil) profile picture

yet another person prev @BoundlessHQ @RISCZero | Computing PhD @imperialcollege | Informatik MSc @ETH

Joined July 2010
Photo by Rami Khalil (@ramikhalil)Photo by Rami Khalil (@ramikhalil)Photo by Rami Khalil (@ramikhalil)Photo by Rami Khalil (@ramikhalil)Photo by Rami Khalil (@ramikhalil)Photo by Rami Khalil (@ramikhalil)Photo by Rami Khalil (@ramikhalil)Photo by Rami Khalil (@ramikhalil)Photo by Rami Khalil (@ramikhalil)Photo by Rami Khalil (@ramikhalil)
Rami Khalil retweeted
I’m looking for a new team to join! Past 7 years: distributed systems, developer tooling, and cryptography. Led teams at ChainSafe and NEAR, most recently at RISC Zero/Boundless. DMs are open to any team working on difficult problems, especially in new domains for me!
4
13
3
116
6,998
Reality of our days: 1. I am sitting on a set of vulns (4+) that allows me to remotely take over a consumer home automation device I have (3+ diff ways). 2. The vendor doesn't only serve consumers, but also critical infra (announced partnerships include pharma manufacturing plants, airport logistics, hospitals, etc) across various locations in Europe and the US (not the same product line, ofc, so maybe moot). 3. Claude and Astra refuse to help me explore even my own device, so now I'm using Chinese hosted models for investigating. 4. The vendor (with a revenue of $0.75B) does not maintain any security disclosure practice whatsoever, nor does respond to emails. So, to recap, I need to outsource vulnerability hunting of (potentially, I don't know) critical western infrastructure to Chinese models, because the western models refuse; and the vendor doesn't give a shit. Fun timeline. PS: The vendor is Swiss.
126
290
52
3,824
176,706
Rami Khalil retweeted
I was recently affected by the layoffs at @0xMiden, and I’m looking for my next Product Manager role in the Bay Area. I’d love to keep building in payments, especially stablecoins and agentic payments. Over the past few years at @0xMiden and @0xPolygon, I’ve: • Led a crypto wallet from scoping through launch to reach 20k user registrations, coordinating a seven-person team across engineering, design, and QA. • Worked with roughly 12 external ecosystem partners, translating their requirements, feedback, and integration blockers into roadmap and release priorities. • Built and launched AggLayer’s developer portal and sandbox, with tutorials and code examples to help developers get started. • Researched zero-knowledge infrastructure and develop the Chainless Apps framework, connecting technical research with practical use cases for developers. Working on a crypto wallet made the practical challenges of stablecoin adoption much clearer to me: how people bring funds in, move them across networks, and actually use them. I want to keep working on those problems, including how AI agents will make payments. I’ve also been building an offline, on-device LLM app and AI-assisted workflows for team syncs, product feedback, and moving GitHub issues through implementation. I’m comfortable getting into the code as well as owning the product. If you know PM hiring managers at @Stablecoin, @stripe, @Visa, or other teams building payments infrastructure and AI products, I’d really appreciate an introduction! And to my teammates at Miden and Polygon Labs: thank you. I’ve learned a lot from building with you, and wish you all the best!
46
14
5
213
10,941
Rami Khalil retweeted
Anthropic is releasing an open source library, FlashProve, for provable inference. It was measured on Llama 4 Maverick, a 400B-parameter mixture of experts model. Code: github.com/anthropics/FlashP… Blog post: anthropic.com/research/prova…
51
90
43
745
109,386
Rami Khalil retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
2,367
7,260
5,752
55,939
10,052,356
Ok, but how do older versions of Claude/ChatGPT perform? Can't really tell "how many months behind" without that artificialanalysis.ai/?model…
Announcing Artificial Analysis Intelligence Index v5, featuring 2 new benchmarks and 3 upgrades. For the first time, the majority of the Index is built on private datasets The Artificial Analysis Intelligence Index is our composite measure of model capabilities across tasks reflecting real-world usage today. v5 is our biggest upgrade since v4 launched in January. With v5, 60% of the Index is now based on evaluations with private test sets or held-out answers. Alongside Intelligence Index v5, we’re launching the Artificial Analysis Progress Index v1, which tracks model capability over time on one continuous scale. New to the Intelligence Index: ➤ FrontierCode 2.0: Cognition's newly-launched private benchmark of real-world coding tasks, graded on whether a repository maintainer would merge the agent's change ➤ Terminal-Bench-Science: 70 agentic scientific research tasks, run in a terminal environment Upgraded: ➤ HLE-Diamond (replacing Humanity's Last Exam): a refined 1,000-question version of Humanity's Last Exam, split evenly between reasoning and knowledge questions ➤ AA-Omniscience v1.1: An updated dataset with saturated and public set questions removed, new questions added, and ambiguous or erroneous questions revised ➤ CritPt v2: a private, 40-question hard subset of CritPt v2, revised by the CritPt authors and run with a code-execution tool Staying in the Index: AA-Briefcase, AutomationBench-AA, GDP.pdf and Terminal-Bench 4.0 Removed from the Index: GDPval-AA v2.1, SciCode, and AA-LCR v1.1 Key results: ➤ Claude Opus 5.5 (max) leads at 58, ahead of GPT-6 Astra (max, 57) and GPT-6.1 Sol (max, 55): Anthropic and OpenAI hold the top five models. Gemini 4 Argon (high) is next at 50 ➤ The open weights gap widens from 3 points to 12: Kimi K3.1 Pro (weights not yet released) is the top open weights model at 46. Among released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) lead at 32 - 26 points behind the leader and below Claude Haiku 5.5 (max, 36) ➤ Cost per task varies by more than 13x among the leading models: of the 10 highest-scoring models shown, GPT-6.1 Terra (max) is the cheapest at $0.96 per Intelligence Index task and Claude Fable 5.1 (max) the most expensive at $13.04
1
148
Rami Khalil retweeted
“A single cluster currently serves 155 trillion tokens/day, generating $38 in revenue per GPU-hour (at off-peak pricing).” Huge flex from DeepSeek. Not only did they open-source their inference engine, but they also open-sourced their revenue numbers. They didn't disclose which GPUs they're using here, but DeepInfer supports both NVIDIA Blackwell and Huawei Ascend. Even assuming they're renting B300s at $7 per GPU-hour, that's ~82% gross margin.
126
295
78
3,920
552,421
Rami Khalil retweeted
Next year, I will assign one of my jira tickets to every kid in America. This will be the first of many.
102
454
36
11,079
217,946
Rami Khalil retweeted
Anthropic actually loses far less than I imagined off the $200 subscription. It’s likely that the subsidized plans are actually very sustainable and may never go away. It's widely said that inference is quite profitable. This turns out to be true, and can be computed from first principles by estimating out cost of GPUs, electricity, datacenters, etc. On paper it might cost something like $0.10–0.30 per million input tokens and $0.50–2.50 per million output to serve Opus 5.5, and they sell it for $4 and $20. The problem with coming to the defense of AI economics by saying inference is profitable is that it makes no sense to decouple inference from training. Afaict, no company has yet recouped training costs for a model from profits generated from serving that model. A model today is served for a month or two, maybe less, before the next one replaces it. If only the AI companies could serve these models for a little longer before having to race to train the next one. They would get filthy rich off inference. If only they could all agree to slow down…
So many devs seem to think that the $100 and $200 subscription plans are how Anthropic and OpenAI make their revenue. There are companies spending over $500 MILLION a year with Anthropic. Your subs aren't even a drop in the bucket.
116
32
14
1,156
122,211
Rami Khalil retweeted
using AI to solve millennium problems
We’re releasing a broad range of new mathematical results produced by an internal frontier model. We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results. github.com/openai/math
18
289
13
5,483
143,997
Rami Khalil retweeted
"openai just proved that Riemann zeta is zero-free in the half-plane ℜ(s) > 7/8" "incredible this changes everything" "do you know what ℜ(s) is?" "no, do you?" "no."
33
483
26
9,170
191,562
Rami Khalil retweeted
“I read every line of my AI code”
19
115
9
1,921
39,443
Rami Khalil retweeted
being on 𝕏 all day gives you the most broken view of reality lol. spend 10 minutes on here and you'd think: > AGI is so close > RSI has been achieved > robots are going to kill us all > TPUs are going to space > Opus 5.5 is the best model ever > Astra is insane > Jev is replacing half your LLM calls > some new model is about to end civilization then you go outside... and people are like: "wait, ChatGPT can code?" the difference between AI Twitter and the actual world is hilarious.
401
443
89
6,580
227,224
Rami Khalil retweeted
Dario has written that we need to “pace the frontier,” and Sam has agreed. People may be surprised by my response: go ahead. You guys are the frontier. By any reasonable metric — market share, revenue growth, model capability — the two of you have a duopoly on frontier intelligence. You’ve also claimed the lead is widening because of recursive self-improvement. I don’t see what you see in the lab. If the unreleased models are scary enough that you think you should slow down, I support your decision to be responsible. But stop pretending you need anyone else’s permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability. Stop pretending METR is independent when it is intertwined with Anthropic’s investors and staff. Stop pretending you need those same evaluators to police competitors who aren’t even at the frontier. Most of all, stop pretending the motivation to slow down is purely altruistic. You face massive product-liability exposure if your products enable a truly damaging cyberattack. The market already punishes models that behave in unpredictable or unauthorized ways. After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability. Call it alignment if you want. It is also just giving customers what they want. Pacing the frontier would also create breathing room for a more intelligent conversation about regulation than Bernie Sanders’ “shut it all down.” China is very unlikely to join a global agreement, as you know, and that has to be taken into account as well. So go ahead and pace the frontier. You are the ones setting it. The easiest way not to build superintelligence is for you to agree not to build it. Demanding your preferred regulatory framework as the price of that will look like blackmail of the public and the political system. So just do it. If you do, you’ll buy goodwill for the next conversation. If you don’t, we’ll know this was just another bid for regulatory capture — or an election-season psyop.
3,376
12,227
2,709
71,709
9,649,068
“we sandboxed the agent” meanwhile the agent:
258
3,406
248
35,447
1,781,410
Rami Khalil retweeted
drop the “xyz” — it’s cleaner.
19
4
5
65
14,088
Rami Khalil retweeted
The current state of AI would be nowhere as good as it is if it wasn't for deepseek openly contributing their research. They invented GRPO, they invented the concept of test time compute, they invented agentic LLMs, as well as countless infra inventions. All open sourced.
146
716
50
10,816
403,114
Rami Khalil retweeted
Replying to @mkratsios47
we have information that Anthropic distilled humanity's entire internet for the development of Fable 5
43
213
7
11,052
200,721
Rami Khalil retweeted
Replying to @mkratsios47
I don't remember letting anthropic scrape my GitHub profile as well as all of my online writing!
41
155
2
8,646
215,860