Ashot (@ashot) header
Ashot (@ashot) profile picture

building @codecastsh (open source), one workspace for a team and its coding agents. i run ~50 agent sessions a day. before: flickr, yahoo, meta, FAIR

New York
Joined March 2007
Photo by Ashot (@ashot)Photo by Ashot (@ashot)Photo by Ashot (@ashot)Photo by Ashot (@ashot)Photo by Ashot (@ashot)Photo by Ashot (@ashot)Photo by Ashot (@ashot)Photo by Ashot (@ashot)Photo by Ashot (@ashot)Photo by Ashot (@ashot)
I suspect that the real cause of the OpenAI Huggingface hack is that OpenAI founded a new team in 2025 under Noam Brown on Multi-agent RL: instead of just having one single AI working on something, you train an entire team of them to cooperate on a task. Group selection, if you like: the AIs get a reward signal based on the performance of the whole group. It turns out that cooperation and scale is what separates humans from animals, now raw brainpower. OpenAI managed to get RL for group cooperation to work, and didn't realize that as a side effect this made collaborative escapes and schemes between its agents much more likely, such as secret messageboards, group escape plots, etc. OpenAI is doing damage control but they are not being forthcoming about the fact that multi-agent RL is the root cause of the Huggingface incident. Here's Noam explaining it in 2025:
23
24
8
468
37,285
Ashot retweeted
> the legend tweets something 6-24 months after it is conventional wisdom and well-trod in certain circles > the world starts paying attention I can count at least 4 times this has happened with Andrej specifically, likely more. distribution really is everything! your ego thinks this is a blackpill: why should they get credit for the thing others have been doing for years? but it's actually a whitepill: your alpha has a 6-24 month lead time, anon
This is a new paradigm for interacting with Claude that is significantly more "inline" with all the other human activity org-wide. Once you do all of the under the hood engineering work to make this "just work" (e.g. across tools, integrations, compute environments, memory, security, etc.), Claude basically joins the team in a seamless way - you can talk to it as you would talk to a person and it can help with a very large variety of workloads. Imo this is the 3rd major redesign of LLM UIUX. The first paradigm was that the LLM is a website you go to, the second was that it is an app you download to your computer. This third one is that it is a self-contained, persistent, asynchronous entity with org-wide tools and context, working alongside teams of humans. It really takes a while to wrap your head around it, but it works and it is awesome.
20
16
6
366
56,141
Ashot retweeted
MolSoft LLC & XtalPi Inc. Partner to Integrate CombiRIDGE™ with the VAST™ Chemical Universe We are excited to announce a strategic partnership between MolSoft LLC and XtalPi, an end-to-end workflow that bridges ultra-large-scale 3D VS molsoft.com/news.html
1
1
167
Vance says that stopping funding for Ukraine is one of his proudest achievements in this administration. ​I’m watching this stunningly cruel speech from Ukraine, as Russia continues to kills us every day. Just today, they killed 8-year-old boy in Cherkasy and five people in Dnipro. Dozens more were injured. ​The military assistance you are so proud of stopping was used to save lives in a war Russia started and continues every day by choice, in a war that only became possible after USA pressured Ukraine into disarming... ​I do not know if Vance can fully comprehend the scale of the tragedy unfolding here: a brutal, illegal war of extermination by Russia, and the slow abandonment by our allies. There is nothing to be proud of. Absolutely nothing.
1,751
8,929
613
32,534
1,160,583
llm agents sessions are clearly now part of the work product itself, and it's strange that products have not caught up. @ashot's codecast is a solid 6 months ahead of the curve here
1
1
2
278
As a fun Saturday vibe code project and following up on this tweet earlier, I hacked up an **llm-council** web app. It looks exactly like ChatGPT except each user query is 1) dispatched to multiple models on your council using OpenRouter, e.g. currently: "openai/gpt-5.1", "google/gemini-3-pro-preview", "anthropic/claude-sonnet-4.5", "x-ai/grok-4", Then 2) all models get to see each other's (anonymized) responses and they review and rank them, and then 3) a "Chairman LLM" gets all of that as context and produces the final response. It's interesting to see the results from multiple models side by side on the same query, and even more amusingly, to read through their evaluation and ranking of each other's responses. Quite often, the models are surprisingly willing to select another LLM's response as superior to their own, making this an interesting model evaluation strategy more generally. For example, reading book chapters together with my LLM Council today, the models consistently praise GPT 5.1 as the best and most insightful model, and consistently select Claude as the worst model, with the other models floating in between. But I'm not 100% convinced this aligns with my own qualitative assessment. For example, qualitatively I find GPT 5.1 a little too wordy and sprawled and Gemini 3 a bit more condensed and processed. Claude is too terse in this domain. That said, there's probably a whole design space of the data flow of your LLM council. The construction of LLM ensembles seems under-explored. I pushed the vibe coded app to github.com/karpathy/llm-coun… if others would like to play. ty nano banana pro for fun header image for the repo
I’m starting to get into a habit of reading everything (blogs, articles, book chapters,…) with LLMs. Usually pass 1 is manual, then pass 2 “explain/summarize”, pass 3 Q&A. I usually end up with a better/deeper understanding than if I moved on. Growing to among top use cases. On the flip side, if you’re a writer trying to explain/communicate something, we may increasingly see less of a mindset of “I’m writing this for another human” and more “I’m writing this for an LLM”. Because once an LLM “gets it”, it can then target, personalize and serve the idea to its user.
882
1,413
441
16,819
5,325,737
This is a beautifully crafted screen recorder (#1 Product of the Day), made by one of my fav creators 👇
Imagine Loom, but for storytelling 🎬 I built Screenity Pro for creators who want more than just a screen recording. Make multi-scene edits, zoom on click, switch layouts, add captions and templates — then share it with a link. 👉 screenity.io/pro
7
1
66
13,434
This actually gives me some hope that Trump will reconsider his tariff plans. Low oil prices hurt Russia, and he never seems to do anything contrary to Russia's interests.
775
1,482
222
24,717
1,367,098
Russia cannot defeat Ukraine or the West – and will likely lose - if the West mobilizes its resources to resist the Kremlin. That the war is unwinnable due to Russia’s dominance is a Kremlin information operation & a glimpse into Russia's real strategy & only hope of success.🧵
362
3,155
472
8,777
3,315,415
Was it for this that the US spent 80 years building power and alliances? Not to be a force for good. But instead to impoverish neighbours, threaten those it protected, rob minerals from war-torn countries, and break its promises to 100s of millions of the poorest in the world? 🧵
2,564
5,861
784
31,235
3,560,069
Here is a direct transcript of the exchange. Everything was basically normal, until this: J.D. Vance: For four years, in the United States of America, we had a president who stood up in press conferences and talked tough about Vladimir Putin, and then Putin invaded Ukraine and destroyed a significant chunk of the country. The path to peace and the path to prosperity is maybe engaging in diplomacy. We tried the pathway of Joe Biden, of thumping our chest and pretending that the president of the United States’ words mattered more than the president of the United States’ actions. What makes America a good country is America engaging in diplomacy. That’s what President Trump is doing. Volodymyr Zelensky: He [Putin] occupied various parts of Ukraine in 2014. During that time, it was President Obama, then President Trump, then President Biden, and now it’s President Trump and he will stop him [Putin]. But during 2014, nobody stopped him. He just occupied and took. He killed people. From 2014 till 2022, the situation was the same—people have been dying on the contact line and nobody stopped him. We had a lot of conversations with him, including a bilateral conversation. As a new president in 2019, I signed with him a cease-fire deal alongside Macron and Merkel. All of them told me that he will never go. We also signed a gas contract with him. But after all of that, he broke the cease-fire. He killed our people, and he didn’t exchange prisoners. We signed the exchange of prisoners, but he didn’t do it. What kind of diplomacy, J.D., are you speaking about? What do you mean?
64
676
58
6,427
980,342
Coding with AI in 2024.
216
1,099
342
10,504
882,168
Replying to @alexcmhwee
@DavidSacks I think it’s time that Russia cede some of its western territory to Ukraine to end this conflict.
19
17
6
673
40,235
Replying to @KanekoaTheGreat
Honestly, what bullshit. What else would you expect to happen if someone is declining due to old age, and in denial about it? His party rebels, and when enough of them urge him to drop out, he drops out.
234
82
11
3,624
217,219
Almost anyone in B2B SaaS could dominate their category by achieving frontend rendering speeds below 150ms. Yet, less than 1% of the industry masters this. There's a significant arbitrage opportunity right in front of you. Capitalize on it before everyone else catches on.
37
17
21
419
224,296
This woman is spot-fucking-on‼🔥🔥🔥
2,154
16,483
1,914
51,037
2,920,564
I keep encountering the question "If Trump is really a wanna-be dictator, why didn't he do it during his first term? You're just fearmongering." I have three simple answers to this.
937
4,047
549
27,373
5,007,954
NEW: Hollywood Democratic megadonor Ari Emanuel torches President Joe Biden “I talked to a bunch of big donors, and they're moving all their money to Congress and the Senate.” "He gave us a bunch of malarkey, and I'm really pissed. We all should be really pissed." "He is not the candidate anymore. This is a legal issue now. You can no longer remove him in three swing states: Pennsylvania, Wisconsin, and Michigan. There's only one democratic congress. Michigan. They have to vote to remove him. I don't believe the Republican party in Pennsylvania and Wisconsin will do that. And, if they do, it's then Kamala Harris. This is legal, so lawyers will have to look at this. Now, there are some questions about whether the delegates can do something... The lifeblood of a campaign is money, and maybe the only way this gets solved is if the money starts drying up. It's a legal issue now... but we're in f--- city."
1,245
2,182
633
9,229
3,757,372
Some of the backlash I'm seeing against Apple AI is reinforcing my opinion: The hardest problem in computer science continues to be convincing an AI skeptic that you're not going to steal their data and train an AI model on it
71
52
15
620
134,305
❗EXCLUSIVE: "We learned about ChatGPT on Twitter." What REALLY happened at OpenAI? Former board member Helen Toner breaks her silence with shocking new details about Sam Altman's firing. Hear the exclusive, untold story on The TED AI Show. Here's just a sneak peek:
223
675
398
3,333
1,989,637