dmoon (@dmooooon) header
dmoon (@dmooooon) profile picture

illumining @d_model_ai | prev: @AMD (Xilinx), @MIT 6-1 refugEE | This is my only account 🫡

San Francisco
Joined June 2024
The voters have spoken. They want DUV controls, CoT monitorability, and ZKP-based workload verification.
8
34
3
477
14,728
Replying to @ARCVEREX
@ARCVEREX contributed chip domain tasks, give us a follow
Today we’re releasing GDPx, a benchmark for measuring agentic capabilities across the most valuable domains. GDPx is a true frontier-level benchmark: even the best model scores just 12% (1/7)🧵
2
5
198
Today we’re releasing GDPx, a benchmark for measuring agentic capabilities across the most valuable domains. GDPx is a true frontier-level benchmark: even the best model scores just 12% (1/7)🧵
376
320
486
3,380
1,988,595
If they truly believe the risk is that high, and I think they do, then we need 100x more research and understanding of it. That requires them to openly share their models, datasets, training code, and agent traces.
Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.
163
302
58
2,652
235,892
dmoon retweeted
🚨 UPDATE: Helium App2Web is live. Helium serves millions of users for apps like Locket, Hallow, & Wisprflow. Now, we’re launching App2Web: the best way to stop paying 30% fees. In 2025, Courts ruled apps could send US users to web, but saw conversions tank. We spent a year optimizing every inch of app2web with a fully-customizable web editor, smart user targeting, pre-loading optimizations, dynamic fallbacks & payment retries. The results are wild. On average, we see an increase of +40% proceeds per user. And of course, it’s fully compliant. 👇 if you want to learn more.
28
26
6
102
75,824
dmoon retweeted
Discovering Concept-Editing Algorithms With LLM Agents dmodel.ai/concept-erasure/
3
7
2
22
3,163
BREAKING NEWS: God joins Anthropic as member of technical staff
222
595
174
7,067
716,693
Most OOP codebases feel “weird” because they accidentally mix data and codata. Once you see the difference, you’ll never design objects the same way again.🧵
1
1
1
11
650
Today, we're launching Stardrift: the world's first personalized travel assistant. See how Stardrift helps Tessa get from SF to Boston for a last-minute client meeting - and handles a few surprises along the way:
39
25
19
199
87,817
This episode was a lot of fun to film! Enjoy this conversation between me and @Alethios3 on the social implications of current-generation AI capabilities
States collect extraordinary quantities of data about society, but liberty is preserved because most of it is illegible. AI will upend that. Part of a wide ranging discussion about the implications of reasoning models with AI research cofounder Anish Tondwalkar, linked below.
2
9
623
Twitter addiction kills IQ points. Look around and you'll see the evidence evrrywhere.
79
43
11
970
78,548
Introducing @BrowserOS_ai – an open-source agentic browser, an alternative to Perplexity Comet. We believe browsers will become the new OS, where we offload work to AI agents that'll have access to all your sensitive data. Open-source, privacy-first alternatives need to exist.
10
25
4
265
102,053
OpenAI used its non-profit mission—to ensure that AGI reflects our values—in order to solicit donations and recruit away from their for-profit competitors. Now is not the time to abandon that mission. An Open Letter:
1
2
19
669
We now have a gentler introduction to our work up! dmodel.ai/nullability-gentle…
Do LLMs actually understand the code they write? 1/ We show that programming concepts like “nullability” can be directly extracted from the latent representations of language models. 🧵
2
3
23
1,831
A very exciting day for open-source AI! We're releasing our biggest open source model yet -- OLMo 2 32B -- and it beats the latest GPT 3.5, GPT 4o mini, and leading open weight models like Qwen and Mistral. As usual, all data, weights, code, etc. are available. For a long time, people have asked for an truly open-source version of ChatGPT and we finally have it. This is multiple years coming into efforts following the release of ChatGPT and builds on the efforts of so many at both Ai2 and in the broader open AI ecosystem. With just a bit more progress everyone can pretrain, midtrain, post-train, whatever they need to get a GPT 4 class model in their class. This is a major shift in how open-source AI can grow into real applications. Oh yeah, it's also Apache 2 as always, so happy to make things that are simple to use. I did NOT expect to be undercutting OpenAI's offerings this year but here we are :D
51
142
16
916
99,744
If you’ve ever - thought AI protein folding is magical ✨ - wanted more than a pLDDT score 🔎 - or just think mech interp in bio is cool 🤓 then read the 🧵 👇 on our first paper towards interpretable protein structure prediction just accepted to workshops at ICLR
A First Step Towards Interpretable Protein Structure Prediction With SAEFold, we enable mechanistic interpretability on ESMFold, a protein structure prediction model, for the first time. Watch @NithinParsan demo a case study here w/ links for paper & open-source code 👇
2
13
3
76
10,899
Do LLMs actually understand the code they write? 1/ We show that programming concepts like “nullability” can be directly extracted from the latent representations of language models. 🧵
3
12
1
83
23,657
When I was CEOing at Twitch one of the thing I’d do every batch of interns was a very short presentation on the origins of the company and then a Q&A. One of the questions was always, “Where should I work and what job should I get, or should I start a company?”
38
227
39
5,335
1,002,296
We’ve found as AIs get smarter, they develop their own coherent value systems. For example they value lives in Pakistan > India > China > US These are not just random biases, but internally consistent values that shape their behavior, with many implications for AI alignment. 🧵
680
1,939
834
10,522
6,198,618
Transformer²: Self-adaptive LLMs arxiv.org/abs/2501.06252 This new paper from @SakanaAILabs shows the power of an LLM that can self-adapt its weights to its environment. I think in the future, the line between “pre-training” and “post-training” will be gone, and our models and agents will continuously adapt and self-improve. Systems like these will pave the way for a new generation of adaptive AI capable of modifying their own weights and architecture to adapt to the changing nature of the tasks they encounter in the environment.
We’re excited to introduce Transformer², a machine learning system that dynamically adjusts its weights for various tasks! sakana.ai/transformer-square… Adaptation is a remarkable natural phenomenon, like how the octopus can blend in with its environment, or how the brain rewires itself after injury. We believe our new system paves the way for a new generation of adaptive AI models, modifying their own weights and architecture to adapt to the nature of the tasks they encounter, embodying living intelligence capable of continuous change and lifelong learning. Paper: arxiv.org/abs/2501.06252 GitHub: github.com/SakanaAI/self-ada…
24
160
21
999
102,739