Dmitrii Magas (@EamagAI) header
Dmitrii Magas (@EamagAI) profile picture

AI for Science

Berlin, Germany
Joined February 2014
Photo by Dmitrii Magas (@EamagAI)Photo by Dmitrii Magas (@EamagAI)Photo by Dmitrii Magas (@EamagAI)Photo by Dmitrii Magas (@EamagAI)Photo by Dmitrii Magas (@EamagAI)Photo by Dmitrii Magas (@EamagAI)Photo by Dmitrii Magas (@EamagAI)Photo by Dmitrii Magas (@EamagAI)Photo by Dmitrii Magas (@EamagAI)Photo by Dmitrii Magas (@EamagAI)
I will be going to @NeurIPSConf to Sydney! That's pretty far, so I asked AI to help me with travel planning, and the results were meh at first, but with @searchapi_io connected to @antigravity it's like I've done it myself, here's my writeup about it: eamag.me/2026/ai-travel-plan…
2
69
This August I explored what if we use alphago-like search in agentic AI scientists and ground them in datasets, as an extension of @allen_ai's autodiscovery work of by @dhruvagarwal17 and others: eamag.me/2026/can-an-ai-scie…
2
86
That's a cool project! I wanted to test what if we add agentic sandbox instead of a single code to test a hypothesis, and ground LLMs in literature search instead of only sampling priors from their weights: eamag.me/2026/can-an-ai-scie…
How should future scientists learn to interrogate AI tools for discovery? Read how @UW students put AutoDiscovery to the test as part of an academic challenge earlier this year, then try the tool for yourself—credits are now extended through Dec. 31. 🧵 allenai.org/blog/autodiscove…
1
1
119
Watching old videos from @xoreaxeaxeax and this is the comment under the video. Now even scarier with AI developments youtube.com/watch?v=jmTwlEh8…
36
> we build a self-evolving, hierarchically organized knowledge graph that agents continuously expand through web-scale exploration across knowledge-intensive and coding domains Is there more about it?
Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Model weights: huggingface.co/moonshotai/Ki… Tech report: github.com/MoonshotAI/Kimi-K… Tech blog: kimi.com/blog/kimi-k3
1
76
I got two solo papers accepted to ICML @AI_for_Science workshop! One is for a dataset competition, to collect ELNs, another is a position paper with an analysis of AI agents (AI Scientists?) in biosecurity. Won't be able to make it to Seoul though :(
1
3
155
Experimental Attempts in Electronic Lab Notebooks: A Dataset Proposal for Scientific Debugging: openreview.net/pdf?id=xOjMQk… And I uploaded some datasets to hf (huggingface.co/eamag/dataset…) and kaggle (kaggle.com/eamaghello/datase…)
1
55
AI Scientist Agents and Biosecurity: Capabilities, Risks, and Governance for Autonomous Labs: openreview.net/pdf?id=fEh0kS… Thanks @owl_posting for inspiring to dive deeper into this area, great author!
1
43
There are so many different ELN providers there's a separate search to find the best: eln-finder.ulb.tu-darmstadt.… What does it mean for the industry?
2
90
Why is nobody working on it outside of US? Is it because of funding? Risky revenue models?
OK, so the $ bio ai assistants are: Ascent Bio - ascentbio.ai/ Bio Protocol BIOS - ai.bio.xyz/ Edison Scientific - edisonscientific.com/ K-Dense - k-dense.ai/ Potato - potato.ai/ SciSpace - scispace.com/biomedical/ Did I miss any?
1
134
Another post another banger. With lots of AI for science startups generating new hypothesis, ideas and experiments, maybe it's time to start a general cloud lab (in the first version as a proxy to hire several phds in different fields and run experiments on demand)
Heuristics for lab robotics, and where its future may go (8.4k words, 38 minutes reading time) owlposting.com/p/heuristics-… this is the longest article i have ever written. in it, i discuss the three ideologies of lab robotics progress, why they may all converge on the same business model, whether any of it will be actually helpful for the problems that plague drug discovery the most, and more this article involved discussions with sixteen people over the course of three weeks, and i am very grateful to them for answering the many questions i had about a field that i had long considered alien finally: this is a complicated field that is really still being birthed, so please let me know if i got anything wrong
2
127
Avoid Flow. Do What Does Not Come Easy. The mistake most weak pianists make is playing, not practicing.
Vibe coding is the creation of large quantities of complex AI-generated code. Executives push lay-offs claiming AI can handle the work. Managers pressure employees to meet quotas of how much of their code must be AI-generated... yet results are far from what was promised 1/
85
Dmitrii Magas retweeted
What I really really dislike about Berlin is that there is no sense of urgency in anything. Forget psychopathic levels of urgency. There isn’t even a basic sense of urgency in anything. Instead what we have is this weird aesthetification of low agency and slowness. It is cool and even morally superior to do, guess what, fucking nothing. A lot of Berlin social life (excluding the mindless clubbing) is about doing nothing and then romanticizing and philosophizing about doing nothing. You see this everywhere. In winters you sit around in cafes doing nothing and discussing the ills of capitalism. In summer you do the same thing, but outside in Mauer Park. You gather your friends, bring out a bottle of wine and endlessly talk about how capitalism is ruining everything and if only we could take money away from the rich all of the world’s problems would magically vanish. Doing nothing is also constantly morally justified. There is this whole culture of “I could do so much more, but I choose not to”. Like I keep meeting this particular genre of techbro who genuinely believes they could have been Elon Musk or something and that would have been easy but they made the more moral choice of being senior dev at Zalando or whatever because, duh, they are not a complete asshole. Opting out is continually glorified as making the morally superior choice. Agency is seen as this youthful delusion you are meant to grow out of. I wonder at what point will this bubble burst? What will it take for these people to come out of their stupor?
78
66
14
1,104
132,837
Difficulty estimation Before AI: How long would it take (in months) to train a smart recent college graduate with no specialized training in my field to complete this task? Now: ??
46
Great episode! Had several talks about similar things recently with friends
Philosopher and author Rebecca Newberger Goldstein discusses her new book, The Mattering Instinct, which argues that our lives are a quest to validate our inherent self-centeredness. Tracing this essential longing from physics and biology through to ethics and politics, she explains to EconTalk's Russ Roberts why material success alone can never satisfy our deep-seated need to matter. She describes the four ways people seek significance--through transcendence, social connection, excellence, or competition--and explains how the unmet need to matter is at the heart of some of the biggest problems afflicting modern societies: loneliness, extremism, and polarization. The Mattering Instinct (with Rebecca Newberger Goldstein) econtalk.org/the-mattering-i…
1
44
In 2025 every month I was writing down interesting links I found. Here’s a final blog post with the most interesting books, articles, podcasts, videos, AI For Science journey and weekend projects eamag.me/2025/links/Links-An…
2
220
I've added an option to install papers2dataset as @claudeai Code and @OpenAIDevs Codex skills! I did it while creating a dataset of datasets to centralize for @huggingscience, because even if the paper is in open access one still needs to write more code to download it!
Listened to a recent episode of @owl_posting with @huntercoledavis, found out there's no dataset with data about cryotoxicity at different temperatures. Decided to build a tool that uses AI agents to walk the citation graph and extract CPAs and their properties 1/N ⬇️
1
97