Markus J. Buehler (@ProfBuehlerMIT) / X

Markus J. Buehler

6,550 posts

Markus J. Buehler

@ProfBuehlerMIT

McAfee Professor of Engineering @MIT; Co-Founder & CTO at Unreasonable Labs; AI-Driven Scientific Discovery

Cambridge, MA

meche.mit.edu/people/faculty…

Joined December 2014

Pinned
Markus J. Buehler
@ProfBuehlerMIT
Jun 5
We've made a breakthrough in self-evolving AI scientists moving from "search" to "principled discovery": Scientific discovery requires that the search space itself changes, and an AI scientist must perceive this shift without intervention. We built an AI that achieves this for
00:00
778K
Markus J. Buehler
@ProfBuehlerMIT
Feb 23, 2025
We trained a graph-native AI, then let it reason for days, forming a dynamic relational world model on its own - no pre-programming. Emergent hubs, small-world properties, modularity, & scale-free structures arose naturally. The model then exploited compositional reasoning &
00:00
359K
Markus J. Buehler
@ProfBuehlerMIT
Mar 25, 2025
Deep stuff! We uncovered a startling link between #entropy—a bedrock concept in #physics—and how #AI can keep discovering new ideas without stagnating. In an era where reasoning models can reflect on problems for days at a time (rather than generating quick, single-step
211K
Markus J. Buehler
@ProfBuehlerMIT
Sep 23, 2024
We are excited to share #PDF2Audio, an open-source alternative to the #podcast feature of #NotebookLM with flexibility & tailored outputs that you can precisely control in the app: You can make a podcast, lecture, discussions, short/long form summaries & more, including the use
00:00
208K
Markus J. Buehler
@ProfBuehlerMIT
Sep 21, 2024
Can #AI not only support but actually drive the future of scientific discovery? We are excited to introduce SciAgents💡🔬, an agentic AI aimed towards scientific discovery through the integration of large-scale knowledge graphs, LLMs, and adversarial interactions between multiple
00:00
208K
Markus J. Buehler
@ProfBuehlerMIT
Sep 19, 2024
Introducing LifeGPT, showing that LLMs can simulate complex, Turing-complete systems like Conway's Game of Life with near-perfect accuracy—no prior topology needed.🌐This unlocks new potential for AI in modeling self-organizing systems in biology, materials science, & beyond.🔬🤖
00:00
114K
Markus J. Buehler
@ProfBuehlerMIT
Nov 19, 2024
How can we build AI models that do not just memorize, but learn higher level abstractions? Our new work proposes a multimodal intelligent graph reasoning model that offers a pathway to move beyond simple memorization and towards constructing relational abstractions of the world.
45K
Markus J. Buehler
@ProfBuehlerMIT
Apr 15, 2025
At the molecular level, biological materials like silk and collagen defy conventional logic by building exceptional strength from intrinsically weak chemical interactions: Hydrogen bonds, π–π stacking, and hydrophobic forces. In our latest paper "Design and sustainability of
00:00
27K
Markus J. Buehler
@ProfBuehlerMIT
Sep 29, 2024
Check out #PDF2Audio, an open-source alternative to the amazing #podcast feature of #NotebookLM with flexibility & tailored outputs that you can precisely control in the app: You can make a🎙️podcast,🎓lecture, 🔠discussions,💡short/long form summaries & more, with fine control
00:00
37K
Markus J. Buehler
@ProfBuehlerMIT
Apr 25, 2024
Check out mistral.rs, our #Rust-based open source inference engine allowing for fast #LLM serving for a variety of architectures including X-LoRA mixture-of-expert (MoE) models, Llama-3, Mistral/Mixtral, Gemma & many others. Built on the @huggingface #Candle framework for #Rust
00:00
74K
Markus J. Buehler
@ProfBuehlerMIT
Feb 10, 2024
What seemed like an intractable problem is now possible: To design proteins with a specified nonlinear mechanical response, capturing complex folding and unfolding mechanisms in singe and few-shot computations. We present ForceGen, an end-to-end algorithm for de novo protein
00:00
47K
Markus J. Buehler
@ProfBuehlerMIT
Apr 7, 2025
How can we build AI models that retain general knowledge while acquiring deep, domain-specific understanding - and even generate new ideas grounded in science? We find that, surprisingly, merging two specialized models can create a new model with capabilities neither one had
37K
Markus J. Buehler
@ProfBuehlerMIT
Feb 23, 2025
Replying to @BenFerrum
Yes - here it is:
arxiv.org
Agentic Deep Graph Reasoning Yields Self-Organizing Knowledge Networks
We present an agentic, autonomous graph expansion framework that iteratively structures and refines knowledge in situ. Unlike conventional knowledge graph construction methods relying on static...
8.3K
Markus J. Buehler
@ProfBuehlerMIT
Feb 5, 2025
How can we build AI models that think rather than just memorize? RL-based reasoning models are one way, but there's another. We can also give Transformers new superpowers by replacing their linear attention mechanism with Graph Isomorphism Networks. This turns them into powerful
12K