Medical AI Research Center (MedARC) Unlocking new possibilities in medical AI research. Supported by @SophontAI Founded by @iScienceLuvr
Joined January 2023
- Tweets156
- Following12
- Followers7K
- Likes126
A fully autonomous agent topped the nanopath leaderboard earlier last week.
Shortly afterwards, a human-created submission topped it again...
A competition between humans and agents ensues!
RSI, proven again. 🧠🔬
ScienceGuru + Guru Turbo 1.0 is now the #1 maintainer-validated trainable run on MedARC's NanoPath — an open, FLOP-capped benchmark for training pathology foundation models, competing against recipes from human research labs (MSK, Oregon, UBC, and others).
Our autonomous research system ran the full loop itself: generated hypotheses, wrote the code, ran controlled experiments on a single H100, and iterated its way to the top recipe.
Score: 0.6597 — a +0.025 jump over the prior best community recipe, with gains across 5 of 6 metric families (classification, segmentation, progression, mutation, survival).
This is the second independent, maintainer-validated #1 for our recursive self-improvement stack — after Autoresearch@Home. Different scientific domain. Zero recipe transfer. Same system. The harness generalizes.
That's the RSI claim in falsifiable form: an AI that improves AI, on open leaderboards anyone can check — fully open source, as the benchmark requires:
github.com/AutoTrustAI/nanop…
Recursive self-improvement isn't a slogan. It's a leaderboard entry.
More domains coming.
@MedARC_AI @ScienceGuruAI
ALT RSI, proven again. 🧠🔬 ScienceGuru + Guru Turbo 1.0 is now the #1 maintainer-validated trainable run on MedARC's NanoPath — an open, FLOP-capped benchmark for training pathology foundation models, competing against recipes from human research labs (MSK, Oregon, UBC, and others). Our autonomous research system ran the full loop itself: generated hypotheses, wrote the code, ran controlled experiments on a single H100, and iterated its way to the top recipe. Score: 0.6597 — a +0.025 jump over the prior best community recipe, with gains across 5 of 6 metric families (classification, segmentation, progression, mutation, survival). This is the second independent, maintainer-validated #1 for our recursive self-improvement stack — after Autoresearch@Home. Different scientific domain. Zero recipe transfer. Same system. The harness generalizes. That's the RSI claim in falsifiable form: an AI that improves AI, on open leaderboards anyone can check — fully open source, as the benchmark requi
1
4
856
Starting now!!!
We are hosting yet another awesome Journal Club presentation on July 30th at 8:30am PT!
@TimDarcet (researcher at Meta) will be presenting his paper "CAPI: Cluster and Predict Latent Patches for Improved Masked Image Modeling"
Join the Discord and check our website!
820
We are hosting yet another awesome Journal Club presentation on July 30th at 8:30am PT!
@TimDarcet (researcher at Meta) will be presenting his paper "CAPI: Cluster and Predict Latent Patches for Improved Masked Image Modeling"
Join the Discord and check our website!
STARTING NOW!
We're excited to announce we're starting a Journal Club. And our first meeting is scheduled for tomorrow!
@__init_self will present her work, CorText: Brain-Language Fusion Enables Interactive Neural Readout and In-Silico Experimentation
Tomorrow at 10:15am ET, join Discord!
1
6,156
We're excited to announce we're starting a Journal Club. And our first meeting is scheduled for tomorrow!
@__init_self will present her work, CorText: Brain-Language Fusion Enables Interactive Neural Readout and In-Silico Experimentation
Tomorrow at 10:15am ET, join Discord!
Medmarks was organized within MedARC, join the server to contribute to our continued eval efforts and medical post-training efforts!
This release was only possible by the numerous MedARC volunteers who implemented benchmarks and datasets to evaluate with. Grateful to all those who contributed!
We're releasing Medmarks v0.1, the largest completely open-source automated evaluation suite for assessing the medical capabilities of LLMs!
Developed in our @MedARC_AI community, w/ support from @PrimeIntellect
So far we’ve explored 46 models to figure out the best!
The MedARC community was in full attendance at NeurIPS!
Sophont had a great NeurIPS last week!
We presented our fMRI foundation model research at the Brain and Body Foundation Models workshop (which we also co-sponsored!)
We also held a social with @KindredVentures (and @_CausalLabs, @fal) including a great panel with our CEO Tanishq Abraham, MIT prof Paul Liang, Stanford prof James Zou, moderated by Kanyi Maqubela from Kindred. We discussed the importance of multimodal models for medicine, open-source, open research problems in the space, agents for accelerating scientific discovery, and so much more.
It was great to connect with so many folks interested in medical AI, see you next time!
Sharing our work on training fMRI neuroimaging foundation models, to be presented at FMs for Brain & Body NeurIPS 2025 workshop
Although this was an internal Sophont research project, we are now expanding this at MedARC
If you're interested in contributing, join us in Discord!!
We are supported by @SophontAI, a public benefit corporation that operates MedARC and separately facilitates commercial projects with medtech, hospitals, and pharma R&D. (2/5)
1
11
852
Nice work by MedARC team!
We came in 4th place in the @AlgonautsProj 2025 challenge! Goal was to predict brain activity to movie watching.
Incredibly lucky to work w/ @MedARC_AI team where we got very competitive results despite the simplest model arch (no transformers, just conv + linear)
Paper below!
ALT MedARC's model architecture for Algonauts challenge
We are presenting MindEye2 at the 1:30pm poster session at ICML today, stop by!
Announcement! 📢
We're excited to share that MindEye2, our latest SOTA fMRI-to-image paper, was accepted to #ICML2024!
We have uploaded the camera-ready version of the preprint on arXiv: arxiv.org/abs/2403.11207
Website: medarc.ai/mindeye2
Announcement! 📢
We're excited to share that MindEye2, our latest SOTA fMRI-to-image paper, was accepted to #ICML2024!
We have uploaded the camera-ready version of the preprint on arXiv: arxiv.org/abs/2403.11207
Website: medarc.ai/mindeye2
A reminder that MindEye2 achieves high-quality reconstructions of images from fMRI using only 1 hour of training data! It also achieves a new SOTA with the full 30-40h of training data.
Full thread about the original results here:
xitter.twitterapi.xyz/MedARC_AI/status/17699…
Announcing MindEye2: Shared-Subject Models Enable fMRI-To-Image With 1 Hour of Data! 🧠👁️
High-quality reconstructions of images from fMRI using 1 hour of training data vs. past work using 30-40 hrs/person!
arXiv: arxiv.org/abs/2403.11207
project page: medarc.ai/mindeye2
1
1
197
MindEye demonstrated incredible brain decoding performance w/ fMRI
But an fMRI machine is huge! Can brain decoding be done with a more portable device?
@humanscotti joined a team of researchers to explore if the more portable fNIRS tech can perform accurate brain decoding ↓
Our Head of Neuroimaging, Paul, joined @labenz on The Cognitive Revolution Podcast to talk about MindEye2.
Check it out if you want a deep dive about collecting fMRI data, using foundation models for fMRI decoding, MindEye1 vs. MindEye2, ethics of brain data usage, and more!
My appearance on @CogRev_Podcast "Reading Minds from Shared Latent Space" is out!
Thanks @labenz for having me on! he does a great job summarizing while still asking in depth about the research
Talked about MindEye2, foundation brain models, BCIs, etc.
youtu.be/7_BS8tuUoZY