A short retrospective on the EleutherAI Summer of Open AI Research
In August of last year, EleutherAI hosted the Summer of Open AI Research (SOAR) for the first time. Despite being organized on short notice without any budget, it was quite successful! As we are now organizing the second edition of SOAR, it seemed like a good time to write a retrospective on the event.
How it started
In true Eleuther fashion, the idea for SOAR came from a conversation in the server’s #off-topic channel. Unlike the usual low signal chatter, this particular conversation was about the rise of open source community driven research as done by EAI and other organizations, and the influx of people, often with no prior research experience, joining the EAI discord willing to contribute to projects. We noticed a lack of onboarding mechanism for these people, and decided to address this by organizing a hackathon-style event.
How it went
After gathering projects from researchers from Eleuther, Cohere, Apart, and other organizations, we ended up with 12 projects on topics ranging from mechanistic interpretability to AI safety. Our initial estimates for the number of applications was around 100, which was way off the 500+ applications we ended up receiving over two weeks. After a somewhat chaotic review process, we ended up with 142 people accepted.
From there, things went on reasonably well. Two projects didn’t continue past week one, but the rest kept going all the way, with 99 participants still active at the end of the event. This is a surprisingly decent retention rate in our opinion, though we expect it to be lower for this year’s edition.
Impact
A number of participants kept working on their project after the end of the event, often over several months. This resulted in at least four publications accepted in various conference workshops with a few more under submission. For the majority of the involved participants, this was their first ever publication. Other projects resulted in publishing a Python package or a website.
Overall, participants who made it to the end reported being satisfied with their experience, and that SOAR was a learning experience that enabled them to get accepted into other fellowships, find a job or simply perform better in their current position. For some mentees, SOAR allowed them to learn research skills that they would not have had the opportunity to learn otherwise.
Conclusion and lessons learned
We were very impressed with the outcome of SOAR and what it was able to accomplish despite being initially a community-driven event with little means. Allowing anyone to join and contribute to a research project is at the core of the EleutherAI ethos, and we are quite proud of contributing to that spirit via SOAR.
That doesn’t mean everything went perfectly smoothly however, and a number of rookie mistakes were made during the organization of the event. We have taken note of these issues and made steps to address them in this year’s edition of SOAR, for which you can apply here.
We are excited to see what will come out of this year’s projects!
Appendix: Testimonials
“Just after SOAR ended, I went on to be a MATS fellow in a very theory-heavy research stream. I think having participated in SOAR made the difference between me looking like ‘aspiring AI/ML researcher with lots of math background’ instead of ‘physicist bandwagoning to AI’.”
Charles R. W.“As someone from industry, this was a great opportunity for me to connect with folks looking to contribute to the AI ecosystem, as well as gain a better understanding of academia and ongoing research.”
Abdul R.
“SOAR was my first experience in an AI research program. Working remotely with experienced mentors pushed me to think like a researcher, from reading literature critically to designing experiments systematically to knowing when to ask the right questions. I’d recommend SOAR to any early-stage researcher who wants to bridge the gap between coursework and real research.”
Eryawan P. Y.
“SOAR as a program helped me tremendously as an early-career researcher. Not only did it help me with research and introduce me to previously unknown research areas, it also greatly helped my academic development. My mentor was really kind and helpful, introducing me to research beyond the program’s topics. Highly recommend for students or people who intend to go into research.”
Luis F. S.
“SOAR gave me the opportunity to get my foot in the door and actually contribute to a publication-worthy AI research project. It connected me with talented mentors who helped me build the foundations for my research journey.”
Ayesha I.
Appendix: Papers that came out of SOAR 2025
The following is the list of publications that resulted from SOAR 2025. We may update it in the future as some other papers are still under development.
Christian Zhou-Zheng, John Backsund, Dun Li Chan, Alex Coventry, Avid Eslami, Jyotin Goel, Xingwen Han, Danysh Soomro, Galen Wei, A Traditional Approach to Symbolic Piano Continuation, Accepted as MIREX session at ISMIR 2025. https://arxiv.org/abs/2509.12267
Prerana Rane, Amitesh Vatsa, Yash Pethe, Ogan Batu Aktolun, Kevin Li, Ishan Singh, Attention-Guided Audio Compression for Multimodal LLM. Accepted at the ICASSP 2026 Workshop “Low-Resource Audio Codec”.
Ayesha Imran, Soham Chatterjee, Persona-Vector Routing: A Lightweight, Interpretable Guardrail for Mitigating LLM Hallucinations. Accepted at the NeuRIPS 2025 Workshop on LLM Persona Modeling. https://neurips.cc/virtual/2025/loc/mexico-city/129955
Nahid Alam, Leema Krishna Murali, Siddhant Bharadwaj, Patrick Liu, Timothy Chung, Drishti Sharma, Akshata A, Kranthi Kiran, Wesley Tam, Bala Krishna S Vegesna, The Spatial Blindspot of Vision-Language Models. Accepted at the ICLR 2026 Workshop “I Can’t Believe It’s Not Better”, https://arxiv.org/abs/2601.09954
