UniverseTBD
@universetbd.org
We are on a mission to democratise science for everyone. Join our Discord at https://discord.gg/RH2jgT3vtQ, or contact us at [email protected].
Pinned
UniverseTBD
@universetbd.org
Β· Apr 13
π HAPPY BIRTHDAY, UniverseTBD! π
As we turn 2, weβre going 2^2.
Launching a new project per day for the next four days.
We hope that you all enjoy these works as much as we have enjoyed working on them. Stay tuned for the big reveals!
As we turn 2, weβre going 2^2.
Launching a new project per day for the next four days.
We hope that you all enjoy these works as much as we have enjoyed working on them. Stay tuned for the big reveals!
Reposted by UniverseTBD
New AstroPT models are out ππ This time trained with an improved DESI galaxy image dataset. Link here: huggingface.co/Smith42/astr...
Check out these new scaling curves!
We are still seeing improvement at 800M parameters where before we stalled at 100M. Maybe high quality data is all you need π€
Check out these new scaling curves!
We are still seeing improvement at 800M parameters where before we stalled at 100M. Maybe high quality data is all you need π€
June 2, 2025 at 6:56 PM
New AstroPT models are out ππ This time trained with an improved DESI galaxy image dataset. Link here: huggingface.co/Smith42/astr...
Check out these new scaling curves!
We are still seeing improvement at 800M parameters where before we stalled at 100M. Maybe high quality data is all you need π€
Check out these new scaling curves!
We are still seeing improvement at 800M parameters where before we stalled at 100M. Maybe high quality data is all you need π€
HypoGen concludes our 2^2 fest and we truly hope you enjoyed it πβ¨. Thank you for your great support with our mission to democratise science for everyoneπ.
April 18, 2025 at 5:57 PM
HypoGen concludes our 2^2 fest and we truly hope you enjoyed it πβ¨. Thank you for your great support with our mission to democratise science for everyoneπ.
8/n A huge thank you to our partners at @msftresearch.bsky.social for enabling our research through the AFMR grant. And to our many friends around the world in academia and industry - this wouldn't be possible without your support π.
April 18, 2025 at 5:57 PM
8/n A huge thank you to our partners at @msftresearch.bsky.social for enabling our research through the AFMR grant. And to our many friends around the world in academia and industry - this wouldn't be possible without your support π.
Huge thanks to Pranav Agarwal for the last minute eval request, we couldn't have done this without you! π«
April 18, 2025 at 5:57 PM
Huge thanks to Pranav Agarwal for the last minute eval request, we couldn't have done this without you! π«
7/n HypoGen was led by the absolute star @charlesoneill.bsky.social working with our wonderful mentors Tirthankar Ghosal, Roberta Raileanu, Mike Walmsley, Thang Bui, @kevinschawinski.bsky.social, @errai34.bsky.social and our teamπ.
April 18, 2025 at 5:57 PM
7/n HypoGen was led by the absolute star @charlesoneill.bsky.social working with our wonderful mentors Tirthankar Ghosal, Roberta Raileanu, Mike Walmsley, Thang Bui, @kevinschawinski.bsky.social, @errai34.bsky.social and our teamπ.
6/n Future directions: expand HypoGen to domains like astrophysics, biology, materials science, and build AI that doesnβt just answer questions but sparks them. ππ
Let us know here if you want to dive in & letβs push scientific discovery forward!
#HypoGen #AI4Science #DemocratisingScience
Let us know here if you want to dive in & letβs push scientific discovery forward!
#HypoGen #AI4Science #DemocratisingScience
April 18, 2025 at 5:57 PM
6/n Future directions: expand HypoGen to domains like astrophysics, biology, materials science, and build AI that doesnβt just answer questions but sparks them. ππ
Let us know here if you want to dive in & letβs push scientific discovery forward!
#HypoGen #AI4Science #DemocratisingScience
Let us know here if you want to dive in & letβs push scientific discovery forward!
#HypoGen #AI4Science #DemocratisingScience
5/n Humans come out on top (~85% win rate) - a comforting result that hints at a vision for the future when AI and human researchers work together to advance scientific discovery π€.
April 18, 2025 at 5:57 PM
5/n Humans come out on top (~85% win rate) - a comforting result that hints at a vision for the future when AI and human researchers work together to advance scientific discovery π€.
4/n We fineβtuned LLaMA 3.1 8B and its R1βdistilled variant on HypoGen (4βbit quant + LoRA), then evaluated with perplexity, IAScore, and a couple of LLM judges coupled with human verification. We obtain significant gains in hypothesis novelty & feasibility with transparent reasoning steps! π
April 18, 2025 at 5:57 PM
4/n We fineβtuned LLaMA 3.1 8B and its R1βdistilled variant on HypoGen (4βbit quant + LoRA), then evaluated with perplexity, IAScore, and a couple of LLM judges coupled with human verification. We obtain significant gains in hypothesis novelty & feasibility with transparent reasoning steps! π
3/n HypoGen deets:
β’ 5,478 samples from NeurIPS 2023 & ICLR 2024
β’ JSON fields: bit, spark, flip, chain_of_reasoning
β’ Extraction courtesy of @OpenAI's tireless o1 model (no coffee requiredβ¦ maybe). π€β
β’ 5,478 samples from NeurIPS 2023 & ICLR 2024
β’ JSON fields: bit, spark, flip, chain_of_reasoning
β’ Extraction courtesy of @OpenAI's tireless o1 model (no coffee requiredβ¦ maybe). π€β
April 18, 2025 at 5:57 PM
3/n HypoGen deets:
β’ 5,478 samples from NeurIPS 2023 & ICLR 2024
β’ JSON fields: bit, spark, flip, chain_of_reasoning
β’ Extraction courtesy of @OpenAI's tireless o1 model (no coffee requiredβ¦ maybe). π€β
β’ 5,478 samples from NeurIPS 2023 & ICLR 2024
β’ JSON fields: bit, spark, flip, chain_of_reasoning
β’ Extraction courtesy of @OpenAI's tireless o1 model (no coffee requiredβ¦ maybe). π€β
2/n Whereβs the creativity? HypoGen reframes scientific hypothesis generation as a conditional LM task: feed it the Bit (problem) β get the Spark (4β6 word insight), Flip (solution), plus an explicit ChainβofβReasoning (How did the Bit turn into the Flip). π§ π
April 18, 2025 at 5:57 PM
2/n Whereβs the creativity? HypoGen reframes scientific hypothesis generation as a conditional LM task: feed it the Bit (problem) β get the Spark (4β6 word insight), Flip (solution), plus an explicit ChainβofβReasoning (How did the Bit turn into the Flip). π§ π
π’ New dataset out!
We introduce HypoGenπ₯, a dataset of ~5.5K structured problemβhypothesis pairs (BitβFlipβSpark + ChainβofβReasoning) to advance LLM-driven scientific ideationπ‘.
Fineβtuned LLaMA 3.1 8B & R1βdistilled models show significant gains. Humans are still the bestπ₯.
We introduce HypoGenπ₯, a dataset of ~5.5K structured problemβhypothesis pairs (BitβFlipβSpark + ChainβofβReasoning) to advance LLM-driven scientific ideationπ‘.
Fineβtuned LLaMA 3.1 8B & R1βdistilled models show significant gains. Humans are still the bestπ₯.
April 18, 2025 at 5:57 PM
π’ New dataset out!
We introduce HypoGenπ₯, a dataset of ~5.5K structured problemβhypothesis pairs (BitβFlipβSpark + ChainβofβReasoning) to advance LLM-driven scientific ideationπ‘.
Fineβtuned LLaMA 3.1 8B & R1βdistilled models show significant gains. Humans are still the bestπ₯.
We introduce HypoGenπ₯, a dataset of ~5.5K structured problemβhypothesis pairs (BitβFlipβSpark + ChainβofβReasoning) to advance LLM-driven scientific ideationπ‘.
Fineβtuned LLaMA 3.1 8B & R1βdistilled models show significant gains. Humans are still the bestπ₯.
This work was supported by Microsoft's Accelerating Foundation Models Research program and the ITER Teide HPC cluster. Thanks to all collaborators across our many institutions!
April 16, 2025 at 12:17 PM
This work was supported by Microsoft's Accelerating Foundation Models Research program and the ITER Teide HPC cluster. Thanks to all collaborators across our many institutions!
For researchers wanting to collaborate, we're available at discord.gg/PUR2FbFRZ4 and our DMs are open. Check out our code at w3id.org/UniverseTBD/..., and come find us at SCI-FM@ICLR if you would like to chat in person!
Join the UniverseTBD Discord Server!
Check out the UniverseTBD community on Discord - hang out with 161 other members and enjoy free voice and text chat.
discord.gg
April 16, 2025 at 12:17 PM
For researchers wanting to collaborate, we're available at discord.gg/PUR2FbFRZ4 and our DMs are open. Check out our code at w3id.org/UniverseTBD/..., and come find us at SCI-FM@ICLR if you would like to chat in person!
We see a future where multimodal models can reason across astronomical data types beyond just imagery: from spectra to light curves to data cubes.
Join the UniverseTBD Discord Server!
Check out the UniverseTBD community on Discord - hang out with 161 other members and enjoy free voice and text chat.
discord.gg
April 16, 2025 at 12:17 PM
We see a future where multimodal models can reason across astronomical data types beyond just imagery: from spectra to light curves to data cubes.
We've evaluated AstroLLaVA on the Galaxy 10 DECaLS dataset and are releasing the model weights, code, and training dataset under the MIT license to support open science and further development by the community.
April 16, 2025 at 12:17 PM
We've evaluated AstroLLaVA on the Galaxy 10 DECaLS dataset and are releasing the model weights, code, and training dataset under the MIT license to support open science and further development by the community.
Our two-stage fine-tuning process adapts the model for both image captioning and visual question answering in the astronomy domain, making complex astronomical concepts more accessible through natural conversation
April 16, 2025 at 12:17 PM
Our two-stage fine-tuning process adapts the model for both image captioning and visual question answering in the astronomy domain, making complex astronomical concepts more accessible through natural conversation
We fine-tuned LLaVA on ~30k astronomical images with captions & QA pairs from NASA APOD, ESO, and Hubble archives to create a model that understands astronomical concepts in visual form π hf.co/datasets/UniverseTBD/AstroLLaVA_convos
April 16, 2025 at 12:17 PM
We fine-tuned LLaVA on ~30k astronomical images with captions & QA pairs from NASA APOD, ESO, and Hubble archives to create a model that understands astronomical concepts in visual form π hf.co/datasets/UniverseTBD/AstroLLaVA_convos
Excited to announce our new paper as part of our 2^2 week: AstroLLaVA, a vision language model for astronomy that enables natural dialogue with astronomical imagery! Shout out to Sharaf Zaman for leading this work arxiv.org/abs/2504.08583 πβοΈ
April 16, 2025 at 12:17 PM
Excited to announce our new paper as part of our 2^2 week: AstroLLaVA, a vision language model for astronomy that enables natural dialogue with astronomical imagery! Shout out to Sharaf Zaman for leading this work arxiv.org/abs/2504.08583 πβοΈ
The UniverseTBD is extremely grateful to our partners at @msftresearch.bsky.social for their continuous support that enables our research.
April 15, 2025 at 12:48 PM
The UniverseTBD is extremely grateful to our partners at @msftresearch.bsky.social for their continuous support that enables our research.
For more updates and behind-the-scenes breakthroughs, follow us at bsky.app/profile/univ... as we continue to break through the barriers of the sky! π
bsky.app
April 15, 2025 at 12:48 PM
For more updates and behind-the-scenes breakthroughs, follow us at bsky.app/profile/univ... as we continue to break through the barriers of the sky! π
Dive into the full paper and explore the future of hypothesis generation here π arxiv.org/pdf/2504.054...
#HypoGen #AI
#HypoGen #AI
arxiv.org
April 15, 2025 at 12:48 PM
Dive into the full paper and explore the future of hypothesis generation here π arxiv.org/pdf/2504.054...
#HypoGen #AI
#HypoGen #AI
We break down innovative approaches like direct prompting, adversarial methods, fine-tuning, knowledge integration, and even multi-agent systems, transforming how we turn vast scientific literature into actionable, testable ideas
April 15, 2025 at 12:48 PM
We break down innovative approaches like direct prompting, adversarial methods, fine-tuning, knowledge integration, and even multi-agent systems, transforming how we turn vast scientific literature into actionable, testable ideas
Authored by Atilla Kaan Alkan and the UniverseTBD team, this paper provides a single, comprehensive resource that covers everything from human-centric methods to cutting-edge LLM-driven techniques
April 15, 2025 at 12:48 PM
Authored by Atilla Kaan Alkan and the UniverseTBD team, this paper provides a single, comprehensive resource that covers everything from human-centric methods to cutting-edge LLM-driven techniques
Our 2^2 celebration is still in full swing! π
Today weβre launching our latest, must-read survey paper:
βA Survey on Hypothesis Generation for Scientific Discovery in the Era of Large Language Models.β
Check it out! arxiv.org/pdf/2504.054...
π
Today weβre launching our latest, must-read survey paper:
βA Survey on Hypothesis Generation for Scientific Discovery in the Era of Large Language Models.β
Check it out! arxiv.org/pdf/2504.054...
π
April 15, 2025 at 12:48 PM
Our 2^2 celebration is still in full swing! π
Today weβre launching our latest, must-read survey paper:
βA Survey on Hypothesis Generation for Scientific Discovery in the Era of Large Language Models.β
Check it out! arxiv.org/pdf/2504.054...
π
Today weβre launching our latest, must-read survey paper:
βA Survey on Hypothesis Generation for Scientific Discovery in the Era of Large Language Models.β
Check it out! arxiv.org/pdf/2504.054...
π