Process for Adapting Language Models to Society (PALMS) with Values-Targeted Datasets
Paper
โข
2106.10328
โข
Published
Hugging Face Science
company
AI & ML interests
None defined yet.
Recent Activity
View all activity
science's activity
irenesolaimanย
authored
2
papers
about 12 hours ago
Post
1969
๐ Just launched: A toolkit of 20 powerful AI tools that journalists can use right now - transcribe, analyze, create. 100% free & open-source.
Been testing all these tools myself and created a searchable collection of the most practical ones - from audio transcription to image generation to document analysis. No coding needed, no expensive subscriptions.
Some highlights I've tested personally:
- Private, on-device transcription with speaker ID in 100+ languages using Whisper
- Website scraping that just works - paste a URL, get structured data
- Local image editing with tools like Finegrain (impressive results)
- Document chat using Qwen 2.5 72B (handles technical papers well)
Sharing this early because the best tools come from the community. Drop your favorite tools in the comments or join the discussion on what to add next!
๐ JournalistsonHF/ai-toolkit
Been testing all these tools myself and created a searchable collection of the most practical ones - from audio transcription to image generation to document analysis. No coding needed, no expensive subscriptions.
Some highlights I've tested personally:
- Private, on-device transcription with speaker ID in 100+ languages using Whisper
- Website scraping that just works - paste a URL, get structured data
- Local image editing with tools like Finegrain (impressive results)
- Document chat using Qwen 2.5 72B (handles technical papers well)
Sharing this early because the best tools come from the community. Drop your favorite tools in the comments or join the discussion on what to add next!
๐ JournalistsonHF/ai-toolkit
Post
2595
We now have a Deep Research for academia: SurveyX automatically writes academic surveys nearly indistinguishable from human-written ones ๐ฅ
Researchers from Beijing and Shanghai just published the first application of a deep research system to academia: their algorithm, given a question, can give you a survey of all papers on the subject.
To make a research survey, you generally follow two steps, preparation (collect and organize papers) and writing (outline creation, writing, polishing). Researchers followed the same two steps and automated them.
๐ฏ For the preparation part, a key part is find all the important references on the given subject.
Researchers first cast a wide net of all relevant papers. But then finding the really important ones is like distilling knowledge from a haystack of information. To solve this challenge, they built an โAttributeTreeโ object that structures key information from citations. Ablating these AttributeTrees significantly decreased structure and synthesis scores, so they were really useful!
๐ For the writing part, key was to get a synthesis that's both short and true. This is not easy to get with LLMs! So they used methods like LLM-based deduplication to shorten the too verbose listings made by LLMs, and RAG to grab original quotes instead of made-up ones.
As a result, their system outperforms previous approaches by far!
As assessed by LLM-judges, the quality score os SurveyX even approaches this of human experts, with 4.59/5 vs 4.75/5 ๐
I advise you to read the paper, it's a great overview of the kind of assistants that we'll get in the short future! ๐ SurveyX: Academic Survey Automation via Large Language Models (2502.14776)
Their website shows examples of generated surveys ๐ http://www.surveyx.cn/
Researchers from Beijing and Shanghai just published the first application of a deep research system to academia: their algorithm, given a question, can give you a survey of all papers on the subject.
To make a research survey, you generally follow two steps, preparation (collect and organize papers) and writing (outline creation, writing, polishing). Researchers followed the same two steps and automated them.
๐ฏ For the preparation part, a key part is find all the important references on the given subject.
Researchers first cast a wide net of all relevant papers. But then finding the really important ones is like distilling knowledge from a haystack of information. To solve this challenge, they built an โAttributeTreeโ object that structures key information from citations. Ablating these AttributeTrees significantly decreased structure and synthesis scores, so they were really useful!
๐ For the writing part, key was to get a synthesis that's both short and true. This is not easy to get with LLMs! So they used methods like LLM-based deduplication to shorten the too verbose listings made by LLMs, and RAG to grab original quotes instead of made-up ones.
As a result, their system outperforms previous approaches by far!
As assessed by LLM-judges, the quality score os SurveyX even approaches this of human experts, with 4.59/5 vs 4.75/5 ๐
I advise you to read the paper, it's a great overview of the kind of assistants that we'll get in the short future! ๐ SurveyX: Academic Survey Automation via Large Language Models (2502.14776)
Their website shows examples of generated surveys ๐ http://www.surveyx.cn/
eliebakย
updated
a
Space
3 days ago
Post
3383
Trying something new to keep you ahead of the curve: The 5 AI stories of the week - a weekly curation of the most important AI news you need to know. Do you like it?
For more AI stories and deeper analysis, check out my newsletter: https://open.substack.com/pub/fdaudens/p/ai-competition-heats-up-grok-3-iphone
For more AI stories and deeper analysis, check out my newsletter: https://open.substack.com/pub/fdaudens/p/ai-competition-heats-up-grok-3-iphone
- 1 reply
Post
5727
๐ฏ Perplexity drops their FIRST open-weight model on Hugging Face: A decensored DeepSeek-R1 with full reasoning capabilities. Tested on 1000+ examples for unbiased responses.
Check it out: perplexity-ai/r1-1776
Blog post: https://perplexity.ai/hub/blog/open-sourcing-r1-1776
Check it out: perplexity-ai/r1-1776
Blog post: https://perplexity.ai/hub/blog/open-sourcing-r1-1776
- 1 reply
Post
2815
Less is More for Reasoning (LIMO): a 32B model fine-tuned with 817 examples can beat o1-preview on math reasoning! ๐คฏ
Do we really need o1's huge RL procedure to see reasoning emerge? It seems not.
Researchers from Shanghai Jiaotong University just demonstrated that carefully selected examples can boost math performance in large language models using SFT โno huge datasets or RL procedures needed.
Their procedure allows Qwen2.5-32B-Instruct to jump from 6.5% to 57% on AIME and from 59% to 95% on MATH, while using only 1% of the data in previous approaches.
โก The Less-is-More Reasoning Hypothesis:
โฃ Minimal but precise examples that showcase optimal reasoning patterns matter more than sheer quantity
โฃ Pre-training knowledge plus sufficient computational resources at inference levels up math skills
โก๏ธ Core techniques:
โฃ High-quality reasoning chains with self-verification steps
โฃ 817 handpicked problems that encourage deeper reasoning
โฃ Enough inference-time computation to allow extended reasoning
๐ช Efficiency gains:
โฃ Only 817 examples instead of 100k+
โฃ 40.5% absolute improvement across 10 diverse benchmarks, outperforming models trained on 100x more data
This really challenges the notion that SFT leads to memorization rather than generalization! And opens up reasoning to GPU-poor researchers ๐
Read the full paper here ๐ย LIMO: Less is More for Reasoning (2502.03387)
Do we really need o1's huge RL procedure to see reasoning emerge? It seems not.
Researchers from Shanghai Jiaotong University just demonstrated that carefully selected examples can boost math performance in large language models using SFT โno huge datasets or RL procedures needed.
Their procedure allows Qwen2.5-32B-Instruct to jump from 6.5% to 57% on AIME and from 59% to 95% on MATH, while using only 1% of the data in previous approaches.
โก The Less-is-More Reasoning Hypothesis:
โฃ Minimal but precise examples that showcase optimal reasoning patterns matter more than sheer quantity
โฃ Pre-training knowledge plus sufficient computational resources at inference levels up math skills
โก๏ธ Core techniques:
โฃ High-quality reasoning chains with self-verification steps
โฃ 817 handpicked problems that encourage deeper reasoning
โฃ Enough inference-time computation to allow extended reasoning
๐ช Efficiency gains:
โฃ Only 817 examples instead of 100k+
โฃ 40.5% absolute improvement across 10 diverse benchmarks, outperforming models trained on 100x more data
This really challenges the notion that SFT leads to memorization rather than generalization! And opens up reasoning to GPU-poor researchers ๐
Read the full paper here ๐ย LIMO: Less is More for Reasoning (2502.03387)
Post
2256
Will we soon all have our own personalized AI news agents? And what does it mean for journalism?
Just built a simple prototype based on the Hugging Face course. It lets you get customized news updates on any topic.
Not perfect yet, but you can see where things could go: we'll all be able to build personalized AI agents that curate & analyze news for each of us. And users who could decide to build custom news products for their needs, such as truly personalized newsletters or podcasts.
The implications for both readers & news organizations are significant. To name a few:
- Will news articles remain the best format for informing people?
- What monetization model will work for news organizations?
- How do you create an effective conversion funnel?
๐ Try it here: fdaudens/my-news-agent (Code is open-source)
๐ Check out the course: https://huggingface.co/learn/agents-course/unit0/introduction
Just built a simple prototype based on the Hugging Face course. It lets you get customized news updates on any topic.
Not perfect yet, but you can see where things could go: we'll all be able to build personalized AI agents that curate & analyze news for each of us. And users who could decide to build custom news products for their needs, such as truly personalized newsletters or podcasts.
The implications for both readers & news organizations are significant. To name a few:
- Will news articles remain the best format for informing people?
- What monetization model will work for news organizations?
- How do you create an effective conversion funnel?
๐ Try it here: fdaudens/my-news-agent (Code is open-source)
๐ Check out the course: https://huggingface.co/learn/agents-course/unit0/introduction
Post
2106
๐ Meet Kokoro Web - Free, ML speech synthesis on your computer, that'll make you ditch paid services!
28 natural voices, unlimited generations, and WebGPU acceleration. Perfect for journalists and content creators.
Test it with full articlesโsounds amazingly human! ๐ฏ๐๏ธ
Xenova/kokoro-web
28 natural voices, unlimited generations, and WebGPU acceleration. Perfect for journalists and content creators.
Test it with full articlesโsounds amazingly human! ๐ฏ๐๏ธ
Xenova/kokoro-web
Post
2751
๐๐ฟ๐ฒ๐ฎ๐ ๐ณ๐ฒ๐ฎ๐๐๐ฟ๐ฒ ๐ฎ๐น๐ฒ๐ฟ๐: you can now share agents to the Hub! ๐ฅณ๐ฅณ
And any agent pushed to Hub get a cool Space interface to directly chat with it.
This was a real technical challenge: for instance, serializing tools to export them meant that you needed to get all the source code for a tool, verify that it was standalone (not relying on external variables), and gathering all the packages required to make it run.
Go try it out! ๐ https://github.com/huggingface/smolagents
And any agent pushed to Hub get a cool Space interface to directly chat with it.
This was a real technical challenge: for instance, serializing tools to export them meant that you needed to get all the source code for a tool, verify that it was standalone (not relying on external variables), and gathering all the packages required to make it run.
Go try it out! ๐ https://github.com/huggingface/smolagents
- 2 replies
Post
2410
For those who haven't come across it yet, here's a handy trick to discuss an entire GitHub repo with an LLM:
=> Just replace "github" with "gitingest" in the url, and you get the whole repo as a single string that you can then paste in your LLMs
=> Just replace "github" with "gitingest" in the url, and you get the whole repo as a single string that you can then paste in your LLMs
Post
2675
โญ๏ธ The AI Energy Score project just launched - this is a game-changer for making informed decisions about AI deployment.
You can now see exactly how much energy your chosen model will consume, with a simple 5-star rating system. Think appliance energy labels, but for AI.
Looking at transcription models on the leaderboard is fascinating: choosing between whisper-tiny or whisper-large-v3 can make a 7x difference. Real-time data on these tradeoffs changes everything.
166 models already evaluated across 10 different tasks, from text generation to image classification. The whole thing is public and you can submit your own models to test.
Why this matters:
- Teams can pick efficient models that still get the job done
- Developers can optimize for energy use from day one
- Organizations can finally predict their AI environmental impact
If you're building with AI at any scale, definitely worth checking out.
๐ leaderboard: https://lnkd.in/esrSxetj
๐ blog post: https://lnkd.in/eFJvzHi8
Huge work led by @sasha with @bgamazay @yjernite @sarahooker @regisss @meg
You can now see exactly how much energy your chosen model will consume, with a simple 5-star rating system. Think appliance energy labels, but for AI.
Looking at transcription models on the leaderboard is fascinating: choosing between whisper-tiny or whisper-large-v3 can make a 7x difference. Real-time data on these tradeoffs changes everything.
166 models already evaluated across 10 different tasks, from text generation to image classification. The whole thing is public and you can submit your own models to test.
Why this matters:
- Teams can pick efficient models that still get the job done
- Developers can optimize for energy use from day one
- Organizations can finally predict their AI environmental impact
If you're building with AI at any scale, definitely worth checking out.
๐ leaderboard: https://lnkd.in/esrSxetj
๐ blog post: https://lnkd.in/eFJvzHi8
Huge work led by @sasha with @bgamazay @yjernite @sarahooker @regisss @meg
- 1 reply
Post
4735
"๐ฎ๐ฌ๐ฎ๐ฑ ๐๐ถ๐น๐น ๐ฏ๐ฒ ๐๐ต๐ฒ ๐๐ฒ๐ฎ๐ฟ ๐ผ๐ณ ๐๐ ๐ฎ๐ด๐ฒ๐ป๐๐": this statement has often been made, here are numbers to support it.
I've plotted the progress of AI agents on GAIA test set, and it seems they're headed to catch up with the human baseline in early 2026.
And that progress is still driven mostly by the improvement of base LLMs: progress would be even faster with fine-tuned agentic models.
I've plotted the progress of AI agents on GAIA test set, and it seems they're headed to catch up with the human baseline in early 2026.
And that progress is still driven mostly by the improvement of base LLMs: progress would be even faster with fine-tuned agentic models.
Post
4537
Introducing OpenR1-Math-220k!
open-r1/OpenR1-Math-220k
The community has been busy distilling DeepSeek-R1 from inference providers, but we decided to have a go at doing it ourselves from scratch ๐ช
Whatโs new compared to existing reasoning datasets?
โพ Based on AI-MO/NuminaMath-1.5: we focus on math reasoning traces and generate answers for problems in NuminaMath 1.5, an improved version of the popular NuminaMath-CoT dataset.
๐ณ 800k R1 reasoning traces: We generate two answers for 400k problems using DeepSeek R1. The filtered dataset contains 220k problems with correct reasoning traces.
๐ 512 H100s running locally: Instead of relying on an API, we leverage vLLM and SGLang to run generations locally on our science cluster, generating 180k reasoning traces per day.
โณ Automated filtering: We apply Math Verify to only retain problems with at least one correct answer. We also leverage Llama3.3-70B-Instruct as a judge to retrieve more correct examples (e.g for cases with malformed answers that canโt be verified with a rules-based parser)
๐ We match the performance of DeepSeek-Distill-Qwen-7B by finetuning Qwen-7B-Math-Instruct on our dataset.
๐ Read our blog post for all the nitty gritty details: https://huggingface.co/blog/open-r1/update-2
open-r1/OpenR1-Math-220k
The community has been busy distilling DeepSeek-R1 from inference providers, but we decided to have a go at doing it ourselves from scratch ๐ช
Whatโs new compared to existing reasoning datasets?
โพ Based on AI-MO/NuminaMath-1.5: we focus on math reasoning traces and generate answers for problems in NuminaMath 1.5, an improved version of the popular NuminaMath-CoT dataset.
๐ณ 800k R1 reasoning traces: We generate two answers for 400k problems using DeepSeek R1. The filtered dataset contains 220k problems with correct reasoning traces.
๐ 512 H100s running locally: Instead of relying on an API, we leverage vLLM and SGLang to run generations locally on our science cluster, generating 180k reasoning traces per day.
โณ Automated filtering: We apply Math Verify to only retain problems with at least one correct answer. We also leverage Llama3.3-70B-Instruct as a judge to retrieve more correct examples (e.g for cases with malformed answers that canโt be verified with a rules-based parser)
๐ We match the performance of DeepSeek-Distill-Qwen-7B by finetuning Qwen-7B-Math-Instruct on our dataset.
๐ Read our blog post for all the nitty gritty details: https://huggingface.co/blog/open-r1/update-2
Post
1292
๐ฅ Video AI is taking over! Out of 17 papers dropped on Hugging Face today, 6 are video-focused - from Sliding Tile Attention to On-device Sora. The race for next-gen video tech is heating up! ๐ฌ๐
Post
3673
๐๐ฑ๐๐ฒ๐ป'๐ ๐ป๐ฒ๐ ๐๐ฎ๐๐ฎ ๐๐ด๐ฒ๐ป๐๐ ๐๐ฒ๐ป๐ฐ๐ต๐บ๐ฎ๐ฟ๐ธ ๐๐ต๐ผ๐๐ ๐๐ต๐ฎ๐ ๐๐ฒ๐ฒ๐ฝ๐ฆ๐ฒ๐ฒ๐ธ-๐ฅ๐ญ ๐๐๐ฟ๐๐ด๐ด๐น๐ฒ๐ ๐ผ๐ป ๐ฑ๐ฎ๐๐ฎ ๐๐ฐ๐ถ๐ฒ๐ป๐ฐ๐ฒ ๐๐ฎ๐๐ธ๐! โ
โก๏ธ How well do reasoning models perform on agentic tasks? Until now, all indicators seemed to show that they worked really well. On our recent reproduction of Deep Search, OpenAI's o1 was by far the best model to power an agentic system.
So when our partner Adyen built a huge benchmark of 450 data science tasks, and built data agents with smolagents to test different models, I expected reasoning models like o1 or DeepSeek-R1 to destroy the tasks at hand.
๐ But they really missed the mark. DeepSeek-R1 only got 1 or 2 out of 10 questions correct. Similarly, o1 was only at ~13% correct answers.
๐ง These results really surprised us. We thoroughly checked them, we even thought our APIs for DeepSeek were broken and colleagues Leandro Anton helped me start custom instances of R1 on our own H100s to make sure it worked well.
But there seemed to be no mistake. Reasoning LLMs actually did not seem that smart. Often, these models made basic mistakes, like forgetting the content of a folder that they had just explored, misspelling file names, or hallucinating data. Even though they do great at exploring webpages through several steps, the same level of multi-step planning seemed much harder to achieve when reasoning over files and data.
It seems like there's still lots of work to do in the Agents x Data space. Congrats to Adyen for this great benchmark, looking forward to see people proposing better agents! ๐
Read more in the blog post ๐ https://huggingface.co/blog/dabstep
โก๏ธ How well do reasoning models perform on agentic tasks? Until now, all indicators seemed to show that they worked really well. On our recent reproduction of Deep Search, OpenAI's o1 was by far the best model to power an agentic system.
So when our partner Adyen built a huge benchmark of 450 data science tasks, and built data agents with smolagents to test different models, I expected reasoning models like o1 or DeepSeek-R1 to destroy the tasks at hand.
๐ But they really missed the mark. DeepSeek-R1 only got 1 or 2 out of 10 questions correct. Similarly, o1 was only at ~13% correct answers.
๐ง These results really surprised us. We thoroughly checked them, we even thought our APIs for DeepSeek were broken and colleagues Leandro Anton helped me start custom instances of R1 on our own H100s to make sure it worked well.
But there seemed to be no mistake. Reasoning LLMs actually did not seem that smart. Often, these models made basic mistakes, like forgetting the content of a folder that they had just explored, misspelling file names, or hallucinating data. Even though they do great at exploring webpages through several steps, the same level of multi-step planning seemed much harder to achieve when reasoning over files and data.
It seems like there's still lots of work to do in the Agents x Data space. Congrats to Adyen for this great benchmark, looking forward to see people proposing better agents! ๐
Read more in the blog post ๐ https://huggingface.co/blog/dabstep
Post
2096
๐ข SmolLM2 paper released! Learn how the ๐ค team built one of the best small language models: from data choices to training insights. Check out our findings and share your thoughts! ๐ค๐ก
Check it out: SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model (2502.02737)
Check it out: SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model (2502.02737)
plagussย
authored
a
paper
19 days ago
anditoย
authored
a
paper
19 days ago