EleutherAI
@AiEleuther · joined 23 Aug 2022
A non-profit research lab focused on interpretability, alignment, and ethics of AI. Creators of Pythia, VQGAN-CLIP, and using SAEs for interp
28 740Followers
105Following
951Posts total
275.8KViews on collected posts
Neueste Beiträge
Wir suchen neue Beiträge…
AI2 models 🤝 EleutherAI interp tooling
Open science
6.8K views · 60 likes
· 7 reposts · 5 replies
Open on X →
Don’t miss our latest collaboration with @huggingface on building more open data!
2.9K views · 14 likes
· 1 reposts · 0 replies
Open on X →
Are open OCR models good enough to unlock historical knowledge? @vanstriendaniel and I built a leaderboard to find out: 14 open models scored on 2,165 expert-transcribed pages from @BioDivLibrary. Meet FineBooks, from @huggingface and @AiEleuther! https://t.co/cKTAxpy7oJ
14.5K views · 39 likes
· 16 reposts · 1 replies
Open on X →
Frontier models are quite scary and dangerous, but hoping that powerful offensive capabilities are not used for ill is foolish. Open sharing of tooling and info is the bedrock of cybersec, and we are proud to be a part of a collation working to use AI to improve the ecosystem. ht
2.9K views · 10 likes
· 2 reposts · 1 replies
Open on X →
Wonderful to see this work start to take off!
2.8K views · 3 likes
· 0 reposts · 0 replies
Open on X →
🚀 Exciting to see Evaluating Evaluations in the media!
Over the past weeks, our initiative has been featured by both @IBMResearch and @mozilla, highlighting the importance of evaluations for measuring AI progress.
📖 Mozilla: https://t.co/tHnvYEx2gk
1/3
4.6K views · 21 likes
· 3 reposts · 2 replies
Open on X →
@mozilla @shv @Livermore_Comp This dataset was previewed at the Datasets Convening we co-hosted with @mozilla to consult with leading experts in open datasets.
Read more about the event: https://t.co/LtgS7yE32T
And the paper distilling the best practices participants identified
1.5K views · 16 likes
· 1 reposts · 0 replies
Open on X →
This work was supported by @mozilla @shv, the National Sciences and Engineering Research Council of Canada, and @Livermore_Comp. As always, this work represents the opinions of the authors and is not necessarily endorsed by its sponsors.
1.6K views · 17 likes
· 0 reposts · 1 replies
Open on X →
This was a huge effort across a dozen organizations. Thank you to all the authors: @kandpal_nikhil @blester125 @colinraffel @soldni @BlancheMinerva @EnricoShippole A Feder Cooper @aviskowron @Void13950782 @bbrabbasi @storytracer @lintangsutawika @AlbalakAlon @LoubnaBenAllal1
1.2K views · 18 likes
· 0 reposts · 1 replies
Open on X →
@jd_pressman @eliebakouch @SkyLi0n @gui_penedo @dashstander @bartoldson @bkailkhu @tomgoldsteincs @ZhenlinX @guangysong and Tyler Murray.
1.2K views · 20 likes
· 0 reposts · 1 replies
Open on X →
For more, check out...
Paper: https://t.co/FdjRmtPG0N
Artifacts: https://t.co/Ab2qekWqHv
GitHub: https://t.co/1NVQJjuDRj
EleutherAI's blog post: https://t.co/bdP1HADFTM
Coverage in @washingtonpost by @nitashatiku: https://t.co/xCIzcgHsm1
3.7K views · 37 likes
· 2 reposts · 1 replies
Open on X →
We're calling this v0.1 for a reason: we are excited to continue to build the open data ecosystem and hope to train bigger models on more data in the future!
If you know datasets we should include in the next version, open an issue: https://t.co/1NVQJjuDRj
1.7K views · 32 likes
· 2 reposts · 1 replies
Open on X →
Several other groups have put out openly licensed dataset recently, why is ours better? Ablation studies show trained on Common Pile v0.1 out perform them, matching the performance of models trained on the original Pile and OSCAR, though still falling short of FineWeb https://t.c
2K views · 22 likes
· 2 reposts · 1 replies
Open on X →
Our pretrained models, Comma v0.1-1T and -2T perform comparably to leading models trained in the same regime. These plots also include Qwen as a SOTA 8B reference, though it saw 36T tokens https://t.co/eY0W4oO17v
35.8K views · 63 likes
· 1 reposts · 2 replies
Open on X →
We put a lot of work into our metadata, such as having two rounds of manually validating the ToS of websites in Common Crawl, manually identifying trustworthy YouTube channels, and leveraging work by @BigCodeProject and @SWHeritage to build the openly licensed subset of StackV2.
2.3K views · 21 likes
· 1 reposts · 1 replies
Open on X →
What do we mean by "openly licensed" data? Following the lead of orgs like @publicknowledge @Wikimedia @creativecommons we adopt the definition laid out by Public Knowledge: https://t.co/3SCR7sB8I5
Succinctly, it's data that anyone can use, modify, and share for any purpose. htt
3.7K views · 39 likes
· 4 reposts · 2 replies
Open on X →
The Common Pile comprises text from 30 distinct sources, covering a wide variety of domains including research papers, code, books, educational materials, audio transcripts, governmental text, and more. Some of this text is commonplace in AI, but a lot of it is pretty new. https:
3.3K views · 41 likes
· 3 reposts · 2 replies
Open on X →
Can you train a performant language models without using unlicensed text?
We are thrilled to announce the Common Pile v0.1, an 8TB dataset of openly licensed and public domain text. We train 7B models for 1T and 2T tokens and match the performance similar models like LLaMA 1&2 h
182.7K views · 642 likes
· 157 reposts · 22 replies
Open on X →
Im Vergleich zu Konten gleicher Größe
1 Beiträge aus den letzten 90 Tagen, verglichen mit der Größenklasse 10K–100K Follower. wird weit gezeigt, aber nur wenige dieser Zuschauer reagieren.
Medianaufrufe6 796dieses Konto924Median für 10K–100K
Reichweite, %23.65%dieses Konto3.62%Median für 10K–100K
Interaktion, %1.06%dieses Konto1.52%Median für 10K–100K
| Kennzahl | Dieses Konto | Median für 10K–100K | Verhältnis |
|---|---|---|---|
| Medianaufrufe pro Beitrag | 6 796 | 924 | 7.35× |
| Reichweite (Aufrufe ÷ Follower) | 23.65% | 3.62% | 6.53× |
| Interaktionsrate | 1.06% | 1.52% | 0.70× |
Weitere Konten dieser Größe → Mit einem anderen Konto vergleichen → Wie diese Vergleichswerte entstehen →
Growth & engagement
How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.
Views per post
2K6 Jun
1.7K
3.7K
1.2K
1.2K
1.6K
1.5K
6927 Jun
4.6K29 Jul
2.8K
2.9K4 Aug
14.5K10 Aug
2.9K
6.8K22 Aug
Last 14 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.
Engagement rate per post
1.24%6 Jun
2.06%
1.11%
1.79%
1.55%
1.10%
1.16%
0.29%7 Jun
0.61%29 Jul
0.11%
0.48%4 Aug
0.40%10 Aug
0.52%
1.06%22 Aug
Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 1.52%.
What the audience does
Likes63.0%1 117 in total
Reposts11.4%202 in total
Replies2.5%44 in total
Quotes2.5%44 in total
Bookmarks20.6%365 in total
Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.
The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.
Ähnliche Konten
Andrew Wall CPA, CMA ☁️📚 at #AppyCamp22
@WALLCPA
57.5K followers
· 10.3K posts
Abuzar Furqan ✓
@AbuzarFurqan
57.5K followers
· 49.2K posts
LUFCMOT
@LUFCMOTcom
57.5K followers
· 11.8K posts
Jesús Manuel Pérez Triana 🌻 ✓
@jpereztriana
57.5K followers
· 99.7K posts
Ben Boulware
@benboulware7
57.5K followers
· 2.8K posts
Eren ✓
@erenn0189
57.5K followers
· 34.7K posts
jean marc manach
@manhack
57.5K followers
· 65.7K posts
Bussai 🇿🇦🇿🇦🇿🇦
@bussaimathebula
57.5K followers
· 23.6K posts
Frequently asked questions about @AiEleuther
How many followers does @AiEleuther have?
28 740 followers.
How many posts has @AiEleuther published?
951 posts.
How do posts by @AiEleuther perform?
Collected posts: 19. Views: 275 803. Likes: 1 117.