iiinova — where your digital self keeps evolving

Ananya Harsh Jha @ananyahjha93.bsky.social · 20/05/2026

Using Claude for coding and getting experimental configs out of multiple papers at the same time feels exactly like a product manager's version of getting work done: I think I am making progress in parallel, but I neither understand the code nor the config decisions made by the papers

040

Ananya Harsh Jha @ananyahjha93.bsky.social · 04/03/2026

Honestly, skill issue, ask Claude at this point! PS: idk what the culture is in the bay, but you cannot expect people to respond to emails from AI agents and if the argument is use AI agents as well, again back to point 1: ask Claude at that point, why send an email

020

Reposted by Ananya Harsh Jha

Matthew Finlayson @mattf.nl · 17/10/2025

We discovered that language models leave a natural "signature" on their API outputs that's extremely hard to fake. Here's how it works 🔍 📄 arxiv.org/abs/2510.14086 1/

arxiv.org

Every Language Model Has a Forgery-Resistant Signature

The ubiquity of closed-weight language models with public-facing APIs has generated interest in forensic methods, both for extracting hidden model details (e.g., parameters) and for identifying...

48723

Reposted by Ananya Harsh Jha

Kunal Jha @kjha02.bsky.social · 03/10/2025

Forget modeling every belief and goal! What if we represented people as following simple scripts instead (i.e "cross the crosswalk")? Our new paper shows AI which models others’ minds as Python code 💻 can quickly and accurately predict human behavior! shorturl.at/siUYI%F0%9F%...

33814

Reposted by Ananya Harsh Jha

Stella Li @stellali.bsky.social · 22/07/2025

WHY do you prefer something over another? Reward models treat preference as a black-box😶‍🌫️but human brains🧠decompose decisions into hidden attributes We built the first system to mirror how people really make decisions in our recent COLM paper🎨PrefPalette✨ Why it matters👉🏻🧵

172

Reposted by Ananya Harsh Jha

Abhilasha Ravichander @lasha.bsky.social · 22/07/2025

📣 Life update: Thrilled to announce that I’ll be starting as faculty at the Max Planck Institute for Software Systems this Fall! I’ll be recruiting PhD students in the upcoming cycle, as well as research interns throughout the year: lasharavichander.github.io/contact.html

139212

Reposted by Ananya Harsh Jha

Francis Bach @bachfrancis.bsky.social · 18/07/2025

Tired of lengthy computations to derive scaling laws? This post is made for you: discover the sharpness of the z-transform! francisbach.com/z-transform/

0194

Reposted by Ananya Harsh Jha

Valentina Pyatkin @valentinapy.bsky.social · 03/07/2025

💡Beyond math/code, instruction following with verifiable constraints is suitable to be learned with RLVR. But the set of constraints and verifier functions is limited and most models overfit on IFEval. We introduce IFBench to measure model generalization to unseen constraints.

1295

Reposted by Ananya Harsh Jha

Kunal Jha @kjha02.bsky.social · 19/04/2025

Our new paper (first one of my PhD!) on cooperative AI reveals a surprising insight: Environment Diversity > Partner Diversity. Agents trained in self-play across many environments learn cooperative norms that transfer to humans on novel tasks. shorturl.at/fqsNN%F0%9F%...

1267

Reposted by Ananya Harsh Jha

Gabriel Peyré @gabrielpeyre.bsky.social · 29/01/2025

This review paper by @guillaume-garrigos.com on SGD-related algorithms is a fantastic resource, offering elegant, self-contained, and concise proofs in a single, accessible reference. arxiv.org/pdf/2301.11235

118940

Reposted by Ananya Harsh Jha

Alisa Liu @alisawuffles.bsky.social · 21/03/2025

We created SuperBPE🚀, a *superword* tokenizer that includes tokens spanning multiple words. When pretraining at 8B scale, SuperBPE models consistently outperform the BPE baseline on 30 downstream tasks (+8% MMLU), while also being 27% more efficient at inference time.🧵

Segmentation of the sentence "By the way, I am a fan of the Milky Way" under BPE and SuperBPE.

38316

Reposted by Ananya Harsh Jha

Ai2 @ai2.bsky.social · 13/03/2025

Announcing OLMo 2 32B: the first fully open model to beat GPT 3.5 & GPT-4o mini on a suite of popular, multi-skill benchmarks. Comparable to best open-weight models, but a fraction of training compute. When you have a good recipe, ✨ magical things happen when you scale it up!

35814

Reposted by Ananya Harsh Jha

Hamish Ivison @hamishivi.bsky.social · 04/03/2025

How well do data-selection methods work for instruction-tuning at scale? Turns out, when you look at large, varied data pools, lots of recent methods lag behind simple baselines, and a simple embedding-based method (RDS) does best! More below ⬇️ (1/8)

1134

Ananya Harsh Jha @ananyahjha93.bsky.social · 22/01/2025

After biking between UW and Northgate, I can confirm with irrefutable evidence that Asian parents indeed have walked/biked uphill to their schools in both directions!

160

Ananya Harsh Jha @ananyahjha93.bsky.social · 12/01/2025

Which LLM is supposed to replace software engineers in 2025? I need links... I've been trying to get Sonnet 3.5/o1 on perplexity to visualize optimizers on 1/2 xT A x + bT x + c using Qt, and I can't make any sense of the garbage being spit out right now! Yes I have worked in C++ before!

070

Reposted by Ananya Harsh Jha

Kunal Jha @kjha02.bsky.social · 12/12/2024

Really excited to present my work this Sunday @NeurIPS on how we might approach training a generalist agent capable of cooperation at scale: coordinating with many novel partners on many novel tasks has never been easier! Come by the IMOL workshop to check it out and chat more!

0113

Reposted by Ananya Harsh Jha

Jiacheng Liu @liujch1998.bsky.social · 09/12/2024

Want to predict the task performance of LMs before pretraining them? We develop task scaling laws and model ladders, which predict the accuracy on individual tasks by OLMo 2 7B & 13B models within 2 points of absolute error. The cost is 1% of the compute used to pretrain them.

23314

Reposted by Ananya Harsh Jha

Akari Asai @akariasai.bsky.social · 04/12/2024

I’m on the academic job market this year! I’m completing my @uwcse.bsky.social @uwnlp.bsky.social Ph.D. (2025), focusing on overcoming LLM limitations like hallucinations, by building new LMs. My Ph.D. work focuses on Retrieval-Augmented LMs to create more reliable AI systems 🧵

37117

Reposted by Ananya Harsh Jha

Lj Miranda @ljvmiranda921.itch.io · 04/12/2024

We're releasing the largest Universal Dependencies (UD) treebank for Tagalog, UD-NewsCrawl! This dataset has been a long time coming, but glad to see this through: 15k+ sentences versus the previous ~150 sents from older Tagalog treebanks. 🤗 : huggingface.co/datasets/UD-... 📝 : Paper soon!

huggingface.co

UD-Filipino/UD_Tagalog-NewsCrawl · Datasets at Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

1142

Reposted by Ananya Harsh Jha

Nathan Lambert @natolambert.bsky.social · 03/12/2024

We're hiring another predoctoral researcher for my team at Ai2/OLMo next year. The goal of this position is to mentor and grow future academic stars of NLP/AI over 1-2 years before grad school. This ends up being people done with BS or MS who want to continue to a PhD soon. buff.ly/49nuggo

6547

Reposted by Ananya Harsh Jha

Mechanical Dirk @mechanicaldirk.bsky.social · 02/12/2024

We just updated the OLMo repo at github.com/allenai/OLMo! There are now several training configs that together reproduce the training runs that lead to the final OLMo 2 models. In particular, all the training data is available, tokenized and shuffled exactly as we trained on it!

github.com

GitHub - allenai/OLMo: Modeling, training, eval, and inference code for OLMo

Modeling, training, eval, and inference code for OLMo - allenai/OLMo

05411

Reposted by Ananya Harsh Jha

michael ginn @mginn.bsky.social · 13/11/2024

Hey all! I started a second starter pack with people who didn't make the first one, please let me know if you'd like to be added: go.bsky.app/JgneRQk

706530

Reposted by Ananya Harsh Jha

Marc Marone @marcmarone.com · 23/11/2024

I noticed a lot of starter packs skewed towards faculty/industry, so I made one of just NLP & ML students: go.bsky.app/vju2ux Students do different research, go on the job market, and recruit other students. Ping me and I'll add you!

10117654

Reposted by Ananya Harsh Jha

Allen School @uwcse.bsky.social · 26/11/2024

We created an Allen School Starter Pack to help you find and connect with @uofwa.bsky.social #UWAllen labs and researchers on 🦋! We'll add to this list as we grow our community here: go.bsky.app/RyHBLJd #AcademicBluesky #CompSci #AI #CompBio #UbiComp #Accessibility #NLP #HCI #DataViz #mHealth

go.bsky.app

Allen School Starter Pack

Join the conversation

1287

Reposted by Ananya Harsh Jha

Ai2 @ai2.bsky.social · 26/11/2024

Meet OLMo 2, the best fully open language model to date, including a family of 7B and 13B models trained up to 5T tokens. OLMo 2 outperforms other fully open models and competes with open-weight models like Llama 3.1 8B — As always, we released our data, code, recipes and more 🎁

The OLMo 2 models sit at the Pareto frontier of training FLOPs vs model average performance.

515235