Neehar Kondapaneni
@therealpaneni.bsky.social
Researching interpretability and alignment in computer vision.
PhD student @ Vision Lab Caltech
PhD student @ Vision Lab Caltech
Pinned
Excited to share our paper Representational Difference Explanations (RDX) was accepted to #NeurIPS2025! 🎉RDX is a new method for model diffing designed to isolate 🔍 representational differences. 1/7
Reposted by Neehar Kondapaneni
the way this is going the only trustworthy information will be from in-person interviews.
for annotation projects i find it much easier to trust small numbers of expert or local students annotators, compared to broad-audience online recruiting
for annotation projects i find it much easier to trust small numbers of expert or local students annotators, compared to broad-audience online recruiting
new paper by Sean Westwood:
With current technology, it is impossible to tell whether survey respondents are real or bots. Among other things, makes it easy for bad actors to manipulate outcomes. No good news here for the future of online-based survey research
With current technology, it is impossible to tell whether survey respondents are real or bots. Among other things, makes it easy for bad actors to manipulate outcomes. No good news here for the future of online-based survey research
November 18, 2025 at 8:27 PM
the way this is going the only trustworthy information will be from in-person interviews.
for annotation projects i find it much easier to trust small numbers of expert or local students annotators, compared to broad-audience online recruiting
for annotation projects i find it much easier to trust small numbers of expert or local students annotators, compared to broad-audience online recruiting
Reposted by Neehar Kondapaneni
ChatGPT’s signature writing style is everywhere now, and I hate it. It reminds me of when we tried mixing all the beverages at the soda fountain in middle school. We didn’t actually create the perfect drink, we just made a cloying monstrosity that lost everything good about its constituent parts.
November 19, 2025 at 5:17 PM
ChatGPT’s signature writing style is everywhere now, and I hate it. It reminds me of when we tried mixing all the beverages at the soda fountain in middle school. We didn’t actually create the perfect drink, we just made a cloying monstrosity that lost everything good about its constituent parts.
Excited to share our paper Representational Difference Explanations (RDX) was accepted to #NeurIPS2025! 🎉RDX is a new method for model diffing designed to isolate 🔍 representational differences. 1/7
November 19, 2025 at 4:50 PM
Excited to share our paper Representational Difference Explanations (RDX) was accepted to #NeurIPS2025! 🎉RDX is a new method for model diffing designed to isolate 🔍 representational differences. 1/7
You’ve generated 10k concepts with your favorite XAI method -- now what? Many concepts you’ve found are fairly obvious and uninteresting. What if you could 𝑠𝑢𝑏𝑡𝑟𝑎𝑐𝑡 obvious concepts away and focus on the more complex ones? We tackle this in our latest preprint!
July 8, 2025 at 3:43 PM
You’ve generated 10k concepts with your favorite XAI method -- now what? Many concepts you’ve found are fairly obvious and uninteresting. What if you could 𝑠𝑢𝑏𝑡𝑟𝑎𝑐𝑡 obvious concepts away and focus on the more complex ones? We tackle this in our latest preprint!
Have you ever wondered what makes two models different?
We all know the ViT-Large performs better than the Resnet-50, but what visual concepts drive this difference? Our new ICLR 2025 paper addresses this question! nkondapa.github.io/rsvc-page/
We all know the ViT-Large performs better than the Resnet-50, but what visual concepts drive this difference? Our new ICLR 2025 paper addresses this question! nkondapa.github.io/rsvc-page/
April 11, 2025 at 4:11 PM
Have you ever wondered what makes two models different?
We all know the ViT-Large performs better than the Resnet-50, but what visual concepts drive this difference? Our new ICLR 2025 paper addresses this question! nkondapa.github.io/rsvc-page/
We all know the ViT-Large performs better than the Resnet-50, but what visual concepts drive this difference? Our new ICLR 2025 paper addresses this question! nkondapa.github.io/rsvc-page/
Reposted by Neehar Kondapaneni
Our new piece in Nature Machine Intelligence: LLMs are replacing human participants, but can they simulate diverse respondents? Surveys use representative sampling for a reason, and our work shows how LLM training prevents accurate simulation of different human identities.
February 17, 2025 at 4:36 PM
Our new piece in Nature Machine Intelligence: LLMs are replacing human participants, but can they simulate diverse respondents? Surveys use representative sampling for a reason, and our work shows how LLM training prevents accurate simulation of different human identities.