Raphael Pisoni
@4rtemi5.bsky.social
Unsupervised multimodal representation of a learning researcher.
https://www.pisoni.ai
https://www.pisoni.ai
Pinned
After a long hiatus I decided to update my blog and write about some of the things I did over the last few years. Come have a look! pisoni.ai
AI isn't coming for your creativity. It's coming for your lack of diligence.
People talk a lot about #AGI and "super-intelligence," but the immediate disruption is much simpler: AI is killing "vibe-based" decision-making.
People talk a lot about #AGI and "super-intelligence," but the immediate disruption is much simpler: AI is killing "vibe-based" decision-making.
January 31, 2026 at 4:18 PM
AI isn't coming for your creativity. It's coming for your lack of diligence.
People talk a lot about #AGI and "super-intelligence," but the immediate disruption is much simpler: AI is killing "vibe-based" decision-making.
People talk a lot about #AGI and "super-intelligence," but the immediate disruption is much simpler: AI is killing "vibe-based" decision-making.
Over the past year Michał Lewandowski and I published a series of papers on Space Folding , and while Michał went to #AAAI to present the latest one, I worked on a blog-post explaining some the central ideas behind the papers.
Let me know what you think!
www.pisoni.ai/posts/space-...
Let me know what you think!
www.pisoni.ai/posts/space-...
The Shape of Thought: Space Folding in Neural Networks
The mathematical description of deep learning has long been dominated by the language of algebra: matrices, gradients, and optimization landscapes. A parallel and perhaps more intuitive language howev
www.pisoni.ai
January 23, 2026 at 10:05 AM
Over the past year Michał Lewandowski and I published a series of papers on Space Folding , and while Michał went to #AAAI to present the latest one, I worked on a blog-post explaining some the central ideas behind the papers.
Let me know what you think!
www.pisoni.ai/posts/space-...
Let me know what you think!
www.pisoni.ai/posts/space-...
After a long hiatus I decided to update my blog and write about some of the things I did over the last few years. Come have a look! pisoni.ai
January 23, 2026 at 7:00 AM
After a long hiatus I decided to update my blog and write about some of the things I did over the last few years. Come have a look! pisoni.ai
Currently heading to #EurIPS in Copenhagen to present our work on space folding and model interpretability. If you're attending and would like to discuss Representation Learning, SSL, Multimodal LLMs, CV, or other topics that YOU are excited about, feel free to reach out.
December 1, 2025 at 8:54 AM
Currently heading to #EurIPS in Copenhagen to present our work on space folding and model interpretability. If you're attending and would like to discuss Representation Learning, SSL, Multimodal LLMs, CV, or other topics that YOU are excited about, feel free to reach out.
Reposted by Raphael Pisoni
The US government should subsidize Open AI rather than OpenAI
November 7, 2025 at 6:43 AM
The US government should subsidize Open AI rather than OpenAI
Reposted by Raphael Pisoni
On the occasion of the 1000th citation of our Sinkhorn-Knopp self-supervised representation learning paper, I've written a whole post about the history and the key bits of this method that powers the state-of-the-art SSL vision models.
Read it here :): docs.google.com/document/d/1...
Read it here :): docs.google.com/document/d/1...
October 15, 2025 at 10:00 AM
On the occasion of the 1000th citation of our Sinkhorn-Knopp self-supervised representation learning paper, I've written a whole post about the history and the key bits of this method that powers the state-of-the-art SSL vision models.
Read it here :): docs.google.com/document/d/1...
Read it here :): docs.google.com/document/d/1...
We're ready!
might be time
September 21, 2025 at 6:39 AM
We're ready!
The single most undervalued property of neural networks is self-consistency. We should change that!
September 6, 2025 at 12:58 PM
The single most undervalued property of neural networks is self-consistency. We should change that!
Reposted by Raphael Pisoni
July 26, 2025 at 12:51 PM
You and Adam keep beating Sota? Stop doing that! Poor Sota!
July 26, 2025 at 9:50 AM
You and Adam keep beating Sota? Stop doing that! Poor Sota!
Have some cool idea but only evaluate it on small models? Tough luck buddy. You only get your paper accepted if your experimental results are 0.2% above SOTA and too expensive to falsify!
Is academic publishing pay to win yet?
Is academic publishing pay to win yet?
July 26, 2025 at 9:45 AM
Have some cool idea but only evaluate it on small models? Tough luck buddy. You only get your paper accepted if your experimental results are 0.2% above SOTA and too expensive to falsify!
Is academic publishing pay to win yet?
Is academic publishing pay to win yet?
Is there a reason why none of the recent models use RBF-kernel Attention to get rid of the softmax-bottleneck for long context?
I tried replacing dot-product attention with the negative squared KQ-distance and was able to remove the softmax without issues and loss in performance!
I tried replacing dot-product attention with the negative squared KQ-distance and was able to remove the softmax without issues and loss in performance!
July 23, 2025 at 8:14 PM
Is there a reason why none of the recent models use RBF-kernel Attention to get rid of the softmax-bottleneck for long context?
I tried replacing dot-product attention with the negative squared KQ-distance and was able to remove the softmax without issues and loss in performance!
I tried replacing dot-product attention with the negative squared KQ-distance and was able to remove the softmax without issues and loss in performance!
Reposted by Raphael Pisoni
NeurIPS is endorsing EurIPS, an independently-organized meeting which will offer researchers an opportunity to additionally present NeurIPS work in Europe concurrently with NeurIPS.
Read more in our blog post and on the EurIPS website:
blog.neurips.cc/2025/07/16/n...
eurips.cc
Read more in our blog post and on the EurIPS website:
blog.neurips.cc/2025/07/16/n...
eurips.cc
eurips.cc
A NeurIPS-endorsed conference in Europe held in Copenhagen, Denmark
eurips.cc
July 16, 2025 at 10:05 PM
NeurIPS is endorsing EurIPS, an independently-organized meeting which will offer researchers an opportunity to additionally present NeurIPS work in Europe concurrently with NeurIPS.
Read more in our blog post and on the EurIPS website:
blog.neurips.cc/2025/07/16/n...
eurips.cc
Read more in our blog post and on the EurIPS website:
blog.neurips.cc/2025/07/16/n...
eurips.cc
Has anyone experimented with "conditional gradients"?
Thinking about a setup where, within a specific activation range (e.g., right before a ReLU), you'd only permit positive or negative gradients.
Thinking about a setup where, within a specific activation range (e.g., right before a ReLU), you'd only permit positive or negative gradients.
July 8, 2025 at 5:59 AM
Has anyone experimented with "conditional gradients"?
Thinking about a setup where, within a specific activation range (e.g., right before a ReLU), you'd only permit positive or negative gradients.
Thinking about a setup where, within a specific activation range (e.g., right before a ReLU), you'd only permit positive or negative gradients.
Quick question to the SSL experts out there: Usually you evaluate an ssl-model by freezing it and training a linear probing layer. Would it be fair to somehow learn a final layer with more dimensions than classes and do a nearest-neighbor evaluation?
June 29, 2025 at 11:17 AM
Quick question to the SSL experts out there: Usually you evaluate an ssl-model by freezing it and training a linear probing layer. Would it be fair to somehow learn a final layer with more dimensions than classes and do a nearest-neighbor evaluation?
Reposted by Raphael Pisoni
There is an oak forest in central France that was planted 400 years ago by Colbert so that France would have quality hard wood by the 2000s to build ships for its navy.
This is the type of long term planning that Seldonian predictions can help improving.
This is the type of long term planning that Seldonian predictions can help improving.
June 17, 2025 at 8:17 AM
There is an oak forest in central France that was planted 400 years ago by Colbert so that France would have quality hard wood by the 2000s to build ships for its navy.
This is the type of long term planning that Seldonian predictions can help improving.
This is the type of long term planning that Seldonian predictions can help improving.
Reposted by Raphael Pisoni
New anti-censorship jailbreak just dropped ;)
May 13, 2025 at 2:17 AM
New anti-censorship jailbreak just dropped ;)
Currently on my way to #ICLR in Singapore where we'll present our latest paper on space folding in neural networks.
Would be happy to meet some people there so if you're at ICLR as well and want to hang out feel free to pm!🙂
Would be happy to meet some people there so if you're at ICLR as well and want to hang out feel free to pm!🙂
April 18, 2025 at 11:19 AM
Currently on my way to #ICLR in Singapore where we'll present our latest paper on space folding in neural networks.
Would be happy to meet some people there so if you're at ICLR as well and want to hang out feel free to pm!🙂
Would be happy to meet some people there so if you're at ICLR as well and want to hang out feel free to pm!🙂
Grok this! What a roller-coaster of emotions...🤪
April 16, 2025 at 7:01 PM
Grok this! What a roller-coaster of emotions...🤪
Reposted by Raphael Pisoni
ModernBERT or DeBERTaV3?
What's driving performance: architecture or data?
To find out we pretrained ModernBERT on the same dataset as CamemBERTaV2 (a DeBERTaV3 model) to isolate architecture effects.
Here are our findings:
What's driving performance: architecture or data?
To find out we pretrained ModernBERT on the same dataset as CamemBERTaV2 (a DeBERTaV3 model) to isolate architecture effects.
Here are our findings:
April 14, 2025 at 3:41 PM
ModernBERT or DeBERTaV3?
What's driving performance: architecture or data?
To find out we pretrained ModernBERT on the same dataset as CamemBERTaV2 (a DeBERTaV3 model) to isolate architecture effects.
Here are our findings:
What's driving performance: architecture or data?
To find out we pretrained ModernBERT on the same dataset as CamemBERTaV2 (a DeBERTaV3 model) to isolate architecture effects.
Here are our findings:
Reposted by Raphael Pisoni
Just assembled a slide about local feature training time/dataset size.
Anything wrong/missing?
Anything wrong/missing?
April 13, 2025 at 11:20 AM
Just assembled a slide about local feature training time/dataset size.
Anything wrong/missing?
Anything wrong/missing?
Is the project even still worth doing when wandb runs out of funny names or am I cooked?🫠
April 11, 2025 at 11:11 PM
Is the project even still worth doing when wandb runs out of funny names or am I cooked?🫠
Reposted by Raphael Pisoni
Meta introduced Llama 4 models and added this section near the very bottom of the announcement 😬
“[LLMs] historically have leaned left when it comes to debated political and social topics.”
ai.meta.com/blog/llama-4...
“[LLMs] historically have leaned left when it comes to debated political and social topics.”
ai.meta.com/blog/llama-4...
April 5, 2025 at 10:08 PM
Meta introduced Llama 4 models and added this section near the very bottom of the announcement 😬
“[LLMs] historically have leaned left when it comes to debated political and social topics.”
ai.meta.com/blog/llama-4...
“[LLMs] historically have leaned left when it comes to debated political and social topics.”
ai.meta.com/blog/llama-4...