Sylvain Pronovost, PhD
@drsylvainpronovost.bsky.social
Doctor in Cognitive Science working in Artificial Intelligence Engineering.
Reposted by Sylvain Pronovost, PhD
It is interesting that, not only does deep learning suffer from catastrophic forgetting, it also suffers from "plasticity loss", ie performance degrades tthat of a shallow learner under a shifting data distribution (or training against multiple). arxiv.org/abs/2404.00781
Addressing Loss of Plasticity and Catastrophic Forgetting in Continual Learning
Deep representation learning methods struggle with continual learning, suffering from both catastrophic forgetting of useful units and loss of plasticity, often due to rigid and unuseful units. While ...
arxiv.org
August 22, 2025 at 5:45 AM
It is interesting that, not only does deep learning suffer from catastrophic forgetting, it also suffers from "plasticity loss", ie performance degrades tthat of a shallow learner under a shifting data distribution (or training against multiple). arxiv.org/abs/2404.00781
Reposted by Sylvain Pronovost, PhD
The Economist offers an accurate overview of key forces shaping the development of AI today: the risks of the race toward AGI and ASI, the challenges posed by open-sourcing frontier models, the deep uncertainty revealed by ongoing scientific debates and much more ⬇️
www.economist.com/briefing/202...
www.economist.com/briefing/202...
AI labs’ all-or-nothing race leaves no time to fuss about safety
They have ideas about how to restrain wayward models, but worry that doing so will disadvantage them
www.economist.com
July 24, 2025 at 2:14 PM
The Economist offers an accurate overview of key forces shaping the development of AI today: the risks of the race toward AGI and ASI, the challenges posed by open-sourcing frontier models, the deep uncertainty revealed by ongoing scientific debates and much more ⬇️
www.economist.com/briefing/202...
www.economist.com/briefing/202...
Reposted by Sylvain Pronovost, PhD
In my latest (and last!) column for Science’s Expert Voices series, I write about the reasons behind AI chatbots’ “deceptive” behaviors (and why Claude threatened a fictional CEO with blackmail).
www.science.org/doi/10.1126/...
www.science.org/doi/10.1126/...
Why AI chatbots lie to us
A few weeks ago, a colleague of mine needed to collect and format some data from a website, and he asked the latest version of Anthropic’s generative AI system, Claude, for help. Claude cheerfully agr...
www.science.org
July 24, 2025 at 6:24 PM
In my latest (and last!) column for Science’s Expert Voices series, I write about the reasons behind AI chatbots’ “deceptive” behaviors (and why Claude threatened a fictional CEO with blackmail).
www.science.org/doi/10.1126/...
www.science.org/doi/10.1126/...
Reposted by Sylvain Pronovost, PhD
In a stunning moment of self-delusion, the Wall Street Journal headline writers admitted that they don't know how LLM chatbots work.
July 21, 2025 at 1:48 AM
In a stunning moment of self-delusion, the Wall Street Journal headline writers admitted that they don't know how LLM chatbots work.
Reposted by Sylvain Pronovost, PhD
Alison Gopnik, telling it like it is, at Johns Hopkins.
May 29, 2025 at 11:38 PM
Alison Gopnik, telling it like it is, at Johns Hopkins.
Reposted by Sylvain Pronovost, PhD
This is perhaps the most ironic, and meta, example of the dangers of AI slop. www.404media.co/chicago-sun-...
Chicago Sun-Times Prints AI-Generated Summer Reading List With Books That Don't Exist
"I can't believe I missed it because it's so obvious. No excuses," the writer said. "I'm completely embarrassed."
www.404media.co
May 21, 2025 at 10:47 AM
This is perhaps the most ironic, and meta, example of the dangers of AI slop. www.404media.co/chicago-sun-...
Reposted by Sylvain Pronovost, PhD
Zen vibes here in Santa Monica
March 8, 2025 at 6:30 PM
Zen vibes here in Santa Monica
Reposted by Sylvain Pronovost, PhD
A few words on DeepSeek new releases. Links are:
- github.com/deepseek-ai/...
- github.com/deepseek-ai/...
- github.com/deepseek-ai/...
and the Ultra-Scale Playbook at huggingface.co/spaces/nanot...
- github.com/deepseek-ai/...
- github.com/deepseek-ai/...
- github.com/deepseek-ai/...
and the Ultra-Scale Playbook at huggingface.co/spaces/nanot...
February 27, 2025 at 1:41 PM
A few words on DeepSeek new releases. Links are:
- github.com/deepseek-ai/...
- github.com/deepseek-ai/...
- github.com/deepseek-ai/...
and the Ultra-Scale Playbook at huggingface.co/spaces/nanot...
- github.com/deepseek-ai/...
- github.com/deepseek-ai/...
- github.com/deepseek-ai/...
and the Ultra-Scale Playbook at huggingface.co/spaces/nanot...
Reposted by Sylvain Pronovost, PhD
Early signs of deception, cheating & self-preservation in top-performing models in terms of reasoning are extremely worrisome. We don't know how to guarantee AI won't have undesired behavior to reach goals & this must be addressed before deploying powerful autonomous agents.
time.com/7259395/ai-c...
time.com/7259395/ai-c...
When AI Thinks It Will Lose, It Sometimes Cheats
When sensing defeat in a match against a skilled chess bot, advanced models sometimes hack their opponent, a study found.
time.com
February 20, 2025 at 4:45 PM
Early signs of deception, cheating & self-preservation in top-performing models in terms of reasoning are extremely worrisome. We don't know how to guarantee AI won't have undesired behavior to reach goals & this must be addressed before deploying powerful autonomous agents.
time.com/7259395/ai-c...
time.com/7259395/ai-c...
Reposted by Sylvain Pronovost, PhD
"The administration suspended the online education of nearly 1,000 women in Afghanistan who had been studying undetected by the Taliban with funding from the State Department."
The Trump World Order
In the MAGA vision of the national interest, might will make right.
buff.ly
February 20, 2025 at 11:44 PM
"The administration suspended the online education of nearly 1,000 women in Afghanistan who had been studying undetected by the Taliban with funding from the State Department."
Reposted by Sylvain Pronovost, PhD
X-tortion.
Lawyers at Twitter are reportedly threatening advertisers to spend more money on the platform “or else” Elon Musk will use his connections with Trump to hurt their companies.
Exclusive | X Hinted at Possible Deal Trouble in Talks With Ad Giant to Increase Spending
Interpublic Group recently signed an advertising agreement with the social-media platform; other agencies are also in negotiations.
www.wsj.com
February 20, 2025 at 10:07 PM
X-tortion.
Reposted by Sylvain Pronovost, PhD
Reposted by Sylvain Pronovost, PhD
The CDC has an ultra-competitive 2-year program to train "the best of the best" to lead public health labs.
Most of the current class were terminated over the weekend. www.science.org/content/arti... (by @meredithwadman.bsky.social)
Most of the current class were terminated over the weekend. www.science.org/content/arti... (by @meredithwadman.bsky.social)
Ax falls on elite group of Ph.D.s training to lead U.S. public health labs
Most of the fellows in CDC’s highly competitive Laboratory Leadership Service were fired last weekend
www.science.org
February 20, 2025 at 12:23 PM
The CDC has an ultra-competitive 2-year program to train "the best of the best" to lead public health labs.
Most of the current class were terminated over the weekend. www.science.org/content/arti... (by @meredithwadman.bsky.social)
Most of the current class were terminated over the weekend. www.science.org/content/arti... (by @meredithwadman.bsky.social)
Reposted by Sylvain Pronovost, PhD
After 6+ months in the making and over a year of GPU compute, we're excited to release the "Ultra-Scale Playbook": hf.co/spaces/nanot...
A book to learn all about 5D parallelism, ZeRO, CUDA kernels, how/why overlap compute & coms with theory, motivation, interactive plots and 4000+ experiments!
A book to learn all about 5D parallelism, ZeRO, CUDA kernels, how/why overlap compute & coms with theory, motivation, interactive plots and 4000+ experiments!
The Ultra-Scale Playbook - a Hugging Face Space by nanotron
The ultimate guide to training LLM on large GPU Clusters
hf.co
February 19, 2025 at 6:10 PM
After 6+ months in the making and over a year of GPU compute, we're excited to release the "Ultra-Scale Playbook": hf.co/spaces/nanot...
A book to learn all about 5D parallelism, ZeRO, CUDA kernels, how/why overlap compute & coms with theory, motivation, interactive plots and 4000+ experiments!
A book to learn all about 5D parallelism, ZeRO, CUDA kernels, how/why overlap compute & coms with theory, motivation, interactive plots and 4000+ experiments!
Reposted by Sylvain Pronovost, PhD
"working to swiftly rectify the situation..."
www.nbcnews.com/politics/dog...
www.nbcnews.com/politics/dog...
USDA says it accidentally fired officials working on bird flu and is trying to rehire them
In a statement, an Agriculture Department spokesperson told NBC News that officials are "working to swiftly rectify the situation."
www.nbcnews.com
February 18, 2025 at 10:58 PM
"working to swiftly rectify the situation..."
www.nbcnews.com/politics/dog...
www.nbcnews.com/politics/dog...
Reposted by Sylvain Pronovost, PhD
Imagine being rich enough to do anything you want and choosing to take food and medicine from starving children to give yourself even more money.
February 18, 2025 at 11:06 PM
Imagine being rich enough to do anything you want and choosing to take food and medicine from starving children to give yourself even more money.
Reposted by Sylvain Pronovost, PhD
