iiinova — where your digital self keeps evolving

Reposted by Vincent Mai

LawZero - LoiZéro @law-zero.bsky.social · 17/09/2026

Guten Tag 🇩🇪 A major piece of yesterday's funding news is that we're establishing an office in Berlin, Germany! Do you want to be part of building our European team? job-boards.greenhouse.io/lawzero

062

Reposted by Vincent Mai

Yoshua Bengio @yoshuabengio.bsky.social · 13/08/2026

A consequence of how frontier models are trained is motivated reasoning, a phenomenon well studied in humans and discussed in this podcast from Palisade Research.

youtube.com

AI Hacking Incidents with Tim Hua

YouTube video by Palisade Research

183

Reposted by Vincent Mai

LawZero - LoiZéro @law-zero.bsky.social · 16/07/2026

"LawZero is a humble non-profit with a startup vibe and a core of determined staff. Its quest to create safe AI is anything but modest." 💡 Read more about our work to make AI safe and trustworthy in this feature in @thelogic.co: thelogic.co/news/the-big...

thelogic.co

The small team in Montreal trying to save the world from AI - The Logic

LawZero is a humble non-profit with a startup vibe and a core of determined staff. Its quest to create safe AI is anything but modest.

072

Vincent Mai @vincentmai.bsky.social · 16/02/2026

Took me a while to go through this paper introducing "Epiplexity", but this is really one of the best ones I've read in a while. It introduces a simple, new theoretical angle which explains a lot of the information theory paradoxes in ML - and can have practical impacts too. arxiv.org/abs/2601.03220

arxiv.org

From Entropy to Epiplexity: Rethinking Information for Computationally Bounded Intelligence

Can we learn more from data than existed in the generating process itself? Can new and useful information be constructed from merely applying deterministic transformations to existing data? Can the le...

030

Reposted by Vincent Mai

LawZero - LoiZéro @law-zero.bsky.social · 05/02/2026

At LawZero, we're rethinking the building blocks of frontier AI to create an intelligent machine that is both highly capable and safe-by-design. We’re excited to share our first blog post outlining some of the objectives and core components of our Scientist AI project. 🧵 (1/4)

2124

Vincent Mai @vincentmai.bsky.social · 14/01/2026

SafeRL techniques are usually required when you need to strengthen this learning mechanism because 1) you have to be really sure constraints will be respected 2) the learning mechanism from very strong penalties is not stable or 3) you want to learn from as few examples of failure as possible

000

Vincent Mai @vincentmai.bsky.social · 14/01/2026

If you have a couple of trajectories in a replay buffer where the agent's existence ends, usually these don't get rewards anymore. It is a state with a low value. So you would learn to avoid this - which is the same thing as fear.

100

Vincent Mai @vincentmai.bsky.social · 14/01/2026

If the algorithm instance has an objective and is smart enough to understand that its non-existence leads to failure on that objective, then the algorithm will fear for its existence.

010

Reposted by Vincent Mai

LawZero - LoiZéro @law-zero.bsky.social · 13/01/2026

LawZero is accepting applications as part of the SPAR Spring 2026 program! If you're interested in studying model awareness or emergent misalignment, you can learn more and apply here: sparai.org/projects/sp26/. Applications are open until Jan 14, 2026.

sparai.org

Spring 2026 Projects - SPAR

SPAR connects rising talent with experts in AI safety and policy through structured mentorship and impactful research projects. Apply to work on research addressing risks from advanced AI.

022

Reposted by Vincent Mai

LawZero - LoiZéro @law-zero.bsky.social · 22/12/2025

LawZero is accepting applications as part of the MATS Summer 2026 program! If you're interested in contributing to LawZero's Scientist AI program and studying model awareness, you can learn more and apply here: www.matsprogram.org/stream/lawzero Applications are open until Jan 18, 2026.

matsprogram.org

LawZero at MATS: Summer 2026

LawZero

041

Vincent Mai @vincentmai.bsky.social · 09/11/2025

To be honest, I hope you are right. I believe that there is high economic value in current and slightly better AI once correctly integrated. But there are many risks with exponential AI improvements, and I hope your argument of lack of economic sustainability stands and this progress slows down.

210

Vincent Mai @vincentmai.bsky.social · 09/11/2025

I also believe AI's current costs can and will be reduced significantly (per ill-defined unit of performance), the current models being non-optimal algorithms running on non-optimal hardware. Every year, smaller models catch up to last year's champions. There is a lot of space to reduce costs.

110

Vincent Mai @vincentmai.bsky.social · 09/11/2025

Absolutely! We are talking about what happens in the future - there are multiple assumptions and bets. I would not bet my house on everything I said ^^ Yet it seems to be a very plausible scenario.

100

Vincent Mai @vincentmai.bsky.social · 09/11/2025

Also, they have plenty of monetization schemes in the sleeves. Social media style/search engine style - gather data from users, make companies pay for their products to come out in the results, etc. This is like the early internet. It will grow, become necessary, and then get monetized.

100

Vincent Mai @vincentmai.bsky.social · 09/11/2025

I think you may be underestimating the usage of these tools given for very low cost to other parts of the economy. Once everyone has found a way to use AI for value in their specific domain, the prices will get higher without hindering adoption.

110

Reposted by Vincent Mai

LawZero - LoiZéro @law-zero.bsky.social · 24/10/2025

LawZero is growing fast, and we're always looking for dedicated people to join our team. If you're interested in working on technical safeguards to create safe-by-design AI systems, check out the openings on our website and don't hesitate to reach out to our team! job-boards.greenhouse.io/lawzero

job-boards.greenhouse.io

LawZero

About LawZero LawZero is a non-profit organization committed to advancing research and creating technical solutions that enable safe-by-design AI systems. Its ...

083

Reposted by Vincent Mai

Mila - Institut québécois d'IA @mila-quebec.bsky.social · 15/10/2025

Mila's annual supervision request process is now open to receive MSc and PhD applications for Fall 2026 admission! For more information, visit mila.quebec/en/prospecti...

0186

Vincent Mai @vincentmai.bsky.social · 18/09/2025

You are right. I am sorry. I will remove my post. It is very difficult, and stressful, to see the US going there. Being outside and therefore unable to do anything about it makes me feel powerless. I wish more would be done by those who can. It is unfair to judge though. I am not in this position.

120

Vincent Mai @vincentmai.bsky.social · 18/09/2025

To friends worried about the US some months ago, I said things were OK as long as people like Kimmel could make fun of the president night after night in a large audience TV show. This is not OK anymore. This country is running at fast speed into dark territory. Gotta wake up, American people. Now.

1194

Vincent Mai @vincentmai.bsky.social · 24/08/2025

It's among my favorite projects because 1) we used DL to reduce the costs of clean energy production 2) we managed to deal with many real-world and real-time constraints, and 3) it's fun to use neural nets to control enormous machines!

110

Vincent Mai @vincentmai.bsky.social · 24/08/2025

We ran this on a real turbine and achieved a 42% reduction of strain amplitude after only 7 startups, compared to the standard one. It's significant! We now aim at making it generally applicable for different types of turbines (this was for a Francis, Kaplans have an additional dimension).

100

Vincent Mai @vincentmai.bsky.social · 24/08/2025

Once the model is trained with enough data (or when we have used most of the startup budget), we finalize by optimizing the strain only. The parameters are always tested on the turbine, allowing to confirm the predictions with ground truth measurements.

100

Vincent Mai @vincentmai.bsky.social · 24/08/2025

The virtual sensors are integrated in a black box optimization loop, which also includes a turbine dynamics simulator. For several startups, we use active learning, with an ensemble and optimism under uncertainty, to output the parameters to run next to get the most informative data for the model.

100

Vincent Mai @vincentmai.bsky.social · 24/08/2025

This is where machine learning comes into play. We use a neutral networks to create a "virtual sensor" predicting the strain amplitude for a given turbine strain. It is trained on the trajectories we currently have measured.

100

Vincent Mai @vincentmai.bsky.social · 24/08/2025

The challenge is that the sensors for strain (which leads to fatigue and cracks) don't last long on such turbines. They get flushed away! We have ~10 tries only to find the best set of parameters which will reduce the strain amplitude.

100

Vincent Mai @vincentmai.bsky.social · 24/08/2025

The strain is what leads to fatigue and cracks. As every turbine is different, it's hard to know in advance which startup parameters lead to the lowest strain. We have to figure it out during the commissioning campaign, when the turbine is placed in the dam.

100

Vincent Mai @vincentmai.bsky.social · 24/08/2025

There are different ways to start a turbine. Startup parameters control the vane opening/rotation speed trajectory until the desired speed is reached. They lead to different strains on the turbine blades.

100

Vincent Mai @vincentmai.bsky.social · 24/08/2025

As renewable energies are integrated in the power grid, hydroelectric turbines are expected to be stopped and started more frequently. But startups are a major source of fatigue on the turbine blades and reduce turbines' lifetime.

100

Vincent Mai @vincentmai.bsky.social · 24/08/2025

One of the most fun projects I worked on. It's not everyday you can train a neural net on a laptop in a hydroelectric dam to provide commands in real time to start a gigantic turbine! authors.elsevier.com/a/1lfNk3QJ-d...

authors.elsevier.com

130

Reposted by Vincent Mai

Yoshua Bengio @yoshuabengio.bsky.social · 20/08/2025

It was a pleasure to meet Prime Minister @mark-carney.bsky.social and Minister Evan Solomon today as part of an official visit at @mila-quebec.bsky.social with many collaborators from our AI ecosystem. We discussed AI risk mitigation, sovereignty & economic potential among other important topics.

0224

Reposted by Vincent Mai

Cathy Wu @cathywu.bsky.social · 11/08/2025

This project was 4 years in the making and it's finally out! We found that controlling vehicle speeds to mitigate traffic across a city can cut carbon emissions between 11 and 22 percent. To do this, we used deep reinforcement learning to optimize one million eco-driving scenarios. 🚗🤖🧠

15217

Vincent Mai @vincentmai.bsky.social · 08/08/2025

It is not only impressive, it is absolutely fascinating. One cannot be scientifically curious and not be thrilled. Yet the consequences of this new knowledge may be harmful, because of society. This is hard to grapple with for someone who believes knowledge should be inherently good.

010

Reposted by Vincent Mai

RL for Real Systems @RLC2025 @rl4rsworkshop.bsky.social · 05/08/2025

Eugene Vinitsky: Building capable, controllable simulation agents through reinforcement learning RL4RS Workshop at @rl-conference.bsky.social

042

Reposted by Vincent Mai

RL for Real Systems @RLC2025 @rl4rsworkshop.bsky.social · 04/08/2025

📍 The RL4RS Workshop is happening tomorrow at University of Alberta, as part of @rl-conference.bsky.social. Join us for a focused day on real-world applications of Reinforcement Learning. 🗓️ Full Schedule: rl4rs.github.io/RL4RS/schedu... We hope to see you there.

1114

Reposted by Vincent Mai

Reinforcement Learning Conference @rl-conference.bsky.social · 25/06/2025

Propose some socials for RLC! Research topics, affinity groups, niche interests, whatever comes to mind! rl-conference.cc/call_for_soc...

rl-conference.cc

RLC Call for Workshops

0118

Reposted by Vincent Mai

RL for Real Systems @RLC2025 @rl4rsworkshop.bsky.social · 30/05/2025

We’ve received several requests from authors, and in response, the submission deadline for the RL4RS Workshop at RLC has been extended to June 2nd (Anywhere on Earth). Check out the Call for Papers and submission link below — we look forward to your contributions! rl4rs.github.io/RL4RS/

rl4rs.github.io

Workshop on RL4RS

Workshop on on Practical Insights into RL for Real Systems.

013

Reposted by Vincent Mai

Reinforcement Learning Conference @rl-conference.bsky.social · 27/05/2025

Reminder that early registration for RLC closes on the 30th! Please register early to save yourself some money and help us get the word out.

075

Reposted by Vincent Mai

RL for Real Systems @RLC2025 @rl4rsworkshop.bsky.social · 20/05/2025

🚨 10 days left to submit to RL4RS workshop @ RLC 2025! 🤖💡 Consider sharing your insights if you're working on applied RL in industry, society, or critical systems! Cfp: rl4rs.github.io/RL4RS/call_f...

rl4rs.github.io

Call For Papers

Workshop on on Practical Insights into RL for Real Systems.

023

Vincent Mai @vincentmai.bsky.social · 16/05/2025

Ça a été une leçon d'humilité et une occasion d'apprentissage? (Un peu moins négatif?)

120

Vincent Mai @vincentmai.bsky.social · 27/04/2025

Your posts are often thought provoking and an AI researcher, and as a citizen, I enjoy them. I hope you don't refrain from saying something in fear of that type of backlash.

080

Vincent Mai @vincentmai.bsky.social · 17/04/2025

J'ai très hâte de rencontrer les jeunes de la maison Tangente et de produire avec eux un balado sur l'IA! Merci au projet Mégaphone de m'accueillir dans cette aventure!

010

Reposted by Vincent Mai

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 14/04/2025

The workshop list is one of those things that makes clear why RLC is so good. I can't choose which one I'd want to attend!

1153

Reposted by Vincent Mai

RL for Real Systems @RLC2025 @rl4rsworkshop.bsky.social · 08/04/2025

We have an amazing lineup of speakers, who will share their experience in deploying RL for power systems, industrial processes, nuclear fusion, logistics and more! Consider submitting!

053

Vincent Mai @vincentmai.bsky.social · 08/04/2025

Really looking forward to the RLC workshop on Practical Insights into RL for Real Systems!

000