Rémi Debette

MLOps manager at @instadeepai / @BioNTech_Group,

Bass
Interested in foreign cultures, esp. Japan

The opinions in my tweets are mine only and not my employer's

[Rémi Debette @RmiDebette]

The band
https://t.co/BR4EEDfcSS

[atomic.chat @atomic_chat_hq]

Mistral OCR 4 turned a handwritten calculus exam into clean LaTeX!

We gave it a photo of a hand-written exam page. The model read the handwriting and rebuilt every formula into structured digital text

Output: Time: 5.1s · Cost: $0.09

Formulas came through exactly right - the hard part was nailed. The graph, unfortunately, it didn’t redraw. But that’s the telling part: most OCR tools just dump the text and quietly drop the figure. OCR 4 caught the plot, boxed it, and tagged it as a chart. It doesn’t get redrawn, but it gets read and accounted for

[Rémi Debette @RmiDebette]

This has been one of the best tools I have used this year

[Justin Poehnelt @JPoehnelt]

Two months ago I was fired by Google for creating the Google Workspace CLI. It went viral, hit #1 on Hacker News, gained thousands of GitHub stars and many thousands of actual users in just a couple days.

It was an incredible, confusing journey, from directors and leaders asking what they could learn from the tool to getting grilled by legal about why the Google logo and brand colors are on the Google Workspace GitHub code repositories.

I think the cause was that Workspace and certain leaders (and projects) were afraid of being disrupted. But the fear wasn't specific to my CLI, it was a broader fear in what agents meant for Workspace. Either way, the irony of my termination was the announcement at Google Cloud Next two days before I was fired that an official Workspace CLI was coming.

I want this out there because it is easier for me to explain my story and it is an experience I want to fully own. It's also part of my healing.

Nearly 7 years at Google was an incredible opportunity for me and I was fortunate to have wonderful teammates and a manager that fully supported me through these last few months. Thank you.

[DeepPCB @DeepPCB]

🤖 Cooper is live: your layout expert, on demand.

It checks your board before @DeepPCB routes it.

Upload a board. Cooper tells you what’s wrong before you route, fixes it once you confirm, then drives the engine.

96% mean completion, DRC-clean.

It flags it. It fixes it. Done.

Try it now: https://t.co/cV9g6pWi2a

#ElectronicsEngineering #PCBDesign #PCBLayout #AIAgent #DeepPCB #Hardware

[InstaDeep @instadeepai]

NTv3 is the latest addition to the Nucleotide Transformer family and our unified DNA foundation model for long-context genome understanding and design. 🧬

[InstaDeep @instadeepai]

🚀 Introducing MLIPAudit, InstaDeep’s benchmark suite for Machine Learning Interatomic Potentials (MLIPs).

Trained on vast amounts of quantum data, MLIPs deliver near-quantum accuracy at classical speeds. But with so many models emerging, knowing which to trust is challenging…

[GrapheneOS @GrapheneOS]

Here's another French journalist participating in fearmongering about GrapheneOS. That article is not measured. It provided a platform to make both unsubstantiated and provably false claims about GrapheneOS while providing no opportunity to see and respond to those claims.

https://t.co/gGEiBwAbZt

The claims the article platforms are conflating closed source products from European companies infringing on our copyright and trademarks with GrapheneOS. GrapheneOS doesn't have the features they claim it does, isn't distributed in the ways they claim and they don't understand open source software.

GrapheneOS is obtained from https://t.co/21l8TYXrws and https://t.co/6FssqXFT1L. There are a bunch of legitimate companies in Europe selling devices with real GrapheneOS including NitroKey. We aren't partnered with those companies and don't get funding from it but there's nothing shady about it.

Products using operating systems partially based on our code are not GrapheneOS. There's no such thing as a fake Snapchat app wiping the device in GrapheneOS. It has no remote management or remote wiping built into it. It does not have a subscription fee / licensing system built into it either.

Vast majority of the code for those products comes from elsewhere: Android Open Source Project, Linux kernel, Chromium, LLVM and other projects. Of course the non-profit open source project writing a small portion of the code being used by those companies being targeted rather than IBM, Google, etc.

Both Android and iOS try to defend users from the same attack vectors we do. We developed far better protections against exploits which we release as open source code. Open source means anyone can freely use it for any purpose, exactly like the Android Open Source Project used by GrapheneOS itself.

Open source is why we can build GrapheneOS based on the Android Open Source Project. It doesn't make Linus Torvalds, IBM, Google, etc. responsible for what we do. Similarly, others can make their own software based on GrapheneOS. A fork of GrapheneOS contains a small portion of code written by us.

France supposedly has a right to reply which we intend to exercise to respond at length to these articles containing libel from the French state.

We're going to be ending the small amount of operations we have in France as we don't feel the country is safe for open source privacy projects anymore.

GrapheneOS doesn't host services storing sensitive user data. We have signature verification and downgrade protection for updates to the OS, apps and app store metadata. We're going move our website and discussion server away from OVH. Our update mirrors and authoritative DNS are already elsewhere.

Our discussion forum, Matrix, Mastodon, etc. in OVH Bearharnois can be moved to local or colocated servers in Toronto instead. We can use Netcup (owned by Anexia, both German) as one of the main providers for website/network service instances. The majority of our servers are already not on OVH.

We won't travel to France including avoiding conferences and will avoid having people working in the country too. A simple heuristic for the EU is avoiding countries supporting Chat Control. We genuinely believe we cannot safely operate in France anymore as an open source project privacy project.

Our pinned post on this platform shows a great example of why they're actually upset with us:

https://t.co/aJFYRQqtZG

It almost makes us willing to contribute to AOSP again to try to wipe out their ability to exploit a subset of unon-GrapheneOS Android devices too. Google is welcome to reach out.

[InstaDeep @instadeepai]

🚀 Proud to share that our research showing how inference-time strategies can break the reinforcement learning performance ceiling has been selected for an oral presentation at @NeurIPSConf 2025! 🎉

👇 Hear from lead authors @ChalumeauFelix and @ruanjohn in the ▶️ below.

🧵⬇️

[InstaDeep @instadeepai]

We are heading to @NeurIPSConf 2025! 🎉

Our Africa-based Reinforcement Learning team are making headlines with:

3️⃣ Accepted papers
✨ 1 Spotlight (top 3%)
🎤 And our first-ever Oral presentation at NeurIPS (top 0.3%)!

🧵⬇️

[InstaDeep @instadeepai]

As seen in @NatureMethods, meet SegmentNT, our transformer model for fast, single-nucleotide, multi-task genome annotation. 🧬

[DeepPCB @DeepPCB]

Manual 𝐏𝐂𝐁 𝐝𝐞𝐬𝐢𝐠𝐧 can’t keep up with today’s complexity. ✨ 𝐀𝐈 𝐜𝐚𝐧.
👉 Discover how @DeepPCB uses reinforcement learning to deliver DRC-clean layouts in hours in our new White Paper: link in comment!
#PCBDesign #AIinEngineering #ReinforcementLearning #InstaDeep⁣

[InstaDeep @instadeepai]

We’re back with the BFN story in Episode 2 of the Let’s Talk Research podcast.

Hear Alex Graves dive into real-world applications of Bayesian Flow Networks, including their use in protein sequencing and antibody design. 🧵⬇️

[Rémi Debette @RmiDebette]

Have a look at @instadeepai and @BioNTech_Group latest AI-enhanced healthcare advancements shared on our AI day!

[InstaDeep @instadeepai]

At this year’s AI Day, we came together with our colleagues from @BioNTech_Group to showcase how we are advancing the frontier of Bio AI, with a look ahead at what the future has to hold in our rapidly accelerating field 🌠

Learn the highlights below: 🧵⬇️

[InstaDeep @instadeepai]

Hear Alex Graves tell the story behind Bayesian Flow Networks, over a decade in the making. 🚀

In Episode 1 of our new Let’s Talk Research podcast, Alex shares the inspiration behind BFNs and how he sees them evolving. 🧵

[derek guy @dieworkwear]

I debated whether to share my story on here, but I guess I will.

I think there's an idea out there that millions of violent criminals are pouring across the border, carrying machetes and drugs, looking to harm Americans. Certainly, while some people fall into that category, the vast majority of undocumented immigrants don't.

My family escaped Vietnam after the Tet Offensive and went through an arduous journey that eventually landed them in the Canada. My father worked there for a time as a janitor; my mother, a secretary. When work fell through, my dad was offered to work with his sister in the United States, so he went, as our family needed money. He ended up staying in the US longer than he was supposed to — not knowing immigration laws — and asked my mom to come be with him. Of course, she went and carried me over the border while I was still a baby.

I'm still unsure whether we technically broke an immigration law. The border between Canada and the United States was pretty porous (as it is today, for the most part). But either way, since I came here without legal documentation, I eventually fell into the category of being an undocumented immigrant. Yet, I've been in the United States since I was a baby. My identity and roots are very much based in this country, no different from anyone else.

The lack of legal immigration has totally shaped my life. It has made every interaction with the law much scarier. It has shaped which opportunities I could or could not get. It has taken an emotional toll, as this legal issue hangs over your head like a black cloud.

There are millions of people in my situation — people who were brought here as children, some babies, and were unable to raise their hand during this process and say "maybe we should hire a lawyer." Parents may have crossed the border not knowing about the law, as the law can be pretty confusing and lawyers are expensive. Perhaps, in the end, they should have hired a lawyer; but sometimes life is messy. In the end, they crossed for the same reason many people rise every day: to support their families.

I think it's unreasonable to deport millions of people who have contributed positively to society. It's simply not true that the majority of these people are violent criminals carrying drugs. Many own small businesses; many work as skilled laborers (including garment workers!). The vast majority are good, honest people. For those of us who grew up in the United States, sending us "back" to our "home" country would mean sending someone to a foreign land, as they grew up in the United States. Tearing families apart based on immigration status seems inhumane.

About twenty years ago, I watched with some small measure of hope that the Dream ACT would pass and help a portion of undocumented immigrants. I saw people march angrily in the streets (something I supported). But I was dismayed to see nothing really come about. In fact, when Obama passed DACA, I was skeptical that it would be sensible to give your information to the federal government without a guaranteed path to citizenship. It seems that the Trump administration may use those lists to hunt down immigrants.

I think the ICE sweeps are inhumane. I support and admire the protestors who are putting their bodies on the line for non-violent resistance. Ultimately, I think we need to solve this issue on a systemic level. It's unreasonable to me to expect that the government will deport some 10-20 million people. Even deporting 1 million will cause an insane amount of chaos, not to mention an incredible amount of wasteful government spending. The militarization of law enforcement is over the top and only escalate the situation.

I have seen people march on this issue, only for it to be dropped later and dissolve into quiet grumbles. In the end, nothing is done for immigrants. Time passes until decades later, we have another political crisis like the one we have today. This cycle seems unsustainable. I hope people sustain this energy to solve the immigration issue and put undocumented immigrants — most who are peaceful, skilled, and hard-working citizens — on a path to citizenship. As time continues to go by, many undocumented immigrants are seeing their lives wasted away.

If you care about this issue, I hope you get involved on the long haul: volunteer for pro-immigrant groups, donate to mutual aid, and perhaps pursue career paths that help people in this situation. I was moved to read someone on here say that, when they found their employee was undocumented, they helped them get a green card. The path was difficult, as our immigration system is broken (it took them 11 years). But I can tell you that this is like pulling someone drowning in raging waters.

Ultimately, I hope me sharing this story helps push back against the idea that all undocumented immigrants are MS-13 members. I know many people in my position and they are all like your neighbors.

[DeepPCB @DeepPCB]

🆕 Just dropped: @InstaDeepAI's Alain-Sam Cohen on how DeepPCB uses RL to transform PCB layout featured in Emerj’s AI in Business podcast with Editorial Director Matthew DeMello
✅ Week-long designs done in a day
✅ Engineers freed from repetitive tasks
✅ Real-world enterprise results
🎧 Listen via @Emerj: https://t.co/ICSPvc9l9x

[InstaDeep @instadeepai]

🔧 Introducing AbBFN2: our multi-modal antibody foundation model.

AbBFN2 jointly models 45 data modes spanning sequences, genetic information and developability attributes to provide a rich framework with which to define conditional generation tasks.

Join Research Scientist Bora Guloglu as they explore AbBFN2's integrated approach to antibody design— using a unified generative model to potentially enhance scientific efficiency.

[InstaDeep @instadeepai]

👩‍💻Hear from the women making history in AI today.

On #WomensHistoryMonth, join women from the InstaDeep team around the world as they share how they're building, coding, researching and connecting to create a better, fairer world through AI innovation.

[DeepPCB @DeepPCB]

Quick technical outline for anyone interested in DeepPCB 🚀 (but we’re not revealing too much 👀)
https://t.co/FcGcdYKKDf

[G Milgram @GGmilgram]

🔥Véronique Louwagie,députée LR, a posé à 2 reprises, en 2018 et en 2024 (!), une question au Gouvernement au sujet des géobiologues. Sa demande ? Madame la députée souhaite une « reconnaissance officielle » par l'État, par l'Éducation Nationale, de la formation et du diplôme..⬇