Cléo Poulin 🇨🇦

Science, space, EVs, music, whitewater, ecommerce.

[Lucy Hargreaves @lucyhargreaves4]

I emigrated to Canada in the summer of 1987. I was 8 years old and had only ever visited once before.

My idea of what Canada was came from a brief trip a few years earlier. My family had visited relatives in London, Ontario, went to Ontario Place, did an Easter egg hunt at the Toronto Zoo, and went up the CN Tower – then less than a decade old and a total marvel. It still is.

Beyond Toronto, I thought Canada was a nation of vast mountain ranges, wilderness, bears, and lumberjacks. While slightly silly in retrospect, I felt full of anticipation at the potential adventure and excitement this country could offer.

There was a rail strike that summer. Our furniture arrived a month after we moved in to our new house. Our next-door neighbours in the suburbs of Toronto (where a house within walking distance of Lake Ontario was still affordable to a junior professor and a teacher) rallied around and gave us their spare camping mattresses, linens, a card table and chairs, and dishware.

I became fast friends with the family's daughter, a girl my age. We're still close. She introduced me to ketchup chips, Coffee Crisp, dill pickles, and bannock. We roamed the neighbourhood parks and made countless trips to the nearby 7-Eleven for 5 cent candy.

By the fall of 1995, the Quebec referendum was the nation's focus. The vote came just a couple of weeks after our Canadian citizenship ceremony. We were so proud to become Canadians. Even my grandmother made the trip over from the UK to share in the celebration. Our neighbours threw us a citizenship party and made us t-shirts that read "My Canada Includes The Hargreaves Family", a play on the "My Canada Includes Quebec" signs from the unity rallies.

The t-shirt is now long gone, lost in the many life shuffles since. But the sentiment holds. I think a lot about the Canada that took us in, the one that showed up with mattresses for a family it barely knew, and about what that same country could still become.

The Canada I landed in was generous before it was anything else. I want it to stay generous. But it can't be generous if it has nothing left to give. Canada must get as ambitious for itself as it's always been for the people it takes in. A country that dreams and builds the way it welcomes: without hesitating, without apologizing.

Somewhere this summer, an 8-year-old is getting off a plane with their family, knowing no one. Another is out on her bike past dark, on a street she's known her whole life -- my own kids among them. I want the Canada they all grow up in to be ambitious, dream big, champion risk takers, and celebrate success.

That Canada truly will have the potential to be the most prosperous country in the world. I know we can do it.

Happy Canada Day! 🇨🇦

[Jimmy Heaters @CathPoaster]

a big ugly data center will be built next to my small town that will provide the necessary compute that researchers use to eradicate a terrible disease. hundreds of thousands of lives will be saved thanks to autonomous vehicles but you might know somebody who becomes paralyzed in a self driving car crash. a frontier model gets released that brings us new math and science yet it induces mania in a small group of people. the economy is more dynamic than ever but your best friend is depressed because his job became obsolete. jagged, imperfect progress.

[shira @shiraeis]

This is actually a much more interesting result than the inevitable “meta can read your mind and will use this for advertising” takes will imply.

One caveat first is the MEG setup is still very much a large neuroscience lab apparatus, not so much something that enables your RayBans to extract your inner monologue.

The actual technical result is still cool tho! The model is trained on healthy skilled typists using MEG while they produce sentences.

Their v1 showed non-invasive MEG could decode typed sentence production at the character level. v2 moves toward continuous sentence decoding.

One interesting ablation finding is removing the MEG tokens makes performance worse, meaning the language model is actually getting useful residual signal from brain activity.

More caveats: avg performance is still imperfect (but it’s a move in a cool direction), and the authors are explicit about the limitations. There’s inter-subject variability, it uses healthy volunteers rather than disabled patients it’s aiming to help eventually, there’s actual typing during data collection, sentence level latency, and MEG hardware that’s nowhere close to a deployable clinical device.

Nonetheless, this feels like one of those moments where we start to see the beginning of something cool that may actually scale.

Morally, your takeaway shouldn’t be panic about the future of mind reading and the possibility piping ads straight to your brain, but rather that AI is becoming a very serious interface layer between noisy biological signals and useful communication.

Yes, this could be enormous for non-verbal disabled people.

It’s also another reminder that “thought” isn’t just mystical vapor, but rather something that leaves residual structure in a system models can potentially learn to navigate.

The thing I’d watch here is the scaling curve tho. If more subject-specific MEG data keeps improving decoding log linearly, then non-invasive BCI starts looking plausible but data/architecture constrained for now.

[Moll @Moleh1ll]

If a country allows its companies to train models on data from around the world, but then bans the rest of the world from using the result, other countries should have the right to restrict the use of their own data for training those models.

[Cléo Poulin 🇨🇦 @Coolbeanspoulin]

@DanielleFong Canada

[the tiny corp @__tinygrad__]

To every researcher in US frontier labs. You used to publish. You justified not because you were shipping something millions of people love. Now you are shipping only to Trump's cronies.

Consider your impact and legacy, you will be fine money wise. Monday is a good day to quit.

[Beff (e/acc) @beffjezos]

The real nuclear level power move here for xAI would be to release as Mythos level model as open source.

Since they own the infra, they can commoditize their complement.

It would save American open source AI from being conquered by China.

The most entertaining outcome...

[kache @yacineMTB]

I've always been a proponent of free software, free as in freedom, from day one. Open Source AI, for myself, doesn't mean free as in free beer, but rather free to do what I want, what I deem necessary, and not give up control.

With software this powerful -- software so powerful that my job isn't worth doing without access, letting someone else vend it to me is intolerable. I can barely tolerate using a close source image editor. You're telling me to trust you with the keys to god in a bottle?

You see, I don't really care about the fabled doom. To yield control to another person is to lose my freedom. At a press of a button, you could destroy me. My ability to create positive economic output, output I rely on. A false prophet's prophesied doom is hardly as frightening as an unaligned human master.

I'm a fighter. I'm not going to roll over and die. The guillotine of API sanctions weighs heavy. So I will grasp onto the technology as tightly as I can. Myself! I will help create a future where everyone can control it. Themselves! If it means short term concessions, so be it. If I have to be clever, I will be clever. I'll have faith, cope, some of you might call it. Even if I can increase the chances just a hair.

If I, an individual, feel this way, how do you think enterprises feel? I see many of my twitter commentator compatriots happy to roll over and die. Unfortunately, enterprises aren't as agreeable and soft.

I'm still learning, but I know enough about how much room is left. Room for our own computers to sputter to life. They're fast, you know? The phones, laptops, GPUs I've been hoarding. The game consoles, the smart watches, even my dishwasher. The dollars I make, which I carefully allocate.

There's enough compute in most households to run AIs powerful enough to materially change their lives. They will get more efficient. More intelligent.

Knowledge, artificial intelligence, it will tread. This isn't stopping. The tiger treads. Human computers, fields of women, were replaced by machines. 27 tonnes of vaccuum tubes. From ink and paper to punch cards. IBM machines, bank tellers no more. The personal computer, commodore 64, with assembly hand typed by a 14 year old. The open internet, connecting people across universities. Crushing all attempts to close it. Information transfer from every last corner of the planet, spreading culture, music, media.

Hasn't even been a century.

It treads, it tramples. Ultimately, it democratizes. Will AI be any different? What is the bottleneck to AI anyways? Human programmers? The lid on the compute multiplier genie isn't going back on. Every day that slips by is a few billion in valuation, your sleepless competitors are catching up. Your researchers are going to anthropic, then to openai, then to deepmind. How much longer can you hold?

Technology moves in one direction. It's as bad as its ever going to be. It's as expensive as it's ever going to be. I thank you for your work. You crazy, crazy motherfuckers. I know you've been staring at your bedroom ceiling every night. You and me both. Is your grip slipping? Or are you holding on? Plenty of space on the tiger. There are sights to see.

[z80.wei 👌☀️👌 @0xz80]

it’s so funny that “evil oppressive china” is just releasing open source frontier models

while the “land of the free” is single-handedly dictating one by one exactly who gets to use our frontier models

[Moll @Moleh1ll]

UBI? Rosy singularity? Heaven on Earth? Forget it.

Strong AI is the new nuclear weapon, the most important strategic resource. And weapons aren’t shared.

So now it’s going to be a closed club for elites and «proper» citizens with the right passport.

Welcome to cyberpunk, with the construction of a digital caste society.

[Cléo Poulin 🇨🇦 @Coolbeanspoulin]

@DanielleFong Please do! We need to figure out how to stop the brain drain. We've got everything we need over here, energy, resources, we would be killing it if we could keep folk like you around.

[Cléo Poulin 🇨🇦 @Coolbeanspoulin]

@DanielleFong We can only hope the rest of the world starts building compute. If we're at the rsi stage it's likely too late to lead. But even being a year or two behind the frontier may be preferable to being locked at an arbitrary intelligence lvl dictated by the US or China.

[Danielle Fong 🔆 @DanielleFong]

@BexelInitiative what fascist doesn't like super weapons that they get and others don't

[Gandalf Stormdrain @auroter]

This is the most narcissistic take imaginable.

"Distillation Attack"

You mean that they read the output of your product?

The output that you derived from all of our public information?

The output that you derived from countless authors, teachers, and other individuals who post information on the web out of the sheer desire to help educate people who share their passions?

What makes these accounts fraudulent? Did they not pay to access your product?

The answer is always spoken clearly in the lie itself The lie is that this was fraudulent. The truth is that *you* are fraudulent.

Intelligence is not private property if it's trained on public data.

The training is the only legitimately "private" thing, and the methodologies for training are well established and can be replicated by any commercial enterprise.

[Larry Pixel @LarryPixel]

I no longer understand what the United States is becoming under Trump.

Artificial intelligence is a legacy that belongs to the entire world. Every person on this planet has contributed, in one way or another, to the development of human knowledge and technology - everyone who has ever lived has left a trace in this shared progress.

They divided our common planet into countries and drew physical borders between us. Now they want to place borders around our minds.

Where is democracy?

It is nowhere to be found.

[Ejaaz @cryptopunk7213]

andrew’s spot on. what do you think anthropic’s doing with all that spare fable compute right now?

they’re using it to build mythos 6 which will make current models look like toys.

the spice will continue to flow regardless of what the government bans and there’s only one loser in all of this:

you.

[antirez @antirez]

Modern AI resulted from research made also by many non-US scientists (Hinton, the French folks, Linnainmaa, many others). The pre-training corpus was produced worldwide with massive code contribution from Europe OSS. What is happening with frontier LLMs is unacceptable.

[Zvi Mowshowitz @TheZvi]

Our new AI policy is that the White House decides ad hoc, for whatever reasons it likes, who does and does not get access to frontier intelligence. This seems rather maximally terrible.

[Peter H. Diamandis, MD @PeterDiamandis]

If we strip away the headlines about Mythos 5, the story is brutally simple: someone, somewhere, is deciding what level of intelligence you and your company are allowed to access.

[Taelin @VictorTaelin]

In ancient Egypt, writing was concentrated among scribes, giving a small educated class control over taxes, laws, land, religion, and state power while most people depended on them to understand the systems ruling their lives.