Umer @ssh_umer

@damian_b i want to do the same

few understand that authenticity is the highest form of looksmaxxing

₊˚ 🦢・₊ @itsmepspsps

3 days ago

I’m obsessed with people who are genuinely themselves. Weird little interests. Strong opinions. Specific taste.

241 27K 135K 1.6M 10K

AI lets you try more ideas than you'll ever ship. Each one can teach you something. Combine what you learn and that's what ships. Ideally something of value to users.

Thariq @trq212

3 days ago

the right way to use model capabilities is not to ship 10x more features to prod it's to spend more time understanding your users, trying experiments, building prototypes, learning about things you don't understand so that you can ship things that actually work

275 614 8K 381K 2K

the right way to use model capabilities is not to ship 10x more features to prod it's to spend more time understanding your users, trying experiments, building prototypes, learning about things you don't understand so that you can ship things that actually work

a lot of students have a hard time with this so let me spell it out: - class is for asking dumb ass questions to someone that knows the material - learning happens on your own outside the classroom - office hours is to ask adversarial questions to the main grader about exams

Mathieu @miniapeur

4 days ago

23 3K 19K 984K 612

@willcb @creatine_cycle Old school Do this instead Make another domain specific harness

Nice write up. I also think we're at the beginning of software engineering completely changing forever. Not just "muh vibecoding yolo", but something more fundamental: libraries are currently the cornerstone of swe. I think the future will be without libraries. You'll do ~everything in your codebase, tailored specifically to exactly what you do. The agents see the whole code. You can change the behavior of anything exactly the way you need it. No need to be backwards compatible. No need to worry about breaking others. What i think might become more ubiquitous instead is something akin to "starter packs" which could really just be something like tutorials explaining a concept. Like if you make a game, get the "rpg(.)md from SquareEnix" and the "entity-component(.)md from Lucas" files into context and go. Not "no libraries at all" but rather "only direct vendor libraries" like of course Vulkan or DirectX are still needed, but Unity or Unreal? Not so much. And i think this will happen in all domains. It will take a while, but i think e are at the very beginning of this future. I'm already starting to live in it. I used to love coding for the sake of coding, and was a bit worried of it going away. However, it turns out it's not really the coding itself that i love, it's more the combined act of creating something and having "puzzles" to think about while doing so. And it turns out that in this new way of coding, these two parts are just as fun.

mike @mike64_t

8 months ago

I think with Codex 5.3, the need for off-the-shelf deep learning libraries will fade away. Reasoning models operate best at the boundary of exact verifiabilty, so ever venturing too far into "well this is kinda correct" is no longer the best strategy. Exact verification now

keyboard_arrow_left Previous keyboard_arrow_right Next

17 26 480 108K 416

Humans try hard things, fail, learn, and get better through repetition. AI models aren't as different as you might think. Here's how models "learn" explained in simple terms. x.com/i/article/2080…

Solving a Rubik's Cube with graph theory.

Based based based based based based based Do not fall for Big Labs that make Frontier Intelligence look like a magic spell, it's all simple tech layered on top. Most of the big labs evals suck anyway - anyone not dumb can write better stuff. Do you think big labs employees can write good code? Do you think they can design anything well abstracted? No, not really. Their programs prove it, all of it sucks. Kudos to Kimi for making all this tech easily inspectable by everyone.

Kimi.ai @Kimi_Moonshot

2 months ago

Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside

2K 7K 46K 14.6M 14K

we're not so different from them

@ChShersh Any regrets?

Last month I wrote about how we can build a positive and safe future for everyone: meta.com/thefutureisfor… Every lab has the responsibility and incentive to move at the pace required to train its models safely, and the ability to take its own actions to ensure that happens. The reality is: - People won't want to use agents that are misaligned with them and that don't do what they ask, so labs have a strong natural incentive to make their models more aligned. There is a lot of debate about slowing progress on capabilities until alignment catches up. My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models. Any lab that doesn't focus on alignment will fall behind. - Labs face significant liability if their models cause harm, so they have a strong incentive to prevent this as well. Meta delayed shipping Muse for several months to focus on safety and security. We didn't call for everyone else to do this before we would. We just did it as part of our day-to-day work because it was clearly the right thing for people and for us. I'm proud of the security foundations we've built. - Engaging independent evaluators and advisors is industry best practice. MSL already does this today in several areas because it helps produce better work. Other labs can just do this too. In general, it would be helpful for there to be a larger and more diverse ecosystem of evaluators. - Committing the significant majority of compute towards serving people rather than racing towards recursive self-improvement is one of the best ways to ensure we develop this technology safely. Meta has made this commitment and other labs can do this as well. I believe the key to building a positive future for everyone is maintaining the right balance of power. This is within our power to do.

To have enough mental bandwidth to think deeply about what's going on in physics or any math-heavy field, you need to be very comfortable executing the underlying math. And that's not going to happen if the math you need is close to the edge of your ability. Sure, you can execute it... but not *comfortably,* and that makes all the difference. Your high-level train of thought is going to get continually derailed by the low-level details of the math you have to do. You're going to have a hard time seeing the forest for the trees.

Introducing CUDA Rust! CUDA Rust lets you write GPU kernels natively in Rust, not just launch them from it. Two paths: cuda-oxide for SIMT kernels compiled to PTX, and cutile-rs for Tile-based programming on stable Rust. Both can catch aliasing errors at compile time. Technical blog: nvda.ws/4hm1bHS

@rsuyoy @rahulchhabra07 Commonn bro plz share the code, i have burned so many tokenss for this

after working on Go backend my take is backend developers are UX designers too

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution

@PatrickToulme Our C/C++ training stack was written by humans. AI is not yet good enough for extremely high performance software (but it will be).

Building secure agents comes down to one major architecture decision and a few important sandbox configuration decisions. x.com/i/article/2099…