jimsimmons

What exactly do SMT systems "solve" in cases like this?

If I wrote a simple BFS or DFS and enumerated the search space how far would I get.. Is that not what TLA+ does in principle.

I am surprised people prefer having a dependency of something like Z3 at compiler level.

Right, how do you know what you want to do is not sleeping 15 hrs a day versus getting inspired by some idea

JAX tries to be a functional language that has a Python front end. The problem is if you are outside Google and don't really understand the XLA compiler, then you are screwed.

I agree that nested loops are verbose. But einsum is unstructured and not extensible, hard to debug in pieces, hard to document etc.,

It comes down to honesty. Are you idle because you want to be idle or are you just being lazy and not going after the opportunity right before you? Are you not pursuing the low-hanging fruit out of principle or is your ego holding you back from doing humbler things?

If you can be honest with such questions then the duality goes away and everything boils down to doing the right thing

Einsums are the regexes of tensor programming. Should be avoided at all costs IMO. Ideally we should be able to write native loops that get auto-vectorized into einsums for which there is a CUDA/PTX emitting factory. But for some reason neither PyTorch nor JAX/TF took this route and now we are here.

Some of the einsum expressions I have seen for grouped multi headed/query attention is mind-boggling and they get shipped to prod.

Why wouldn't you be able to implement this in Rust or Python?

Seems like typical exception handling. Erlang isn't even type checked

Don't use it.

I tried using it for coding but it made up APIs constantly or had subtle errors. I don't know what people mean when they say it works for them. I can see it help as an icebreaker though. It just helps me get in the mood sometimes.

Other factual stuff ex nutrition questions, health stuff, historic events etc it's quite useless.

I am an LLM skeptic. I think there're a lot of people who're claiming that they find it valuable just like crypto was claimed to be valuable

The Gemini white paper reports higher scores on HumanEval and other tasks.

So one of Google lied, this eval has bugs, they borked the deployment is true

Crazy that Apple gets to ship a proprietary messaging app as the default

Clearly people very close to Altman were accessed for this article. I'd doubt that'd happen if they deemed that the reporting was going to be prejudicial

Very interesting theory about GDB being CEO. I thought it was more of a gratitude tour

Paul always has this delicate restraint in praising Altman it's hilarious. It's like he knows there might be a scandal one day and he doesn't want to have those positive endorsements lying around

I work in AI research in the Bay Area where hustle culture is strong. OpenAI and hot startups have normalised working long hours and taking few vacations.

Hustle is great if you have a well defined, measurable objective. With such an aim it is easy to keep track of your progress over time and maximize your work.

But when you want to do research, design great composible systems, be creative, write good code, it is better to do it casually. With such endeavors the goal is not well defined or measurable. Pressure to be productive or have demonstrable results often leads to myopic thinking, local minima, unsustainable designs, technical debt and spaghetti code.

I haven't quite been able to capture why there exist these two modes. The moment I try to talk about the creative type of work my colleagues look at me like I'm making excuses or that I'm a slacker.

There's a GitHub issue on the repo that explores this!

Wrong link/title

n00b question: what are some important problems for which quantum computation has shown to have an advantage

They already use closed form summations. That's their whole point

Most problematic is when articles critical of OpenAI or Altman get off, despite seeing a lot of activity.

Very hard to prove

A can of coke is a few inches. 150K cans of coke is huge. That's enough to get to the moon?

Everything is something. Question is what this nomenclature gymnastics buys you? Unless you answer that this is no different than claiming neural networks are a projection of my soul

How to hire a good stylist? (In SF)

What is the best way to deduplicate a corpus of documents

Some journalists on X are claiming that this story is fabricated

Sunday night: this is the only path forward.

Monday morning: she cried!

I thought it was one of the fastest web frameworks but this article completely destroys that image. What gives

Wow kinda respect them testing staffs willingness to quit.

The juicy tender is fueling a lot of the motivation here, more than any loyalty or preference to Altman

MSFT comp is shit though

OAI comp was high based on equity and its crazy valuations

How will equity and compensation work

AI peeps are not cheap