eachro

Anyone got a list for the agentic LLM age?

Do you think this would be appropriate for a command line tool that hits various apis as the function calls? Ex: "what's the weather in SF tomorrow?" Or "daily price change of apple, Tesla stock for past week"? (Let's assume I have documented the apis thoroughly somewhere that the model has access to or fine tuned it on this data)

Does anyone know what the state of the art industry solvers do for these problems? I had dabbled a bit in ml approaches to combinatorial optimization with great interest a few years back, but I don't think any of these rl based methods ended up being used in production.

Didn't Amazon aquihire Adept Labs?

If you wanted to train it from scratch, how long would it take on a reasonable GPU setup?

I'm reminded of the nixon quote: "When the president does it, that means it's not illegal."

What would it take to make NYC more like Tokyo where you have consumer/retail level things on the not-ground floor level.

From what I've heard, the llama3 models are fairly easy to fine-tune (please correct me if I'm wrong or if there are more amenable models here). How easy is it to finetune smollm3? I know a lot of the MoE LLMs have been quite fickle in this regard.

"And 50% of the time they work 50% of the time."

I think this is still an incredible outcome given how many dice rolls you can take in parallel with multiple claude/o3/gemini attempts at a problem with slightly different prompts. Granted, each rollout does not come for free given the babysitting you need to do but the cost is much lower than going down the path yourself/having junior colleagues make the attempt.

Yeah haha I got autocorrected

Is there a reason to use data classes over pedantic base models anymore?

A lot of people are saying 12gb is too small to do anything interesting with. What's the most useful thing people __have__ gotten to work?

During the openai gym era of RL, one of the great selling pts was that RL was very approachable for a new comer as the gym environments were small and tractable that a hobbyist could learn a little bit of RL, try it out on cartpole and see how it'd perform. Are there similarly tractable RL tasks/learning environments with LLMs? From the outside, my impression is that you need some insane GPU access to even start to mess around with these models. Is there something one can do on a normal MacBook air for instance in this LLM x RL domain?

For those in quant finance, how much of this is useful in your day to day?

This looks great! Is there going to be exercises or a project based component as well?

Thats a great choice. High trust city, super high density.

Do the rankings ever change much year over year? For instance, with intro fountain pens, it will always be things like the platinum preppy, pilot metro, kaweco sport.

Is there consensus about whether gnn architectures are better than transformer based ones at this point? I am aware that transformers can be viewed as a gnn too.

Reproduce nanogpt.

Then find a small dataset and see if you can start getting close to some of the reported benchmark numbers with similar architectures.

I'd love to see linear regression taught by say a quant researcher from Citadel. How do these guys use it? What do they particularly care about? Any theoretical results that meaningfully change the way they view problems? And so on.

This is cool that simd primitives exist in the std lib of rust. I've wanted wanted to mess around a bit more with simd in python but I don't think that native support exists. Or your have to go down to C/C++ bindings to actually mess around with it (last I checked at least, please correct me if I'm wrong).

Training in int8 is noteable (to me). I've been out of date with ML research for a bit now but last I recall, people were mostly training at full precision and then quantizing after training and finetuning a bit on the quantized model afterwards.

Is student debt dischargeable in bankruptcy?

I wonder how he regards his success as a productivity guru vs professor. All credit to him for achieving a level of success, notoriety that most will never get close to in their life. At the same time, I suspect productivity guru is not quite what he wanted to become. Or maybe it is. I don't know.

When people learn about distributed systems outside of work, how do they actually get hands on experience with it (assuming they don't go spinning up a bunch of machines on aws/gcp/azure/etc)? I find it easiest to learn by doing, writing simple proof of concepts but that seems a bit harder to do in this area than others? What is the hello world/mnist of messing around with distributed systems?

Can this actually be observed from data though?

Because there are real problems to work on that benefit from the sort of problem solving with algorithms that students are hoping to do.

Every year I see tons of CS students captivated by the beauty of algorithms in theoretical CS. I lose track of all the bright eyed undergraduates saying they love thinking about algorithms and would ideally like to spend their summer doing research on these sorts of problems. More often than not, I end up telling them they should focus their efforts on the systems side of things and chat with an OS or DB professor rather than a prof in TCS, but very few of them actually take this advice.

Do people use numba outside of scientific computing code? feel like for most "normal" uses, you'd probably be fine with having your code written in numpy or pandas, so I'd love to hear what people generally use numba for outside of scientific computing.

Super fun game. I really wish there were a larger online community that plays it. Jane Street made a great website for playing it online: https://figgie.com/. But alas, there's pretty much never anyone online to play against.