alexmolas

Yes, good point. I meant that the Jev can't be calibrated for everyone out of the box. I find it important because this is one of their main claims. They literally say

> Outcomes assigned a probability of 0.2 should occur about 20% of the time.

and this is not true

"Not especially good at classifying" does sound like a serious drawback for a classifier

If you want calibrated probabilities you'll be forced to fine-tune it

I don't know if it's reasonable. What it isn't is calibrated.

But "problems about actual probabilities whose answers can't be known at all" are exactly the problems where calibration is important. Since calibration is one of the big claims about Jev I'd expect it to perform well in these problems.

Read the first paragraph of the Wikipedia page [1] for a lot of sources claiming the opposite.

[1]: https://en.wikipedia.org/wiki/Gaza_genocide

Very interesting idea, but the result is a bit disappointing. It doesn't matter what I answer, the questions are always repeating

I used to work on a logistics company and we had to map latitude and longitude to specific directions. One of the first things I learnt was to avoid storing 6 decimal precision coordinates. Also, this XKCD was shared a lot https://xkcd.com/2170/

This is textbook survivorship bias. Out of 133 electable cardinals, someone was bound to guess Robert Prevost. If they were wrong, no one would remember. You could probably find 132 others who guessed wrong.

how does it compare against mypy? is it much faster?

> Religion needs science to keep it away from superstition and keep it close to reality, to protect it from creationism, which at the end of the day is a kind of paganism – it's turning God into a nature god

I love this quote!

There are two examples in the repo, one with CoRT and another one without. And the one without it it's much better than the one that uses it. Weird choice of examples...

The astral team (behind uv and ruff) is already working on a type checker in Rust, and given the quality of their existing tools, I'm inclined to wait and see what they release. Pyrefly looks interesting, but from the repo it seems pretty early-stage and not intended for external use yet.

is this the same as error propagation? I used to do a lot of that during my physics degree

I guess the biggest limitation of this approach is that the max output length is fixed before generation starts. Unlike autoregressive LLM, which can keep generating forever.

how did they solved the problems with pypi in China?

Exactly! They're evaluating the model using the training data

Cursor and Continue.dev

And the literature Nobel will go for ChatGPT

They can't spend more than 6 months without a drama...

Looks very nice! But I'm missing a "how to run it" section in the README.

Barcelona. It has good weather all the year. A great start-up environment. Great WLF. Great people. And a lot of things to do after work. Geographically it's placed in one of the best spots in the world. In winter you can go to ski in the Pyrenees, and in summer you can go to the beach yo Costa Brava or Mallorca/Menorca

But with FTS you don't solve the "out-of-context chunk problem". You'll still miss relevant chunks with FTS. You still can apply the approach proposed in the post to FTS, but instead of using similarity you could use BM25.

I was using data from the group stages. And I agree that just using the last 5 games had a lot of variance, but unfortunately, I didn't have data from before the Euro.

But the expected value of past games may not be the same as the expected value of future games.

I don't have the numbers to back this opinion, but I believe that consistently ranking in the top 5% in all the games will greatly increase your chances of winning the league. It's not necessary to win every stage, but just having high scores consistently will likely make you win the league.

If you're addicted to your phone and you want to quit it maybe it's a good idea to use a dumbphone with a shitty UX. Otherwise you're going to get addicted to your dumbphone.

This is why it's called Computer Science, and not Software Engineering.

And sometimes numpy is all you need, especially for POCs when the embeddings fit in memory.

my bad, you're right. Thanks for pointing it out :)