agucova

Also, training over ChatGPT outputs is terribly expensive and Anthropic already has a working competing paradigm for fine-tuning (Claude worked before ChatGPT was released).

It seems more likely that this is just the consequence of training over up-to-date websites (which would include plenty of discussions about LLMs, and particularly ChatGPT). Claude is probably conflating its own identity.

An explanation of how they achieve the higher performance: https://million.dev/blog/virtual-dom

And how they implement it: https://million.dev/blog/behind-the-block

The article mentions this explicitly.

Did the prompt include my country as derived from geoip? This was a weird exchange:

CGPT: That's a great list! Do you have any specific companies in Argentina that usually hire for these positions?

Me: I'm not in Argentina, I'm in Chile. But I'm willing to move to other countries or work remote.

CGPT: That's great! There are many companies in Argentina that hire for these positions. Some of the most popular ones include Mercado Libre, Globant, Despegar, and OLX Group. Additionally, there are a number of startups in the country that may be looking for people with your skillset. What courses or skills do you think would help you increase your chances of getting hired by one of these companies?

Pop!_OS is much more than just reskinning Ubuntu. System76 is even developing their own Rust-based desktop environment [1].

[1]: https://blog.system76.com/post/more-on-cosmic-de-to-kick-off...

This page got me through my engineering calculus I, II and III, linear algebra and ODEs courses.

I'm eternally grateful ;)

For experienced programmers, I usually recommend:

0. Watching this [0] Fireship video to get the “big picture” on the language.

1. Reading the “Noteworthy Differences from other Languages” [1] section of the manual, if your language is on the list.

2. Then, if you're hurried, read Learn Julia in Y Minutes [2], or otherwise read the official manual [3] in order, skipping sections you're not interested in.

In case you're interested in scientific computing specifically, you can start from Chris Rackauckas' course on Parallel Computing and Scientific Machine Learning [4]. And if you don't have a straightforward application to test Julia on while learning, I would recommend doing the exercises available in Exercism [5], which are well-made and provide access to awesome mentors.

As for books, Think Julia is pretty good imho, even if it's a little bit outdated.

[0]: https://www.youtube.com/watch?v=JYs_94znYy0

[1]: https://docs.julialang.org/en/v1/manual/noteworthy-differenc...

[2]: https://learnxinyminutes.com/docs/julia/

[3]: https://docs.julialang.org/en/v1/manual/getting-started/

[4]: https://www.youtube.com/watch?v=-lJK92bEKow&list=PLCAl7tjCwW...

[5]: https://exercism.org/tracks/julia

[dupe of https://news.ycombinator.com/item?id=33878216]

She's the CEO of Alameda Research. SBF is the CEO of FTX.

I literally spent all day Tuesday trying to find a decent alternative to bit.ly. I'm so glad this exists.

At least compared to classical REST, usually access is limited to whatever the programmer explicitly chose to add, instead of being open by default.

I've wanted exactly this for years. I love LaTeX, but collaboration through platforms like Overleaf is severely limited, and the language itself has always felt so unnecessarily complex.

And while they haven't announced the platform itself to be open source, I'm glad they're at least already contributing some of the stuff they're making [1].

[1]: https://github.com/typst/

More recently, "wiki" seems to be a more concrete concept: a collaboratively edited website, managed by its audience [1]. The project references Wikipedia as an inspiration, and I hope it can live up to similar ideals.

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

While I found the concept interesting, having to request an account through a contact form just to edit an article [1] doesn't really live up to the “wiki” part.

[1]: https://wikenigma.org.uk/info/notes_for_contributors/become

Whatever your definition of “social movement” is (which appears at the very least, quite peculiar), you seem to be implying those two things are well defined, unrelated, things, and they're not.

You can make up whatever categories you feel are useful (the author here chooses to include values as part of technology, by example), but don't forget that categories tend to make people overestimate the difference between things.

Is arithmetics not part of mathematics?

That was probably a mistake on their part, considering that he points to such evidence in the phrase that follows.

There is a big causal step between "phones slightly warm brains" and "that causes headaches".

Fortunately we don't need to rely on the fact that it would be “hardly surprising” because there is good quality empirical evidence disproving the relationship.

SIKE is promising, but lacks research and maturity. It is mostly not ready for real-world use.

At this point, you might be tired of requests, but I'm also interested :).

The attack in question was only tested on SIKE, so it seems logical to start targeted disclosure on the community using and developing it, while using the general disclosures to target the broader cryptographic community.

Both Cloudflare and Microsoft are one of the few companies that have put significant investments into developing SIKE for post-quantum cryptography. Microsoft has a SIKE research team, and Cloudflare has been exploring SIKE for post-quantum TLS for years.

Both companies also maintain the key open-source implementations of SIKE [1][2], and Microsoft is spearheading the effort to standardize SIKE through NIST. Most open source cryptographic libraries don't implement SIKE.

[1]: https://github.com/cloudflare/circl [2]: https://github.com/Microsoft/PQCrypto-SIKE

They've explicitly justified not sharing it, for good ethical reasons.

There's a big difference between explicitly denying to share data for justified ethical reasons, and saying you will share the data and then failing to follow through.

There is absolutely no contradiction here.

I was somewhat expecting this argument when I wrote the comment.

The thing is that argument only works at the policy and governance level of EA. If I'm an individual who is convinced on the philosophy behind EA, and I'm willing to place it before anything else, then it doesn't matter what's optimal to “ensure commitment” because I'm already committed.

That's why I mean when I say that sustainability isn't a good general argument: it can help you in some cases, but not all. Examining EA as a system of values, that either suggests that EA is insufficient or that self-care shouldn't be that important.

> I just started reading the 80000 hours website. Their argument against the former seems to be that your continued career and income is your most important asset (unless you inherited a fortune), and not being miserable or depressed STRONGLY supports building a sustained income and career, and continuing to give.

I have a fundamental problem with the argument of “mental health maximizes giving” which is that it doesn't do a good job of justifying self-care in many situations.

Fundamentally, the issue is that sustainability can't be an enduring argument when you have finite time horizons (as all humans do).

For example, if I ever retire or become irreparably sick, should I just donate all of my money and make my life miserable? That is, if I'm no longer productive, what's the point of caring for my mental health?

This seems to imply terms can only have and only one “right” definition, which is not how languages work, at all.

Clearly there are two big clusters of meaning for “zero-knowledge” and they seem distinct enough to not overlap. Seems fine to me.

Note that ROI is not the only criteria for investment. Some people invest on lower ROI investments if they think the positive externalities align with their values, and risk is just as important as ROI anyway.

While I don't think this investment is effective either way, there is probably a market for people that want a more visible way to track their impact.

Is there any plan to expand to things like database modelling? I've been searching for a good modelling tool with automatic SQL generation, but most software is either legacy or very expensive.

I assume part of this is because the manuscript hasn't been properly typeset for publishing yet.