nerdponx

22 ms·

391.
by nerdponx 2y ago
Primary care has the opposite problem right now. Practices are closing faster than new ones are opening, and doctors are leaving the profession faster than new doctors are joining. There is an actual shortage of primary care docs.
392.
by nerdponx 2y ago
You don't need to include 0 on every axis. In this case they really made the numbers smaller than they should be, so it's hard to see that the scale is on the order of single digits. It looks like this is about a 3-4% improvement
393.
by nerdponx 2y ago
Lesson 1 about competing with Google should be "don't be even more disrespectful to your users than Google is". Otherwise people will just use Google. Relying on the goodwill of a small number of "never-Googlers" to
394.
by nerdponx 2y ago
I didn't know there was a bad art museum there. However I do strongly recommend the penis museum in Reykjavik.
395.
by nerdponx 2y ago
I flipped through the "unseen forces" section and so far about half of them aren't actually bad. For example Monochrome 006 (supposedly inspired by Schoenberg) would IMO fit right in at MOMA and was actually kind of cool. Lik
396.
by nerdponx 2y ago
It's a Github-ism. Github renders a Markdown-formatted readme file as the "front page" of a Git repository.
397.
by nerdponx 2y ago
> I couldn’t believe they whiffed it Why should we expect a general-purpose instruction-tuned LLM to get this right in the first place? I am not at all surprised it didn't work, and I would be more than a little surprised if it di
398.
by nerdponx 2y ago
I thought the same thing about Spark, coming from R and later Pandas.
399.
by nerdponx 2y ago
Pandas can be easier for interactive poking around on smaller data sets. It feels more like Excel, whereas Polars feels more like SQL. It's no surprise that on a forum full of programmers Polars is widely preferred, but IMO it's w
400.
by nerdponx 2y ago
I think it's worth giving some credit to prior art here. Polars was heavily inspired by the "Spark SQL" API, and Pandas by R data frames. I disagree that Pandas is no longer state of the art: its interface is optimized for a
401.
by nerdponx 2y ago
The problem I see in a lot of cases is putting too much emphasis on solving the problem in the interview, instead of working through it. Many interviewers claim that they care more about the latter, but in practice failing at the former e
402.
by nerdponx 2y ago
I always assumed it was just a misspelling of "wary" by analogy to "wear". You're telling me that people actually say "weary" when they mean "wary"? That's very different from what I thoug
403.
by nerdponx 2y ago
The wary/weary mixup is so strange to me, especially because it seems relatively new (last 5-10 years at most).
404.
by nerdponx 2y ago
But not every business finds it usefulness gradually eroded by its revenue model.
405.
by nerdponx 2y ago
It is and would be useful, but it would be quite a big lie to the public, but more importantly to paying customers, and even more importantly to investors.
406.
by nerdponx 2y ago
Ironically it's probably a lot closer to what a super-human AGI would look like in practice, compared to just an LLM alone.
407.
by nerdponx 2y ago
This is what Google did. Even if you technically don't manipulate the search results around sponsorships, you can devote more and more real estate to the sponsored results, and gradually make the distinction between sponsored and norma
408.
by nerdponx 2y ago
Sort of. 3rd-party recruiters get paid on commission: they make money when their candidate gets hired in favor of the other recruiters' candidates. In that sense, the incentives are actually aligned in your favor when you're apply
409.
by nerdponx 2y ago
The good recruiters are the ones who won't put you forward for a job that you're not qualified for, and will advocate for you during the hiring process in a job that you are qualified for. I have worked with recruiters like this b
410.
by nerdponx 2y ago
A lot of institutions have to do this even for internal hires. It's an astonishing waste of everyone's time and resources, not only the candidates'.
411.
by nerdponx 2y ago
Matrix multiplication and the Gaussian distribution are also like that. A lot of things are like that. I really dislike that this approach is not a core tool for teaching in math.
412.
by nerdponx 2y ago
I wonder if the best benchmark is a Prolog program that generates tests of logical reasoning. You could have a functionally infinite stream of test cases!
413.
by nerdponx 2y ago
Could you run the benchmark by bootstrapping (average of repeated subsampling), instead of a straight-across performance score, and regain some leakage resistance that way? As well as a better simulation of "out of sample" data, a
414.
by nerdponx 2y ago
You might also want to support a Unix timestamp as input, i.e. an integer or decimal number of (mili|micro|nano-)seconds since the Unix epoch. No need to worry about messy date parsing there.
415.
by nerdponx 2y ago
That's actually fair. I'm not anything close to a constitutional expert, but I I'm willing to assume that there is either court precedent backing this up, or there has never been a meaningful challenge to it and so it stand a
416.
by nerdponx 2y ago
> The U.S. Department of Treasury is amending its definitions of a military installation, adding to the list of military installations covered by the Committee on Foreign Investment int he United States (CFIUS) > The new list adds alm
417.
by nerdponx 2y ago
The whole innovation of GPT and LLMs in general is that an autoregressive model can make alarmingly good next-token predictions with the right inductive bias, a large number of parameters, a long context window, and a huge training set. It
418.
by nerdponx 2y ago
> The top LGBT-friendly employer these days is Raytheon. The what? Slapping a pride flag on a jet engine isn't exactly a political bias, it's a marketing / reputation-washing gimmick. > Even pre-Trump the Dems were the
419.
by nerdponx 2y ago
If you are being lied to, then the liars are at fault. If you know that you are being lied to and keep on listening without at least increasing your level of skepticism, then you are a collaborator and an enabler and a public-information-we
420.
by nerdponx 2y ago
Freedom (such as free speech) has always been limited by harm caused to others. Or do you suggest that we somehow impose a Pigouvian tax on letting people lie to you for "entertainment"? Let's also be clear: the people crying

More ›