JoshuaDavid

7 ms·

31.
by JoshuaDavid 1y ago
It is striking how similar these answers are to each other, hitting the same points beat for beat in a slightly different tone.
32.
by JoshuaDavid 1y ago
Google employees collectively have a lot of talent.
33.
by JoshuaDavid 1y ago
There are some patterns you can use that help a bit with this problem. Lowest hanging fruit is to tell the LLM that its tests should test only through public interfaces where possible. Next after that is to add a "check if any non-publ
34.
by JoshuaDavid 1y ago
Are we using the same LLMs? I absolutely see cases of "hallucination" behavior when I'm invoking an LLM (usually sonnet 4) in a loop of "1 generate code, 2 run linter, 3 run tests, 4 goto 1 if 2 or 3 failed". Usuall
35.
by JoshuaDavid 1y ago
Ultimately it's probably not a productive use of time to be commenting here at all from a strict EV perspective. Meaning that if you're posting here, you're probably getting something else out of it. The value of that "s
36.
by JoshuaDavid 1y ago
I think "Who, specifically, claims that [...]?" comes off as less condescending than "Who claims that [...]? Be specific." just by virtue of the latter using imperative language, which triggers a reflexive "you&#x27
37.
by JoshuaDavid 1y ago
> What does it do better than other languages? Shared nothing architecture. If you're using e.g. fastapi you can store some data in memory and that data will be available across requests, like so import uvicorn, fastapi
38.
by JoshuaDavid 1y ago
I'm not sure - a lot of the top comments are saying that this article is great and they learned a lot of new things. Which is great, as long as the things they learned are true things.
39.
by JoshuaDavid 1y ago
One thing that has worked for me when I have a long list of requirements / standards I want an LLM agent to stick to while executing a series of 5 instructions is to add extra steps at the end of the instructions like "6. check if
40.
by JoshuaDavid 1y ago
LLMs largely either succeed in boring ways or fail in boring ways when left unattended, but you don't read anything about those cases.
41.
by JoshuaDavid 1y ago
We can almost certainly use 100x as much code as is currently written. There's a ton of throwaway code that, if written, would produce small but nonzero value. Certainly 100x as much code wouldn't produce 100x as much value though
42.
by JoshuaDavid 1y ago
Yeah, the logic is basically "sure there are lots of structural or root issues, but I'm not confident I can make a substantial positive impact on those with the resources I have whereas I am confident that spending money to preven
43.
by JoshuaDavid 1y ago
The general idea is that on the margin (in the economics sense), more resources should go to the most effective+neglected thing, and.the amount of resources I control is approximately zero in a global sense, so I personally should direct al
44.
by JoshuaDavid 1y ago
I suspect what's happening is that lots of people have a collection of questions / private evals that they've been testing on every new model, and when a new model comes out it sometimes can answer a question that previous
45.
by JoshuaDavid 1y ago
LeakyReLU is reversible in all but the least significant bits, right?
46.
by JoshuaDavid 1y ago
Fun game! Though I dispute that people are "very close" to achieving a perfect pattern. To get a "perfect" pattern you'd need to find three 7 letter words that can stack on rows adjacent to each other to form a 3 le
47.
by JoshuaDavid 1y ago
Along these lines, if the main LLM goes down a bad path, you could _rewind_ the model to before it started going down the bad path -- the watcher LLM doesn't necessarily have to guess that "skip" is a bad token after the word
48.
by JoshuaDavid 1y ago
> Or, most software will become immutable. You'll just replace it. The joys of dependency hell combined with rapid deprecation of the underlying tooling.
49.
by JoshuaDavid 1y ago
Seems like there's a real need that this is aimed to address. I don't understand how this protocol prevents double-spends though. If a payer sends the same signed payment authorization (i.e. same nonce) to two different online s
50.
by JoshuaDavid 1y ago
The training loop asked the model to one-shot working code for the given problems without being able to iterate. If you had to write code that had to work on the first try, and where a partially correct answer was better than complete fai
51.
by JoshuaDavid 2y ago
There are a few things working in favor of community notes there 1. Viral tweets have a longer-than-average time window between the time they start to go viral and the time the median viewer sees them, so a community note can get there befo
52.
by JoshuaDavid 2y ago
Twitter community notes are popular and pretty effective at calling out misleading information that has gone viral (organically or inorganically). I think writing off facts and arguments is premature.
53.
by JoshuaDavid 2y ago
Make the model write annotated tests too, verify that the annotations plausibly could match the test code, run the tests, feed the failures back in, and iterate until all tests are green?
54.
by JoshuaDavid 2y ago
If you're using the term "value" to refer to monetary value, yes. If you're using it to refer to some other kind of value, it probably isn't relevant to the viability of the proposed merge.
55.
by JoshuaDavid 2y ago
> Contrary to this, I'd like to see 3blue1brown actually get two more strikes and get deleted per YouTube "3 strikes" policy. Does the "3 strikes" policy apply to large channels that bring in a lot of views and a
56.
by JoshuaDavid 2y ago
https://www.tumblr.com/kingjamesprogramming
57.
by JoshuaDavid 2y ago
It depends on pronunciation, not on the first letter. This is why you say "an MBA" not "a MBA". If you are from a region that pronounces "historic" as "'istoric", then "an historic occasion&
58.
by JoshuaDavid 2y ago
Yeah, that's a pretty solid approach if the LLM you're using (and the third party that hosts it, if you're not self-hosting) supports that. One minor footgun I've seen with that approach is that while the model is guaran
59.
by JoshuaDavid 2y ago
Liking video games sometimes leads to modding video games, and modding video games is a gateway drug for "real" programming.
60.
by JoshuaDavid 2y ago
I wouldn't be surprised if it was something like, on the dev server SLOT_GAME_WIN_FRACTION=0.95 and then $win_fraction = (float) getenv('SLOT_GAME_WIN_FRACTION'); and then nobody bothered to set the environm

More ›