bubblyworld

10 ms·

91.
by bubblyworld 1y ago
Eh? I'm making no such case. It was a quote from the article.
92.
by bubblyworld 1y ago
It's really odd to me that pointing this out is used as some kind of counterargument. On its own that makes no sense. Consider that the frequency of replies along those lines might be evidence that there's something to it. It&#x27
93.
by bubblyworld 1y ago
Haha, well at least they warned you! > If you're from infosec, you might want to stop reading now — the rest of this article isn't going to make you any happier. Keep your medication close at hand if you decide to continue...
94.
by bubblyworld 1y ago
Me neither, for multiple reasons, especially if you are bundling your frontend. Very easy to accidentally include browser-incompatible code, becomes a bit of a cat and mouse game. On types, I think the real value proposition is having a sin
95.
by bubblyworld 1y ago
Go make an account and add your own links =)
96.
by bubblyworld 1y ago
I agree with you - I learned to program because I found it fascinating, and wanted to know how my computer worked, not because it was the only option available to me at the time... There are always people willing to take shortcuts at long-t
97.
by bubblyworld 1y ago
Oh, wow, that's quite a cool connection. Will look into that, thanks!
98.
by bubblyworld 1y ago
Yeah, not suggesting philosophers are the only people using logics, but they've certainly been using them the longest! Indeed, I've seen various attempts to tackle the problem including what you are suggesting - expressing the sem
99.
by bubblyworld 1y ago
Personally I think caring is the resource in extremely short supply here and I don't think vibe coding has much to do with it causally. AIs are just tools - basically all of the issues the author has raised are present with human jun
100.
by bubblyworld 1y ago
Formalising natural language statements is a minefield of difficulties, for (imo) essentially the same reasons that writing code which interacts with the real world is so difficult. Concepts you take for granted like identity, time, causali
101.
by bubblyworld 1y ago
Yeah, solving this edge case properly can add a lot of complexity (your solution has the same problem, no? deletes would mess it up as would updates, technically). I've seen people using long-lived "idempotency tokens" point
102.
by bubblyworld 1y ago
Doesn't surprise me at all haha. LLMs have anchoring bias in the extreme, anything you say can and will be used against you further down the conversation. In a sense I think it's one of their strengths too, provided you can curate
103.
by bubblyworld 1y ago
Yeah, that makes sense to me for a lot of reasons. Recently decide to take the plunge and start introducing AI tools in my startup. Fairly cheap experiment, all things considered - three months of giving it a try as a team and then we'
104.
by bubblyworld 1y ago
I agree, but my ego is incredibly fragile sometimes.
105.
by bubblyworld 1y ago
Lol, I think making confident claims in either direction is total copium at this point. It's abundantly obvious that LLM-based tools are useful, it's just a question of what we'll settle on using them for and to what degree
106.
by bubblyworld 1y ago
Absolutely matches my experience - it can still be super helpful, but AI have an extreme version of an anchoring bias.
107.
by bubblyworld 1y ago
I don't think the choice of metric matters to my point (I'm just calling it software-level because these are decisions you make while writing software). These decisions come with both pros and cons either way, which it sounds like
108.
by bubblyworld 1y ago
The ARC-1 problem set that they benchmark on is an example of such a problem, I believe. It's still more-or-less completely unsolved. They don't solve it either, mind, but they achieve very competitive results with their tiny (27m
109.
by bubblyworld 1y ago
I mean, I'm curious why this is such a highly upvoted article, which is why I'm engaging with this comment thread. I think criticism of the article is fine (although I acknowledge I could have been less blunt). It's the unn
110.
by bubblyworld 1y ago
Absolutely, everybody gets their say here.
111.
by bubblyworld 1y ago
Obviously engineering is broader than just trade-offs. I agree that broad heuristics can be useful but the ones proposed in this article just seem bad. Dependencies are (again, obviously?) really useful in many contexts. Same with build s
112.
by bubblyworld 1y ago
These aren't "problems" with a "solution". These are software-level decisions that have both pros and cons.
113.
by bubblyworld 1y ago
They don't really, as far as I can tell. There are lots of independent good and bad reasons to go down each of the roads in the article. Dependencies make a lot of sense in a lot of cases (e.g. the recent article on date parsing). So
114.
by bubblyworld 1y ago
No need for the random snark? I mean, again, we're in extremely obvious territory here. Just doesn't seem appropriate for hacker news front-page (to me).
115.
by bubblyworld 1y ago
I don't see the value in posts like this. It's just a random list of things that can go wrong in software projects, with no discussion of trade-offs at all. Where's the engineering?
116.
by bubblyworld 1y ago
> CoT models can, in principle, solve _any_ complex task. The authors explicitly discuss the expressive power of transformers and CoT in the introduction. They can only solve problems in a fairly restrictive complexity class (lower than
117.
by bubblyworld 1y ago
Okay, haha, I'm not sure what we're doing here.
118.
by bubblyworld 1y ago
You definitely need to be careful of the addition mechanics of CC, but is it still a slot machine if there are loads of steps you can take to improve your odds of success? For instance, only allowing it to code within a feedback loop of som
119.
by bubblyworld 1y ago
I don't know the jargon, but for me the main thing is the distinction between humans injecting additional bits of information into the training set vs the algorithm itself discovering those bits of information. So self-play is very int
120.
by bubblyworld 1y ago
True, it leaks information about all the symmetries of the puzzle, but that's about it. I guess someone needs to test how much that actually helps - if I get the model running I'll give it a try!

More ›