6,395 karma · joined April 19, 2017
More interesting was the independent commentary paper they linked near the bottom: https://www-cdn.anthropic.com/files/4zrzovbb/website/cc4be24...
Neel Nanda (of Google Deepmind - his part begins on page 33) discusses his opinions on the paper, and the small-scale replication he performed on an open-weight model.
You can still indirectly call native JavaScript, in a message-passing kind of way (via Ports or custom elements) but these changes were still really disruptive to many codebases.
What has mostly changed is that there are more editors now, and thus more eyes and also more serious discussion (rigor?) about such things.
> Wikipedia has a bunch of languages that were used by like one person, because there is published material on them
No. It's more like, there are plenty of articles on Wikipedia that don't meet Wikipedia's notability guidelines AT ALL, but when you write an article on Wikipedia and enough time passes without anyone noticing that the article is poorly sourced, then eventually the tendency of Wikipedia community is to just keep it.
This is what has led to the what-about-ism regarding Odin's deletion - there are lots of other programming languages that also don't meet the notability guidelines, yet, to this day, still have Wikipedia articles.
Could someone come along and propose deletion for such articles? Yes, of course. You yourself could go do that right now, if you want. But nobody's getting paid for such work, so someone has to want to. The tendency of Wikipedia editors is that, when an article is many years old, they would rather flag it for improvement rather than simply throw away years of fellow editors' work. Whereas an article that's brand new is likely to not have much work put into it, and also more likely to be self-promotion and/or spam.
This is very frustrating for people who create Wikipedia articles and have them deleted. "You mean, whether or not my non-notable article gets deleted or not is just the luck of whether someone comes along and notices that it's not notable?" Yep. Like I said, nobody's getting paid for deletion work.
You could always use setuid to allow the agent to run designated commands whose operation depends on the files, without the agent itself being able to access the files.
For example, if you allow weak passwords, then you have a trivial vulnerability to people guessing other people's passwords. But nothing about the login system is working differently from how anyone intended. It's just that the intentions themselves were naïve.
If you're dealing with a party who tracks currency in cents, then tracking currency with more precision than that is going to lead to rounding disagreements. Vice versa if you deal in cents but they deal in tenths of cents. And so on for all the other advice in this document.
Maybe? This one's kind of subjective. I'm sure there are some people who will feel this way, and many who won't.
Do you respect artists less, now that you can make AI images?
As for pay, it seems unlikely to me that the job title of "software engineer" is going to see a significant decrease in median wage as a result of AI. Though there may simply be fewer "software engineer" jobs and more "prompt engineer" jobs.
> Implication 2: Optionality changes the commitment to software products
It's not clear to me that the typical decision-making process the average company was using to choose (for example) project management software, is going to be significantly different in the AI era than before. "Let's use JIRA, since that's what everyone else uses."
Making decisions this way is low-risk, and lower-cost than the token cost of vibe-coding something custom. The analogy to dating apps doesn't work - dating apps reward searching far and wide for something perfect, whereas the business world rewards going with what you know and making decisions quickly.
> Implication 3: The middle class of software products will disappear
I don't believe the cost of software creation is approaching zero. People are taking this concept too far and too literally. First of all, obviously there are token costs. And secondly, obviously there is still a time and effort requirement involved in maintaining anything, even via vibe-coding. Most companies have absolutely no reason to prefer to incur these costs rather than simply paying the man his $50/month.
But thirdly, and probably most importantly, there's the inherent cost of merely being responsible for something. Like I wrote earlier, decisionmakers want to minimize risk. The mere fact of being responsible for something - of it being someone's fault if something goes wrong - is a dire political cost, which most business leaders try to avoid by buying external rather than creating in-house. The SaaS market isn't going anywhere.
> Implication 4: If you want to win, sell services, not products
Service automation is a fruit that has already been mostly squeezed by conventional software.
That is to say - the space of things that traditional software can't already automate, that LLMs would be capable of automating, and that LLMs would be reliable and efficient enough at to significantly move the needle on real productivity - is small.
(Ironically, software development is one of the few things in that space. Since when you automate software development you can also automate the creation of tests that (at least attempt to) validate the correctness of the software itself. Not so much for legal documents.)
PureScript has much better support for row polymorphism for this reason.
Right now, `if` in expression position is just a syntax error ("unexpected symbol")
I feel like this is the actual problem, but it's not being talked about enough.
Logically, it shouldn't matter how a piece of code was written. It either meets our engineering standards or it doesn't. That's what code review is for.
But what has actually happened is that, because we have AI now, many organizations have normalized practices that weren't normal before, like submitting 10000-line PRs. And just, in general, submitting code that you yourself don't seem to fully understand.
If management doesn't push back on such things (or will even push back against people who try to do so), then reviewers have basically no incentive to push back either. Before I left my last job (for unrelated reasons), I had entered this mode, wherein I decided I couldn't really carefully review these monstrous AI PRs, while also getting my own work done, while also not burning out. So I chose the latter two.
In fact you could literally just buy a separate PS5 and Macbook Neo and spend less than most Steam Machine configurations, so even the "it's also a computer" selling point is not that big of a deal.
If they had paid $60 billion dollars in cash for Cursor, it would have been a ripoff.
Still frustrating (if the taxpayers want it, might as well let them have it), but not purely a semantic technicality.