HNHacker News
TopNewBestAskShowJobs

flimsypremise

301 karma · joined January 12, 2023

submissionscomments
flimsypremise··on Artificial General Intelligence remains a distant dream despite LLM boom
It is quite literally copying and pasting from StackOverflow, and any other code that it has been trained on. I think you're underestimating exactly how much code there is out there, and the degree to which most of it is boilerplate.

Here's an example of what it can't do. I was recently messing around with the Plaid API, building sqlalchemy models for the Plaid API objects and defining relationships between them. After about 45 minutes of playing around, I got about 80% of the way to what I created without ChatGPT. It saved me a bunch of time googling around for documentation, but it gave me an absolute garbage password encryption implementation that was literally copied out of an intro tutorial I had read elsewhere, didn't even seem to recognize that there is a Password field type in sqlalchemy_utils, and omitted a bunch of key aspects of the model relationships. It also used field arguments that were either deprecated or not best practice for what I wanted to do. I also had to ask it to model every object individually, even though I specifically asked it to model all of the fields necessary to implement transactions.

Now I could go and ask it to specifically modify those things, but it doesn't "know" the best way of building this set of models. It gives me the statistically determined output of what it has consumed from all of the answered questions on stackexchange and tutorials it has been fed, and provides a halfway decent starting point, but to get from there to a working application requires actually understanding why you might build a model one way rather than another. You need to understand exactly what behavior you want to see when you delete a user, and what needs to happen to all of the related objects. You need to know that id columns should be UUIDs and not integers or strings, which is something you commonly see in tutorials but you would (hopefully) never see in a production application. All of this knowledge is the result of years of experience building applications and knowing not only what to do, but why you do it and in what context. ChatGPT is very useful reference library. It blows google searching for examples out of the water, but you need to be very careful with the results, and very knowledgeable to use them properly.

flimsypremise··on Artificial General Intelligence remains a distant dream despite LLM boom
Its not reasoning and solving problems. It's a statistical model. It gives the appearance of reasoning and solving problems, but the second you start asking it to solve problems that aren't in its training database it gives you nonsense.

I'm surprising at how little actual testing of an LLM's ability to reason based on well-established hard problems in AI (like halting) is happening. Philosophy of mind has produced a number of logic problems that are designed to test the reasoning ability of an intelligence, and instead LLMs get tested on content that is explicitly in their training dataset.

flimsypremise··on No code payments automation is a lie
You could just copy this and paste it for any "no-code" solution. Beyond a certain degree of complexity and customizability, engineers are always required. You can't just create configurable automation for everything, because at some point the configuration becomes just as (or more) complex than the coding necessary to create the thing in the first place. Then you are just creating a new context for which a trained person needs to be hired with domain-specific knowledge, except in this case the domain is proprietary, which is actually what I think all of this no and low code automation is about.
flimsypremise··on The problems with live coding interviews
I keep seeing people claim that no one offers alternatives to live coding in interviews, but not only does the author provide an alternative (take home coding test/project), there are obviously many alternatives to live coding. I've been working in this industry for over two decades, and I've interviewed more candidates than I can count. I can assess a candidate with a 20 minute conversation, because a candidate's understanding of the context and theory of how and why things are done the way they are determine how good they are at the job. Ask them what happens when you visit a website. The amount of detail they go into, and what details they focus on, will tell you an enormous amount about them. Ask them whether they prefer an OO or functional approach to organizing code. The good candidates will have thorough, well-thought out opinions, and you're handing the bad candidates enough rope to hang themselves.

Of course, in order to be able to assess the answers to these questions in an effective way, you need to be able to very knowledgeable yourself, and that's the real root of the problem here.

flimsypremise··on We Need To Talk About Vercel
I'd say one of the big annoyances I've encountered while building applications with the last few versions of NextJS is the increasingly tight integration with the built-in server. The world of Node servers is pretty well-established, with documented interfaces that the major server platforms all implement and support. Simple stuff like res and req in the context and what those objects contain. With each release NextJS breaks a bit more of those interfaces and replaces them with approaches that are very specific to NextJS and pretty unintuitive to anyone who is used to building servers in Node. I'm talking about stuff like handling response codes (the notFound parameter you need to return to get a 404 from getServerSideProps is a particularly egregious example) or redirects. You are presented with what looks like a standard interface, but doesn't fundamentally actually work anything that interface.

Decades of working in this industry have taught me to value interoperability, modularity and standardization. I love the idea of a framework that makes SSR easy, and the idea of static site generation, and a router that uses the pages approach where each file is a route is handy in some cases (though very clumsy in others), but all of those a different things and aren't really something I need a single monolithic framework for. I may want one or some or none of those things based on what project I'm working on. If NextJS makes it harder to pick and choose what I want to use, or makes some features contingent on using other features I don't really care about, I'm going to start looking really hard for an option that gives me more choices.

flimsypremise··on We Need To Talk About Vercel
My sense of Vercel (mostly from working with NextJS) is that they are more interested in appearing to support an open source framework while making their product as difficult to interoperate with other technologies as possible in an attempt to lock users into their platform and hopefully pay for it.
flimsypremise··on Unpopular Opinion: Don’t Use a Raspberry Pi for That
The big benefit of something like a pi is that it offers the flexibility of a linux-bootable computer with the IO pin interface of an MCU. That being said, maybe there's a cheap, easy way to get a GPIO interface on a NUC? I know there are USB to GPIO boards you can buy, but I have no idea how easy they are to use, and what the drawbacks might be. There are also extremely expensive GPIO interfaces that negate the cost-savings of using a cheap NUC.
flimsypremise··on There Were Half As Many Affordable Homes For Sale In 2022 As There Were In 2021
You do realize that the bulk of the US population lives in the top 100 metropolitan areas? That's probably why it was chosen as a metric.
flimsypremise··on Starbucks illegally fired US workers over union, judge rules
pretty much all of these seem very normal and i'm actually shocked starbucks doesn't already have them in place. you can work 32 hours at a starbucks and not be considered full time? starbucks doesn't supply fresh aprons to everyone every shift?
flimsypremise··on Tech's Elite Hates Labor
there is absolutely no way that these guys are 100x as valuable as the people how actually build the product. if larry page and sergei brin's search engine hadn't won out, someone else's would have. these guys are all just average to dumb, and very often end up vaporizing millions if not billions of dollars with stupid decisions. there is no doubt in my mind that these tech giants would be better off if they fired their ceos and c levels. this should be obvious to anyone who has ever interacted with the ceo of a tech company.
flimsypremise··on Conversion of HP 54520C / 54540C oscilloscopes to VGA display(s)
I wonder if this would work on the A versions of these machines with the monochrome displays.
flimsypremise··on Ask HN: Strategies for working with engineers that are too smart?
What you are describing is a smart, but very junior engineer. As smart engineers gain experience they stop doing this sort of thing.
← PreviousPage 3 of 3