A topic for more in depth study to be sure. However:
1) video streaming has been around for a while and nobody, as far as I'm aware, has been talking about building multiple nuclear tractors to handle the energy needs
2) video needs a CPU and a hard drive. LLM needs a mountain of gpus.
3) I have concerns that the "national center for AI" might have some bias
I can find websites also talking about the earth being flat. I don't bother examining their contents because it just doesn't pass the smell test.
Although thanks for the challenge to my preexisting beliefs. I'll have to do some of my own calculations to see how things compare.
I had Hosking as a professor. Iirc, it was an okay experience. Compilers course I believe.
When the handbook came out, I bought it because "hey, I know that guy". Ultimately, I don't think it's necessary, but having a more in depth knowledge of garbage collection and the problems in the space occasionally comes in handy.
For example, what implication do finalizers have on garbage collection design? Reading about that was kind of an eye opener.
My objection to LLMs is the same that I had for TDD. There's all these people saying that you just gotta try it, but when I do, the effect is lesser than just using my preexisting skills. Oh, it's not for you? Wrong, here's some tautological or contradictory or poetic or nonsensical advice that'll be 'the wrong way' a week from now.
Does TDD and LLMs have a kernel of utility in them, yeah, I don't see why not. But what the majority of people are saying doesn't seem to be true and what the minority of people I can actually see using them 'for reals' are doing just doesn't applicable to anything I care about.
With that in mind, the only thing less real to me than a tool that I have to vibe with at a social zeitgeist level to see benefits from is an award when I already have major financial and industrial success.
Half the people in my team has played the game. For months all I would hear about w.r.t. games was how this game was smashing milestones and causing the entire industry to do some soul searching or putting their fingers in their ears.
I'm sure they can console themselves from having lost this award with their piles of money.
[An LLM did help me with a cryptography api that was pretty confusing, but I still had to problem solve that one because it got a "to bytes" method wrong. So... once in a blue moon acceleration for things I'm unfamiliar with, maybe.]
The output is correct for all inputs but only with the mocked dependency.
The output looks correct but the downstream processors expected something else.
The output is correct for all inputs with real world dependencies and is in the correct structure for downstream processors, but it's not being registered with the schema filtered and it all gets deleted in prod.
While implementing the correct function you fail to notice that the correct in every way output doesn't conform to that thing that Tom said because you didn't code it yourself but instead let the LLM do it. The system works flawlessly with itself but the final output fails regulatory compliance.
Security itself is a journey, not a destination. To say that you are secure is to say that you have been so clever that nobody else in the history of ever again will ever be as clever as you just were. Even knowing that they can study you being clever.
Even a super intelligent AI might not be able to replace lawyerhood unless it is also dynamically going out into the world and investigating new legal theory, researching old legal theory, socializing with the powers that be to ensure that they accept their approach, and carefully curating clients that can take advantage of the results.
* One developer tried to refactor a bunch of graph ql with an LLM and ended up checking in a bunch of completely broken code. Thankfully there were api tests.
* One developer has an LLM making his PRs. He slurped up my unfinished branch, PRed it, and merged (!) it. One can only guess that the approved was also using an LLM. When I asked him why he did it, he was completely baffled and assured me he would never. Source control tells a different story.
* And I forgot to turn off LLM auto complete after setting up my new machine. The LLM wouldn't stop hallucinating non-existent constructors for non-existent classes. Bog standard intellisense did in seconds what I needed after turning off LLM auto complete.
LLMs sometimes save me some time. But overall I'm sitting at a pretty big amount of time wasted by them that the savings have not yet offset.
> One developer tried to refactor a bunch of graph ql with an LLM and ended up checking in a bunch of completely broken code. Thankfully there were api tests.
So the LLM was not told how to run the tests? Without that they cannot know if what they did works, and they are a bit like humans, they try something and then they need to check if that does the right thing. Without a test cycle you definitely don’t get a lot out of LLMs.
You guys always find a way to say "you can be an LLM maximalist too, you just skipped a step."
The bigger story here is not that they forgot to tell the LLM to run tests, it's that agentic use has been so normalized and overhyped that an entire PR was attempted without any QA. Even if you're personally against this, this is how most people talk about agents online.
You don't always have the privilege of working on a project with tests, and rarely are they so thorough that they catch everything. Blindly trusting LLM output without QA or Review shouldn't be normalized.
A LOT of people, if you're paying attention. Why do you think that happened at their company?
It's not hard to find comments from people vibe coding apps without understanding the code, even apps handling sensitive data. And it's not hard to find comments saying agents can run by themselves.
I mean people are arguing AGI is already here. What do you mean who is normalizing this?
I fully believe there are misguided leaders advocating for "increasing velocity" or "productivity" or whatever, but the technical leaders should be pushing back. You can't make a ship go faster by removing the hull.
And if you want to try... well you get what you get!
But again, no one who is serious about their business and serious about building useful products is doing this.
> But again, no one who is serious about their business and serious about building useful products is doing this.
While this is potentially true for software companies, there are many companies for which software or even technology in general is not a core competency. They are very serious about their very useful products. They also have some, er, interesting ideas about what LLMs allow them to accomplish.
I am not saying you should be a LLM maximalist at all.
I am just saying LLMs need to have a change-test cycle, like humans, in order to be effective. But looks like your goal is not really to be effective at using LLMs, but to bitch about it on the internet.
> But looks like your goal is not really to be effective at using LLMs, but to bitch about it on the internet
Listen, you can engage with the comment or ignore everything but the first sentence and throw out personal insults. If you don't want to sound like a shill, don't write like one.
When you're telling people the problem is the LLM did not have tests, you're saying "Yeah I know you caught it spitting out random unrelated crap, but if you just let it verify if it was crap or not, maybe it would get it right after a dozen tries." Does that not seem like a horribly ineffectual way to output code? Maybe that's how some people write code, but I evaluate myself with tests to see if I accidentally broke something elsewhere. Not because I have no idea what I'm even writing to begin with.
You wrote
> Without that they cannot know if what they did works, and they are a bit like humans
They are exactly not like humans this way. LLMs break code by not writing valid code to begin with. Humans break code by forgetting an obscure business rule they heard about 6 months ago. People work on very successful projects without tests all the time. It's not my preference, but tests are non-exhaustive and no replacement for a human that knows what they're doing. And the tests are meaningless without that human writing them.
So your response to that comment, pushing them further down the path of agentic code doing everything for them, smacks of maximalism, yes.
> IMO pivoting to ads is a sign of core weakness for OpenAI.
Yeah, I've had the same thought for a while now. You don't sell investors on an endeavor for 10s of billions of dollars with the endgame being "sell ads". If that was the endgame then there are a lot less resource and capital intensive ways to get to it.
Given all of the discourse of "you need this new tech in your life to continue to participate in society", I would not have expected them to need to stand on the roadside trying to get people to buy low cost fireworks. It smacks of going through the sofa for loose change so you can make rent.
And if they had something impressive coming down the pipeline I would think they could get someone to spot them a few billions yet, unless the billionaire/megacorp economy is really that tapped out.
Yeah that was my thought. Although, I went a bit more paranoid with it.
If it looks like AI cheating software will be a problem for my children (and currently it has not been an issue), then I'm considering recording them doing all of their homework.
I suspect school admin only has so much appetite for dealing with an irate parent demanding a real time review of 10 hours of video evidence showing no AI cheating.
This is so close to my sentiment that I had to double check to see if I wrote it.
And to explore running studies a bit further. The second time you build a system goes so much better because you already know all of the weird edge cases. And the third better still because your failures in the second time has cured you of some of your hubris.
Even if you somehow bankrolled 50 repeat projects and did the statistics etc correctly, you're still going to get some weird artifacts because some of those teams have people who did the thing before. You'll learn the wrong lesson when the real lesson is "make sure Bob is working on Bluetooth because he's done it 10 times before."
Starting with people with no experience is likewise not interesting because nobody really cares what lessons you learn by turning a bunch of muggles loose on something difficult.
What you need to bankroll is 50 teams worth of people who spend their entire careers testing out a hypothesis. (And even then you probably need to somehow control their professional communities in some way because again who cares what some small group of people approaches a problem when you could instead have people who go out and learn things from other people.)
1) video streaming has been around for a while and nobody, as far as I'm aware, has been talking about building multiple nuclear tractors to handle the energy needs
2) video needs a CPU and a hard drive. LLM needs a mountain of gpus.
3) I have concerns that the "national center for AI" might have some bias
I can find websites also talking about the earth being flat. I don't bother examining their contents because it just doesn't pass the smell test.
Although thanks for the challenge to my preexisting beliefs. I'll have to do some of my own calculations to see how things compare.
reply