"The vast majority of Codex is built by Codex," OpenAI told us about its new AI coding agent.
See full article...
See full article...
How OpenAI is using GPT-5 Codex to improve the AI tool itself
- Thread starter JournalBot
- Start date
Great questions! The kinds of questions journalists are paid to ask. I wonder why the journalist writing so many of these AI articles never thinks to ask these types of questions. Even when the comment section is full of suggestions for this very thing
And why doesn't Ars hire a better journalist since their current writer is failing at basic journalism? They could easily find someone like you right in the comment section if they wanted to! Why are they ignoring great options for hard-hitting journalism?
Guess we'll never know!
Upvote
38
(52
/
-14)
I'm going to get a good grade in asking the computer to make something for me, something that is both normal to want and possible to achieve.
Upvote
-9
(9
/
-18)
Upvote
38
(39
/
-1)
Glad I'm retired and can let the next generation figure this out
The developer was not just coding against an API, but (hopefully) had a broader sense of why he was doing it.
In my teams, the “junior developer” sat in on all meeting, including with stakeholders. a lot was written down, but a lot was conveyed that was not. The team communicated in writing and verbally.
The developer was not just coding against an API, but (hopefully) had a broader sense of why he was doing it.
Upvote
78
(78
/
0)
Upvote
64
(65
/
-1)
Missing Minute
Wise, Aged Ars Veteran
I deliberately included some implausible explanations to reinforce the plausibility of other explanations.
Upvote
22
(23
/
-1)
No, it is the number of engineering managers and CTOs telling devs to use them. Then, we do, because a mandate is a mandate.
Upvote
61
(61
/
0)
"We have all this amazing, world-changing, incredibly value and useful technology. We're just, uh, waiting for the right time to reveal it. Yeah, that's it." -- OpenAI
That's the smell of bullshit and desperation...
That's the smell of bullshit and desperation...
Upvote
81
(81
/
0)
I am getting close to retirement as a software engineer. I am now really looking forward to it. I have really enjoyed the creative process inherent in software engineering. That creative process is being eroded away by these so-called advancements.
Upvote
61
(63
/
-2)
Unfortunately this "beat" of Benj's is more of an infomercial than critical journalism...
Upvote
85
(89
/
-4)
Upvote
35
(35
/
0)
Alas, this has been a problem at ars for years, at least. They really don't like pushing or contradicting their interviewees, or independently confirming things they say. Once you realize that it's hard to stop seeing it. I like ars - I've been here a long time now - but I really wish they'd commit more to doing research.
It's not just AI though. I'm a neuroscience nerd, which is where I first noticed the lack of investigation or skepticism. Their reporting on things I knew something about always just seems to be repeating what one source said about a topic. They rarely go to an independent source to try and fact check.
Last edited:
Upvote
74
(74
/
0)
Respectfully, this is incredibly uncritical and even naive coverage.
I’ll tell you right now I absolutely do not believe these claims. This feels like marketing.
I’ll tell you right now I absolutely do not believe these claims. This feels like marketing.
Upvote
120
(123
/
-3)
Man I use this thing, I type commands into it, and the computer follows the commands. Some might call it a chat interface, some might call it a terminal. When I want to make full programs, I open up an editor that natively understands what I type into it, like it's integrated the things needed to develop the programs. What are you going to call your alternative?
Upvote
13
(13
/
0)
I do not believe anything that comes out of OpenAI.
It occurred to me that post-IPO Altman will have to be a lot more careful about the truthfulness of his statements.
Then I remembered that for that to matter the US would need a functioning SEC and Department of Justice.
Oh well. The IPO (and the cash out for early investors that goes with it) may be the thing that pops the bubble.
Upvote
64
(64
/
0)
Missing Minute
Wise, Aged Ars Veteran
Yeah the AI would likely be substantially different, experts in those fields often have safety and ethics training.
Upvote
33
(33
/
0)
Bra, we just started to force push to main, and we started shipping new code in record time!
Upvote
42
(42
/
0)
It isn't. Their parent corporation made a devil's deal with OpenAI, and its reporting on the subject has never been the same since.
Link: Condé Nast Announces Partnership with OpenAI
Upvote
94
(98
/
-4)
But the commentariat is second to none. I saw the article headline and thought "Ohhhh... interesting! Let's read the comments" Sometimes the comments make me think that the article is also worth reading. The comments are always worth the visit.
Thank you, and everyone else.
Upvote
123
(123
/
0)
macosandlinux
Ars Praefectus
Sam Altman's previous eye-scanning venture literally partnered with criminals. Want the sources, I am happy to PM you.
Upvote
28
(28
/
0)
This is definitely what's happening at some companies:
View: https://bsky.app/profile/alexhanna.bsky.social/post/3m4g6eti5mk2m
A company taking the position that whether or not you use "AI" coding tools will be considered in your performance review.
Upvote
29
(29
/
0)
If these tools were actually any good, I would be quite concerned about letting an AI code itself.
Upvote
31
(31
/
0)
developers are the most expensive text workers
on code you can easily check if it is "correct" - atleast if it compiles.
Nasty errors in a text are more difficult to spot and handle
Upvote
-15
(9
/
-24)
I mean, that's the dream/hype/pr-bull. Where is the evidence that anything like that is happening or possible? Still, if it does happen, we know it's going to be paper clips.
Upvote
11
(12
/
-1)
To be fair, this is true of a lot of journalism, even at places like the NY Times. Anytime they report on something that I actually know something about, I'm amazed at how wrong the reporting is. But it's especially bad when it comes to reporting on "AI" (a term that is not meaningful unless it is in quotes). Pretty much all journalism is just regurgitating the "AI" hype.
Upvote
36
(37
/
-1)
The reporting at Ars on "AI" is been pure uncritical hype since always. It long predates Conde Nast's deal with OpenAI.
Upvote
36
(39
/
-3)
At least they didn't go and do that one thing that you do when the AI is about to take over, and it's that thing that it uses to gain the abilities. Damnit, what is that one thing? I'm so forgetful sometimes.
Upvote
6
(6
/
0)
I've been playing with the idea to make a program that can modify itself when I was a student somewhere in the year 2000. No Ai, just a genetic algorithm. Would be nice if it found a way to exit the testing environment through some back door and conquer the world. I never got to it, as you probably figured out by now. I could not find a good scoring function. Not even close. The closest I got was to replicate itself as hell, inspired by rabbits. But the scoring function (not sure what the proper English name is) is actually the big challenge. I bet I am not the first one to think of something like this. Does anyone have any pointers on this subject? No worries, my ambitions to conquer the world have faded away together with a big part of my hair. Cheers!
Upvote
12
(13
/
-1)
I think about turning it off almost everyday. I'm fighting it 75% of the time. It's gotten maybe 5% better a trillion dollars later too...
Upvote
6
(7
/
-1)
What worries me is that my experience with multibillion-dollar clusterfucks is that it feels like we've not yet reached the end of the beginning. We're at the part of the horror movie where it's perfectly safe to go down to the basement alone to grab some stuff.
The thing about multibillion-dollar clusterfucks is that unlike in horror movies, people will just keep going down into that metaphorical basement long after people have started to go missing, and just consider it the price of doing business. No scary music, no ironic "I'll be right back", just an honest, "if I return".
On the other hand, who knows, maybe they'll actually find an inflection point where clusterfucks scale up to the point that they somehow become useful? Like the clusterfuck equivalent of a defibrillator?
Upvote
33
(33
/
0)
Today actual programmers can solve issues generated by AI.
Next step will be AI generates its own programming language for "better efficiency", language that will not be understandable by actual programmers. And goes straight in the wall. Just faster.
Next step will be AI generates its own programming language for "better efficiency", language that will not be understandable by actual programmers. And goes straight in the wall. Just faster.
Upvote
7
(11
/
-4)
Unfortunately, LLMs are also used a lot in HR, Finance and other critical domains. And with far less control than in development.
Upvote
31
(31
/
0)
Compiling isn't a valid test , that's what unit tests , integration tests , loading testing etc is for.
Upvote
49
(49
/
0)
This sounds like it will enable bad actors to cheaply pump out more scams and malware with less sophistication requirements rather than the useful high quality work developers are expected to make for their clients. Much like the deluge of scams image and video generation are enabling in the online advertising industry.
Upvote
31
(31
/
0)