Most large tech companies are just job programmes, i've seen so much effort, time, cost go into things that are truly unimportant.
React Native gives you (with asterisks) - one "source" for things to go wrong (as it sits on top of the native implementations of the UI + Native APIs). You are writing at that source level, and that filters down to the native builds. I am WELL aware to do certain things on each platform you need to get your hands dirty, but most apps are just serialising JSON from a database and displaying it.
AI writes very buggy, sloppy code.
They've decided - instead of containing that code to 1 surface, they would now have 2 surfaces they throw slop on top of.
I look forward to the 2027 version of this where they've gone back
And on top of that, they're losing the ability to ship hotfixes over the air and will always be at the mercy of apple & google to deliver their updates.
But there's a bunch of people already in this thread calling that stuff unfounded speculation (which I disagree with), and my point is that even if that specific thing isn't true, OA's behavior here is obviously awful.
If they're going to try to beat researchers to discoveries like this it disincentives researchers to talk about their progress publicly, and basically breaks the ecosystem of scientific cooperation / discovery. It's also immoral.
Yep. The most uncharitable view of this might be: they stole the work of researchers to build their models, and now they're using said models to steal the proceeds of future work, too.
It's literally a toggle in the options for ChatGPT, one which is on by default and most researchers probably have on without realising it.
So to say that it is unlikely is extremely suspicious. No, they did not literally pull user data. But user data is automatically added to their training set by default, so their latest in-house model would be trained on it if it is from several months ago. It isn't intentional on their part, and they probably realised they could not refute that they trained on Tristan's logs unintentionally, hence why they acted the way they did.
Well, of course Anthropic employees would say that, since they likely do the same. Claiming that your primary competitor doesn't engage in a certain malicious practice is supposed to make it look as if there's no way you would too. If somebody even says that about their competitor, then surely there must be truth to that, otherwise you would never give credit to someone you're opposed to.
By default OA trains their models on codex-sessions. If I understand him correctly this is something Tristan explicitly mentions in his post as a possible reason for the fast results obtained by the internal OA team. Anthropic obviously doesn't want to challenge the idea that training is transformative, even if it means agreeing with their competitor.
I see a lot of enthusiasm from people who work alone or who have total control over a project and nothing but abject misery from people who work in a professional setting with a group of people using coding models.
It's a learning curve, but it can work, you just need to be comfortable with large (and likely quite negative) reviews for the inevitable large MRs that will get generated.
The important things to get right are the same as they were before though. Work from well refined stories that are not too broad in scope. Ensure you have enough good acceptance criteria that will help prove that the code works as intended.
Yeah so unverified PR's is super annoying, but people using LLMs to answer things incorrectly they could look at themselves is a big issue too.
I often hear this get dismissed along the lines of "oh you didn't context engineer hard enough" - but the default state of the model is to very confidently state a thing to be true when it hasn't searched correctly.
That is a trait of a very junior engineer - one who, if they never learned to fix this behaviour would be fired.
It seems objectively _worse_ than what we had before - trained engineers who gained wisdom over a long time horizon and had a reputation they'd lose if they kept incorrectly stating things.
I think he is saying this in reference to this:
"If you can reduce a problem to a clearly verifiable end state, provide the necessary context [...]"
Which for reference i am not an engineer but a Physicist is literally one of the things we make jokes about for engineers. Not that we are much better in that regard as a verfiable end state in Physics is like realy realy dificult to get so is the necessary context.
I don't mean to imply that was easy! I think it's hard, and a skill that needs active investment. It's one of the reasons I'm not afraid for my career.
Such a big disconnect in the coverage of coding models and the output they produce.
If you discard coding purity questions like style, architecture, cleanliness - the stuff they come out with is buggy & error prone.
The problems seems architectural - in that context windows are limited and you need more compute to increase them, married with the fact the models are really over confident. But if you do increase them it causes mode collapse. Yann LeCun has a really good graphic in his slides of a circle (all possible answers) and a red line coming from the centre depicting the one correct path. How do you actually stop the model going into the subsequence of wrong paths? I don't think it's possible.
I've had so many times in my day job someone has told me (Claude told them) there is a bug in my code, I look at it and nope - it just didn't look up the right file. Then you push back on it and it completely crumbles and says sorry.
I wouldn't keep an employee hired who did that over and over again and never learned
People acting like meat proxies add no value. Claude is also very eager to make conclusions without digging deeper. It has no inherent curiosity or prior knowledge about the codebase asside from what it can see.
None of the existing ones really made Linux on Desktop "happen", so their numerous and lengthy existence is kind of a counterpoint. The existing ones didn't work, other approaches may need to be tried — at least I commend that some enthusiasts continue trying.
Really depends your expectations. Linus on desktop is already successful from my point of view. There is no Linux company that needs to care about marketshare. Nowadays installing Linux and getting a working desktop environment with all the bells and whistles is pretty much a solved problem. It has even become a pretty good gaming platform. I would even say, better than macOS in that regards. There are still a lots of enthusiasts working on improving things, in fact using a Linux distribution has never been as popular as it is now, and we have way more resources to learn to get started. We are very far from the experience from 2010 (that was actually already quite good, but not mainstream ready)
reply