More than a quarter of new code at Google is generated by AI.
More than a quarter of new code at Google is generated by AI.
The company had a strong quarter thanks in large part to AI.
I really don't believe the headline. Google has thousands of teams of engineers that are writing code for hundreds of different products... There's no way all of them are generating anywhere near 25% of their new code via AI.
Unless they're doing something like generating massive test fixtures or training data sets using AI and classifying them as "code" 🤔
I really don’t believe the headline.
The The company had a strong quarter thanks in large part to AI. part is what makes it sound strange to me, sounds like shareholder egostrking.
That said all they need to do is mandate use of AI during development like my company's done and they can boast this kind of bullshit easily.
I wonder if "code" means pull requests and they have a load of automated ones to update versions of external and internal libraries
Given the size of lockfiles this would not surprise me but who the hell counts lock files code. Their barely configs :/.
How often does a solution need “new” code and not “basically the same code as a previous issue but with two small details changed”? This is a genuine question, I have only ever coded as a hobby. But 25% of your work being essentially just copy pasted sounds plausible, and that’s sorta all LLMs are doing, right?
Reusable code is usually pulled out into a library and reused that way, rather than copied and pasted into a new project. You might copy and paste some boilerplate to new projects but it wouldn't be anywhere near 25% of the code.
I'm not sure why someone downvoted you (it wasn't me!) because your comment did seem like a genuine question.
Pretty often, but then you can just refactor the code so you can use it for more situations
What LLMs are good at are the opposite - when the thing you want to do is almost exactly the same, but nearly all the details need to be changed
Say you want a page to edit account details, and another page to edit community details. And the API paths to do this will be even more similar - but because they're different things, you'd have to get fancy with the design to make code that works for both... It's possible, but there will be trade-offs
LLMs are great at it though... Pass in the account page, give it the object definition for the community details, and it'll spit it out for you
If they're counting all the auto-completed code that's inserted after pressing Tab on an AI suggestion (such as from Copilot), then I easily believe it.
Tons of places in code only have 1 possible thing that can go on a particular line, given the context, and there is no point in typing it all out manually.
Does this mean "AI was used as a fancy autocomplete"? Because that's my number 1 use case for AI like copilot, and if that's the case, over 25% of my code is written by AI. But let me tell you, it still gets it wrong, repeatedly making the same syntax errors no matter how many times I correct it. It starts to get it right, then later reverts to making the same syntax errors, even making up variable names that violate widely known public APIs.
Auto complete is about... 60% helpful and increases my productivity with about 5-10% as I need to double check everything it does and half the time it's something ridiculously stoopid
Agreed. It's really shit for new code, but if I'm writing glue code stuff or repetitive code it saves a lot of time spent on typing.
25% of all new code written at all? Sure, I guess.
25% of all new code that actually gets used in a real product, not just tested in an IDE? Bullshit.
I wonder if they do the monkey writing shakespeare experiment but with code. If you keep letting it write code, something has to come out of it.
You still need to check the hell out of it because AI is wildly unreliable.
That would explain the decline in quality of everything from Google. Even Gmail is becoming buggy as hell, even though I haven't seen any new features added. I have used Gmail since it's founding and only in the last year or so did it become extremely buggy.
Now they can fill new holes at the google graveyard at twice the speed!
I've read exactly the opposite article a few days ago:
https://www.cio.com/article/3540579/devs-gaining-little-if-anything-from-ai-coding-assistants.html
This company wasn't trying to bullshit financial analysts which was the reason for the google CEO comment.
Ah Elon most love all this extra code being written. If course it's super inefficient but look at all those lines sooo much code.
Are the lines salient though?
How is Elon going to tell the AI to print its code on hardcopy and then fly to Elon's city to show him your hardcopy code like he told actual Twitter developers to do?
Makes sense considering how shitty Google products have become.
NICE TRY, AI !