Live data from Hacker News

GitHub cuts AI deals with Google, Anthropic

bloomberg.com

671–680 of 742 posts

Re: GitHub cuts AI deals with Google, Anthropic

#671

Earlier quoted context omitted.

As soon as something can be automated, people start acting irrationally upset, especially if the thing is seen as even remotely "creative". Those people are going to have a bad time moving forward.

I don't see this being raised as an automation issue. The GP comment is concerned with their work, including licensed code, being used by Microsoft to train LLM models without any kind of agreement or even compensation. There isn't clear legal precedent yet whether training models is an acceptable use of licensed work, but it has nothing to do with automation.

If the legal precedence comes out as "yes, it is legal", as many countries have already done (e.g. Japan), would GP really change their mind and become okay with it? (Note: the Copilot lawsuit have also mostly been dismissed, so this is already close to reality.)

I doubt it. I think the GP will still be concerned about it, and would petition to change the law. So it has nothing to do with the legal precedence either, unless the GP concern genuinely comes from legalities, instead of using it as an argument. But why would the GP continue to be concerned?

The first possible reason I can think of is automation. That is, "no one cared until the models became good enough". The GP might fear for their job, or have "artist envy".

The second possible reason is a distaste for corporations, and wanting one's due for contributing to it in any way, regardless of what the law says (note: assuming that the training is legal as aforementioned). So this is more of a personal morals issue, one that I disagree with but must acknowledge. I must also point out that open weight models exist.

Re: GitHub cuts AI deals with Google, Anthropic

#672
post #388

Earlier quoted context omitted.

You need to spend time learning how to use it. This is difficult because there's no manual, and there's a widespread implication that it should just magically work well without you having to invest any effort in it. If you can figure out HOW to invest that effort it becomes really valuable. I wish I had good resources I could link you to here but I don't, which is a big part of the problem here.

You sound like you’ve had success using this tech for work. Can you tell more about your personal experience, please? I’ve tried ChatGPT a few times a year ago or so, but it was extremely frustrating, and I gave up.

Solve the problem at a high-level in your head, ask the LLM if your concept is correct, and then instruct the LLM to build the solution step-by-step.

Re: GitHub cuts AI deals with Google, Anthropic

#673

Earlier quoted context omitted.

I'm actually very curious why AI use is such a bi-modal experience. I've used AI to move multi thousand line codebases between languages. I've created new apps from scratch with it. My theory is the willingness to baby sit and the modality. I'm perfectly fine telling the tool I use its errors and working side by side with it like it was another person. At the end of the day it can belt out lines of code faster than I…

As a programmer of over 20 years - this is terrifying. I'm willing to accept that I just have "get off my lawn" syndrome or something. But the idea of letting an LLM write/move large swaths of code seems so incredibly irresponsible. Whenever I sit down to write some code, be it a large implementation or a small function, I think about what other people (or future versions of myself) will struggle with when interactin…

It really depends what you're building.

If you're building code that's going to go in some medical system, or a space shuttle, then yeah, you probably want to write every small function with great detail.

If you're creating some silly consumer app like a "what will your baby look like in 5 years", then code quality doesn't matter you just need to ship fast. Most startups just need to ship fast to validate some ideas, 99% of your code will be deprecated within a few months

Re: GitHub cuts AI deals with Google, Anthropic

#674
post #456

Earlier quoted context omitted.

Here's my "How I use LLMs and ChatGPT" series: https://simonwillison.net/series/using-llms/ Also relevant is my ai-assisted-programming tag: https://simonwillison.net/tags/ai-assisted-programming/

Sorry, it’s a bit hard to understand from your blog. I’m a bit dyslexic, so there’s a chance I’ve mussed something obvious, but I couldn’t find any examples beyond ‘explain this short snippet’. Also, are your posts AI assisted? I find it hard to read. I’ve tried asking ChatGPT questions about how a large codebase works, what entities should I use to implement some feature, etc. After a couple hours I realised the cha…

The problem with your example of applying analyzing something as complex and esoteric as a codebase is that LLMs cannot reason they simply return a response based on statistical inference, so unless you followed a standard like PSR for PHP and implemented it to a 't' it simply doesn't have to context to do what you're asking it to do. If you want an LLM to be an effective programmer for a specific application you'd probably need to fine tune and provide it instructions on your coding standards.

Basically, how I've become successful using LLMs is that I solve the problem at a 9,000ft view, instruct the LLM to play different personas, have the personas validate my solution, and then instruct the LLM step-by-step to do all of the monkey work. Which doesn't necessarily always save me time upfront but in the long run it does because it makes fewer mistakes implementing my thought experiment.

Re: GitHub cuts AI deals with Google, Anthropic

#675
post #634
post #633

This sort of makes me sick as a software engineer with licensed code on GitHub. Am I understanding correctly that they have trained data on my code despite my license? Do I receive monetary payment from the deal? Or have I misunderstood this?

How is that different to a human reading your code and building up their experience? Is reading code now also covered by a license? It does not reproduce your code 1:1.

Co-pilot autocompletes my full name and copyright information.

Re: GitHub cuts AI deals with Google, Anthropic

#676

For all those believers in the power of AI who tested it in modifying their front-ends and writing a Python script, I have a test: ask AI to write an operating system kernel or a database. Of course, something simple. I never seen AI being used in writing system software. Perhaps there is a reason behind it?

Dawg, LLMs cannot reason, they simply return a response based on statistical inference (voting on correct answer). If you want it to do anything correctly you need to do a thought experiment, solve the problem at a 9,000ft view, and hold its hand through implementation. If you do that there's nothing it cannot do.

However, if you're expecting it to write an entire OS from a single prompt it's going to fail just as any human would also fail. Complex software problems are solved incrementally through planning. If you do all of that planning its not hard to get LLMs to do just about anything.

Re: GitHub cuts AI deals with Google, Anthropic

#677
post #221

I use cursor and its tab completion; while what it can do is mind blowing, in practice I’m not noticing a productivity boost. I find that ai can help significantly with doing plumbing, but it has no problems with connecting the pipes wrong. I need to double and triple check the updated code - or fix the resulting errors when I don’t do that. So: boilerplate and outer app layers, yes; architecture and core libraries,…

I'm actually very curious why AI use is such a bi-modal experience. I've used AI to move multi thousand line codebases between languages. I've created new apps from scratch with it. My theory is the willingness to baby sit and the modality. I'm perfectly fine telling the tool I use its errors and working side by side with it like it was another person. At the end of the day it can belt out lines of code faster than I…

>I've created new apps from scratch with it.

Would mind sharing those apps to view. Not the code, just the apps. I have a suspicion about the bi-modal experience.

Re: GitHub cuts AI deals with Google, Anthropic

#678

Earlier quoted context omitted.

I'm curious: Suppose I upload some code to GitHub, but I didn't have the authority to share it with anyone at all. And then it was used to train DL models. How would various jurisdictions handle that? Would any of them force the deletion of all resulting model weights? And how might the remedies differ based on the kind of data? E.g., copyright vs. trade secret vs. protected medical info vs. military secrets?

Not a lawyer, but I would expect you to be the one on the hook there as you shared code without permission and likely lied when agreeing to GitHub's T&Cs. Microsoft wouldn't be able to pull that code out of already trained and, given that MS didn't do anything illegal when they used code that you said was yours to share, I wouldn't expect them to liable at all. That means MS wouldn't likely be fined, nor would they h…

> Microsoft wouldn't be able to pull that code out of already trained

I imagine they could, they just wouldn't want to. Because it might require retraining the model from scratch, or at least from some not-very-recent checkpoint.

Re: GitHub cuts AI deals with Google, Anthropic

#679
post #491

Earlier quoted context omitted.

It’s really far from mid level. It’s a weird mix of expert at things it trained on, and complete misleading idiot at anything outside. For a bash script or the first steps of something simple it’s great. For anything complex at all it’s worse than nothing.

For anything complex, move in small steps. For anything truly novel, or on a codebase with a very bespoke in house architecture or DSL, yeah you won't get much out of it.

Even in small steps, it fails. I have two cases I test with, nothing special, just some TS generics in one instance and a schema-to-schema mapping tool in another. Both things that Junior devs could do given a couple days, even though they'd need to study and figure out various pieces.

o1 can't get either, no matter how much I break it down, no matter how much prodding. In fact the more you try the worse it gets. And yes I do try starting new conversations and splitting it out. Simply does not help, at all.

It's not to say it isn't really helpful for really simple things. Or even complex things but that are directly in the training set. But the second you go outside that, it's terrible.

Re: GitHub cuts AI deals with Google, Anthropic

#680
post #221

I use cursor and its tab completion; while what it can do is mind blowing, in practice I’m not noticing a productivity boost. I find that ai can help significantly with doing plumbing, but it has no problems with connecting the pipes wrong. I need to double and triple check the updated code - or fix the resulting errors when I don’t do that. So: boilerplate and outer app layers, yes; architecture and core libraries,…

I use copilot to write boilerplate code that I know how to write but I don't feel like writing it. When it gets it wrong, it's easy to tell and easy to fix.
Post reply on HN