Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Gitlab’s AI-assisted code suggestions

about.gitlab.com

Post reply on HN
Print view View original

11–20 of 160 posts

Re: Gitlab’s AI-assisted code suggestions

#11

Post by colordrops »

RadiozRadioz wrote: post #3

Crucial: do they train on GPL code? If I am to use this tool, I must abide by the license terms of the training data. Even if it is found that the GPL does not cover LLM responses as derivative works, I would prefer to be on the safe side and refuse to use models trained on software with non-permissive licenses unless I am building Free Software.

Unless the LLM is cutting and pasting GPL code, I don't see how this differs from an engineer learning to code by referencing open source projects then creating their own project with its own licensing. I'd imagine these models could be tuned or at least a check could be added that no licensed code is being returned.

Re: Gitlab’s AI-assisted code suggestions

#12

Post by luxurytent »

cassianoleal wrote: post #8
dgrin91 wrote: post #2

I wish them luck. While I'm not a huge fan of copilot coding AIs, I understand how they are going to be a killer feature for many and I worry that it would create a big moat for Github. I hope that Gitlab is able to close that gap because I really like Gitlab.

Codeium [0] seems like a worthy competitor to Copilot already. [0] https://codeium.com/

Don't sleep on Cody either https://about.sourcegraph.com/cody

Re: Gitlab’s AI-assisted code suggestions

#13

Post by ianhawes »

This looks very interesting, but as an FYI using LLMs to do machine translation is a terrible waste of resources. The example on their landing page has a "translate.py" sample which, albeit handy, is not something I would do beyond basic string translations.

Re: Gitlab’s AI-assisted code suggestions

#14

Post by renewiltord »

Going to be interesting. Github had the advantage that they were offering Copilot into an audience of people who just defaulted to them. Gitlab on the other hand is for people who are into license wars, etc. so the AI product is going to be offered to a hostile audience.

Re: Gitlab’s AI-assisted code suggestions

#15

Post by lucasmullens »

ianhawes wrote: post #13

This looks very interesting, but as an FYI using LLMs to do machine translation is a terrible waste of resources. The example on their landing page has a "translate.py" sample which, albeit handy, is not something I would do beyond basic string translations.

> using LLMs to do machine translation is a terrible waste of resources.

Could you elaborate on that? LLMs seem perfectly suited for language tasks like translation. They don't seem particularly expensive either, especially compared to hiring a person.

Re: Gitlab’s AI-assisted code suggestions

#17

Post by nonameiguess »

RadiozRadioz wrote: post #3

Crucial: do they train on GPL code? If I am to use this tool, I must abide by the license terms of the training data. Even if it is found that the GPL does not cover LLM responses as derivative works, I would prefer to be on the safe side and refuse to use models trained on software with non-permissive licenses unless I am building Free Software.

I guess we'll see what the legal system says, but it's nearly impossible for me to imagine how anyone could ever justify saying a system trained on potentially millions of separate open-source projects can then be said to universally produce derivative works of any specific project it trained on.

On the other hand, if a developer using this tool then goes and tells it "please write me a C library in the style of GNU libc," then yeah, that is skirting a fine line. But just don't do that.

Re: Gitlab’s AI-assisted code suggestions

#18

Post by Chiron1991 »

ases wrote: post #4

All the code examples on this page are doubling any brackets ([], {}, ()). How have they managed that? Not a fantastic first impression of its capabilities...

Not only that, but the Golang example is full of errors. Parameter definitions don't have colons between name and type. The map for seen elements is declared as a, but later referenced to as m. The append instruction references uniques, which is undefined. The return statement also references uniques.

Re: Gitlab’s AI-assisted code suggestions

#19

Post by nonameiguess »

lucasmullens wrote: post #15
ianhawes wrote: post #13

This looks very interesting, but as an FYI using LLMs to do machine translation is a terrible waste of resources. The example on their landing page has a "translate.py" sample which, albeit handy, is not something I would do beyond basic string translations.

> using LLMs to do machine translation is a terrible waste of resources. Could you elaborate on that? LLMs seem perfectly suited for language tasks like translation. They don't seem particularly expensive either, especially compared to hiring a person.

There are already machine-translation services trained and created specifically for that purpose. While it's an amazing realization that LLMs can do this and do it pretty well without having to be trained specifically for this one purpose, training something to do literally all text generation tasks is expensive compared to training something specifically to do language to language translation.

For a maybe more obvious example, say that LLMs ever got good enough to do arbitrary precision arithmetic on numbers up to hundreds of digits. Would that be a good use of one when calculators can already do this and are far cheaper to produce? I guess it makes no difference from a free-tier consumer's perspective, but it's still more expensive even if you aren't personally paying the expense.

Re: Gitlab’s AI-assisted code suggestions

#20

Post by frankgrecojr »

RadiozRadioz wrote: post #3

Crucial: do they train on GPL code? If I am to use this tool, I must abide by the license terms of the training data. Even if it is found that the GPL does not cover LLM responses as derivative works, I would prefer to be on the safe side and refuse to use models trained on software with non-permissive licenses unless I am building Free Software.

https://docs.gitlab.com/ee/user/project/repository/code_sugg...

> Google Vertex AI Codey APIs are not trained on private non-public GitLab customer or user data.

Post reply on HN