Gitlab’s AI-assisted code suggestions • Hacker News Board

Re: Gitlab’s AI-assisted code suggestions

#101

Post by pjot » Thu, Jun 29, 2023, 9:34 PM UTC

simonw wrote: post #72

endisneigh wrote: post #68

Earlier quoted context omitted.

Same. It’s funny reading comments on here. Even gpt4 I think is just ok. Maybe the people who see these huge increases are more junior?

I'm definitely not more junior, and I'm seeing massive productivity improvement from Copilot and GPT-4 - but I'm finding it takes a lot of expertise to get the best results, both in my ability as a programmer and in terms of knowing the best ways to use the AI tools. Learning how to get the best results of them takes a great deal of experimentation.

I’m curious what your workflow is like switching between copilot and gpt-4. I typically have an open window for each, though this can feel more cumbersome than necessary at times.

Re: Gitlab’s AI-assisted code suggestions

#102

Post by jameshart » Thu, Jun 29, 2023, 9:35 PM UTC

__loam wrote: post #93

jameshart wrote: post #66

Earlier quoted context omitted.

There’s a clear separation between the training process which looks at code and outputs nothing but weights, and the generation process which takes in weights and prompts and produces code. The weights are an intermediate representation that contains nothing resembling the original code.

I think this view is incredibly dangerous to any kind of skills mastery. It has the potential to completely destroy the knowledge economy and eventually degrade AI due to a dearth of training data.

Okay, that’s an argument from consequences, but is the view factually wrong?

Re: Gitlab’s AI-assisted code suggestions

#103

Post by jameshart » Thu, Jun 29, 2023, 9:36 PM UTC

szundi wrote: post #96

jameshart wrote: post #81

Earlier quoted context omitted.

The trained weights of a GPT model are a frozen, static, transmissible representation. They’re not equivalent to the live state of a brain.

Pretty equivalent to the snapshot of a live brain. Those inside it are even called neurons and neural network

No, they are the weights that are used to configure a neural network. They’re a map of how to build a useful brain, not a neural state.

Re: Gitlab’s AI-assisted code suggestions

#104

Post by axus » Thu, Jun 29, 2023, 9:39 PM UTC

__loam wrote: post #93

jameshart wrote: post #66

Earlier quoted context omitted.

There’s a clear separation between the training process which looks at code and outputs nothing but weights, and the generation process which takes in weights and prompts and produces code. The weights are an intermediate representation that contains nothing resembling the original code.

I think this view is incredibly dangerous to any kind of skills mastery. It has the potential to completely destroy the knowledge economy and eventually degrade AI due to a dearth of training data.

It reminds me of people needing to do a "clean room implementation" without ever seeing similar code. I feel like a human being who read a bunch of code and then wrote something similar without copy/paste or looking at the training data should be protected, and therefore an AI should too.

Re: Gitlab’s AI-assisted code suggestions

#105

Post by lee101 » Thu, Jun 29, 2023, 9:52 PM UTC

[dead]

Re: Gitlab’s AI-assisted code suggestions

#106

Post by olddustytrail » Thu, Jun 29, 2023, 9:55 PM UTC

mitthrowaway2 wrote: post #41

olddustytrail wrote: post #32

Earlier quoted context omitted.

LLMs don't memorize and reuse. They don't really have a memory of any kind. I think the problem is they're more different from human learning than you think.

The LLM's trained parameters are a lossy memory of their training data.

No, they aren't.

Re: Gitlab’s AI-assisted code suggestions

#107

Post by zacmps » Thu, Jun 29, 2023, 10:02 PM UTC

jameshart wrote: post #66

jedbrown wrote: post #29

Earlier quoted context omitted.

Even MIT licensed code requires you to preserve the copyright and permission notice. If a human did what these language models are doing (output derivative works with the copyright and license stripped), it would be a license violation. When humans want to create a new implementation with clean IP, they have one team study the IP-encumbered code and write a spec, then a different team writes a new implementation acco…

There’s a clear separation between the training process which looks at code and outputs nothing but weights, and the generation process which takes in weights and prompts and produces code. The weights are an intermediate representation that contains nothing resembling the original code.

But the original content is frequently recoverable.

You can't just take copyrighted code, base 64 it, sent it to someone, have them decode it, and claim there was no copyright violation.

From my (admittedly vague) understanding copyright law cares about the lineage of data, and I don't see how any reasonable interpretation could consider that the lineage doesn't pass through models.

IANAL

Re: Gitlab’s AI-assisted code suggestions

#108

Post by john_cogs » Thu, Jun 29, 2023, 10:03 PM UTC

lolinder wrote: post #83

john_cogs wrote: post #75

Earlier quoted context omitted.

Code Suggestions can be used in GitLab's Web IDE and VS Code and Microsoft Visual Studio when you have the corresponding GitLab extension installed: https://docs.gitlab.com/ee/user/project/repository/code_sugg... We offer experimental support for additional editors: https://docs.gitlab.com/ee/user/project/repository/code_sugg...

That doesn't answer the question of what the pricing will be. I'm assuming it's not going to be free. Will it be a standalone subscription, or another thing that gets bundled into the main tiers to justify the ever-increasing prices?

Code Suggestions is free while in Beta.

When GA, it will be included in our $9 per user per month AI add-on.

Re: Gitlab’s AI-assisted code suggestions

#109

Post by john_cogs » Thu, Jun 29, 2023, 10:03 PM UTC

ollieglass wrote: post #98

curtis3389 wrote: post #50

Okay, but how much are they charging? It's getting really annoying how many sites force you into a trial just to find out how much it'll cost when it ends. EDIT: Is this even positioned to compete with Copilot? What editors are there plugins for? There is surprisingly little information on the site.

Seems to be $99 per month. Signing up for the free trial funnels me to the trial of Gitlab Ultimate. So assuming that you need an Ultimate subscription to use it after the trial, that's the price. Pricing is here https://about.gitlab.com/pricing/ In contrast, Copilot is $10 per month https://github.com/features/copilot#pricing

Please see this comment: https://news.ycombinator.com/item?id=36527397

Re: Gitlab’s AI-assisted code suggestions

#110

Post by colordrops » Thu, Jun 29, 2023, 10:03 PM UTC

cj wrote: post #44

meowface wrote: post #28

Earlier quoted context omitted.

The same argument could be used to defend image generation models. (And, personally, I'm favorable to that argument for both.)

It's a blurry line, I think. Imagine a prompt "Photo of person, Shutterstock ID 132456, with blue eyes instead of brown eyes, watermark removed" If the prompt returns Shutterstock photo #123456 without the watermark (and with the different color eyes) but otherwise a near identical photo, I think most people would agree the output shouldn't be free to use without buying the original photo license from shutterstock. T…

You could say the same of an artist. If you ask an artist to paint a facimile with modifications, they would. That doesn't mean all of their output is "tainted" by copyright just because they learned to paint by referencing existing works.