Meta Releases Code Llama, a Code-Generating AI Model (techcrunch.com) 20
Meta, intent on making a splash in a generative AI space rife with competition, is on something of an open source tear. From a report: Following the release of AI models for generating text, translating languages and creating audio, the company today open sourced Code Llama, a machine learning system that can generate and explain code in natural language -- specifically English. Akin to GitHub Copilot and Amazon CodeWhisperer, as well as open source AI-powered code generators like StarCoder, StableCode and PolyCoder, Code Llama can complete code and debug existing code across a range of programming languages, including Python, C++, Java, PHP, Typescript, C# and Bash.
"At Meta, we believe that AI models, but large language models for coding in particular, benefit most from an open approach, both in terms of innovation and safety," Meta wrote in a blog post shared with TechCrunch. "Publicly available, code-specific models can facilitate the development of new technologies that improve peoples' lives. By releasing code models like Code Llama, the entire community can evaluate their capabilities, identify issues and fix vulnerabilities." Code Llama, which is available in several flavors, including a version optimized for Python and a version fine-tuned to understand instructions (e.g. "Write me a function that outputs the fibonacci sequence"), is based on the Llama 2 text-generating model that Meta open sourced earlier this month. While Llama 2 could generate code, it wasn't necessarily good code -- certainly not up to the quality a purpose-built model like Copilot could produce.
"At Meta, we believe that AI models, but large language models for coding in particular, benefit most from an open approach, both in terms of innovation and safety," Meta wrote in a blog post shared with TechCrunch. "Publicly available, code-specific models can facilitate the development of new technologies that improve peoples' lives. By releasing code models like Code Llama, the entire community can evaluate their capabilities, identify issues and fix vulnerabilities." Code Llama, which is available in several flavors, including a version optimized for Python and a version fine-tuned to understand instructions (e.g. "Write me a function that outputs the fibonacci sequence"), is based on the Llama 2 text-generating model that Meta open sourced earlier this month. While Llama 2 could generate code, it wasn't necessarily good code -- certainly not up to the quality a purpose-built model like Copilot could produce.
All of you are fired! (Score:2)
Re: (Score:2)
Re: (Score:2)
I'm not so sure.
"Pull the lever!"
"Not that lever!"
"Why do we even have that lever?"
Or:
"Uh-oh."
"Let me guess, giant waterfall ahead."
"Yup."
"Sharp rocks at the bottom?"
"Most likely"
"Bring it on." Followed by screams.
Not the images I would want associated with my code.
Re:All of you are fired! (Score:5, Insightful)
Code Llama will do your jobs now.
Well, those people who had the illusion that the job of a programmer was to write software. For those that understood that the real job is to spend your entire life trying to debug horrible bugs that others have put into the software this vastly increases the potential for subtle and illogical bugs that go beyond human comprehension and require decades of consulting hours to even track down. We can even hardwire AI based LLM debugging into the build process to ensure that any bugs allowed through are ones that can't be picked up using AI based LLMs. What could be better?
Re:All of you are fired! (Score:4, Funny)
I would definitely vote you up, if I had mod points.
Just waiting for the first LLM generated code to hallucinate an API to a non-existant (but plausible) library that is not compatible with another similarly fictional library whose imaginary authors haven't supported it since 1853.
Re: (Score:2)
Just waiting for the first LLM generated code to hallucinate an API to a non-existent (but plausible) library that is not compatible with another similarly fictional library whose imaginary authors haven't supported it since 1853.
I think we're definitely going in that direction. My experiments with Bard and ChatGTP both came up with REST API paths that should exist but didn't. A few could actually be corrected and work, but most were completely non-existent APIs that were just like APIs for other systems. Try asking them for code to automate something which can only be done manually (needs an actual press on a device screen or button) and you'll probably get the same. I'm sure that if we feed them enough steampunk novels with Babbag
can this be tuned for a private code base? (Score:2)
This is great that it can generate Fibonacci function but it wound be great if I could tune it on a private code set to aid with proprietary domain.
does anybody remember (Score:2)
Jeff Minter ?
"At Meta, we believe ..." (Score:4, Funny)
Anything following is probably BS. This goes for all tech companies that start a statement like this.
Re: (Score:1)
I mean, what they wrote sounds about right for what they would believe at Meta? Do you have any specific objections about what they wrote?
Looks like a little jab at OpenAI/Microsoft too.
Re: (Score:3)
Let's come back to this in about a year and see how Meta's belief system has fared.
Winamp! (Score:2)
Man, I am getting old...
Re: (Score:2)
I still install it on a new PC, and it still really kicks ass.
Does it really whips the Llama's ass? (Score:2)
That is the question that should be answered first of all.
I'm not a fan of the new code model (Score:2)
Gave the new 34b code model a spin and it seems quite a lot worse than my uncensored llama2 70b asking both the same questions. Not only does it not even seem to try constantly skipping over details attempting to reason with it is an exercise in futility.
Winamp: It really whips.... (Score:1)