Archive Link: https://web.archive.org/web/20240330224149/https://www.theregister.com/2024/03/28/ai_bots_hallucinate_software_packages/
This is fascinating. I’ve certainly seen AI hallucinating things like imaginary functions in gdscript. Admittedly, it does it a lot more with gpt3 than with gpt4 on a subscription, which is consistent with what 3 vs 4 has access to, but I’m sure the problems apply in a lot of other use cases that might have not had the benefit of more recent documentation.
I suppose it’s not surprising that a number of larger entities have been falling prey to this, as they keep trying to inappropriately jam AI into their production lines where it’s incapable of doing the job. Pretty clever vulnerability to find, though.
Ultimately, this is probably a good thing for human coders, imo. The more LLMs demonstrate that they’re not effective without robust human intervention, the better.
15000 downloads - that’s messed up. This backs my opinion that inflationary usage of those LLMs do more harm than good atm. To copy and paste their answers, without questioning them, is like giving up your own critical thinking to rely on a highly gifted and eloquent LSD addict as your guide.
So… there will be organizations that will train devs for this, and that will outright ban LLMs. I know which mine will be. Time to reconsider my job, and possibly my place of work…
I think there will be a market for “corporate compliant and secure!!” LLMs. “Pay us gobs of money so you don’t get ‘hacked’ by dumb LLM users”
I had a google summary telling me how to use a lodash (js object/array/things helper library) method that sounded like it probably would exist.
It was named how lodash would likely name it, and was summarised to do what i needed.
Except, lodash doesnt have that method. Had to use a couple methods.But that was eye opening for me.
Similar to lawyers citing cases that dont exist.
Saw a meme-ish post recently from an IBM presentation 30-40 years ago along the lines of “computers cannot be held accountable. So dont have them make decisions”.🤖 I’m a bot that provides automatic summaries for articles:
Click here to see the summary
In-depth Several big businesses have published source code that incorporates a software package previously hallucinated by generative AI.
Not only that but someone, having spotted this reoccurring hallucination, had turned that made-up dependency into a real one, which was subsequently downloaded and installed thousands of times by developers as a result of the AI’s bad advice, we’ve learned.
He created huggingface-cli in December after seeing it repeatedly hallucinated by generative AI; by February this year, Alibaba was referring to it in GraphTranslator’s README instructions rather than the real Hugging Face CLI tool.
Last year, through security firm Vulcan Cyber, Lanyado published research detailing how one might pose a coding question to an AI model like ChatGPT and receive an answer that recommends the use of a software library, package, or framework that doesn’t exist.
The willingness of AI models to confidently cite non-existent court cases is now well known and has caused no small amount of embarrassment among attorneys unaware of this tendency.
As Lanyado noted previously, a miscreant might use an AI-invented name for a malicious package uploaded to some repository in the hope others might download the malware.
Saved 83% of original text.