AI don't trust techbros
- Thread starter NovaSaber
- Start date
ChatGPT provides false information about people, and OpenAI can’t correct it
noyb today filed a complaint against the ChatGPT maker OpenAI with the Austrian DPA
noyb.eu
NOYB at first glance appears to basically be a European version of the EFF, focused specifically on GDPR application. Not sure what will come of this, but that is something that hadn't even crossed my mind before - it's basically impossible wiht the current implementation to meet GDPR criteria.
With their "Train on the internet" model and no tracking on that, and AFAIK no way to "un-learn" something without basically going to a previous snapshot of sorts and re-training minus the problematic source(which is a fair impossibility at this point for almost everything that has already been fed into it) there's basically is no way for it to be GDPR compliant. They could maybe re-train it on specific things, but the incorrect associations are basically there forever; even if it's a tiny miniscule chance of those tokens being associated, it's still never going to be zero if they were associated at least once. You could maybe "program" it further with a prompt to avoid that specific association, but if you ahve to keep adding those on top of the base model for every person in the EU who requests a correction, it's going to get unweildly pretty fast IMO. On top of that it's still there if anyone were to acquire a copy of the model without those prompts added.
Then we also come to the fact OpenAI even admits it's entirely just word prediction, not a factual database, so given the previous stipulations, effectively using it as a database may be straight up illegal in EU land due to the requirement for correctness in the GDPR, since that can't be guaranteed.
So it turns out that even AI is susceptible to misinformation. Go figure.
More legal action rolling against the AI-makers:
apnews.com
In addition to a couple from 2023:
While Stack Overflow is explicitly handing its data over to OpenAI(rather than them just stealth-scraping it):
techcrunch.com
Eight US newspapers sue ChatGPT-maker OpenAI and Microsoft for copyright infringement
A group of eight U.S. newspapers is suing ChatGPT-maker OpenAI and Microsoft, alleging that the technology companies have been “purloining millions” of copyrighted news articles without permission or payment to train their artificial intelligence chatbots.
apnews.com
In addition to a couple from 2023:
While Stack Overflow is explicitly handing its data over to OpenAI(rather than them just stealth-scraping it):
Stack Overflow signs deal with OpenAI to supply data to its models | TechCrunch
Stack Overflow, the developer Q&A site, has inked a licensing agreement with OpenAI to supply data to its models.
techcrunch.com