this post was submitted on 17 Oct 2023
22 points (80.6% liked)
Technology
71922 readers
3809 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Last time they hyped up their AI it was really inferior to ChatGPT. Lots of companies are claiming to be on par with GPT-4 and still end up having terrible reasoning, so I doubt it.
Think there's room for more models anyway. Sometimes accuracy isn't as important as running costs and such. When proofing translations I've found that GPT3.5 works better for me than GPT4.
OP isn’t saying there isn’t room. He’s simply stating that non-Chinese companies claim to rival GPT4 which fall through.
Knowing this is heavily China backed, there is going to be a ton of misinformation related to it. Let alone lies about how smart it is.
So, over/under on how long it takes someone to convince it to talk about Tiananmen Square and the CCP nukes it from orbit? Bonus points if you can get it to say Taiwan is an independent country.
I'd be surprised if this was possible without explicitly describing those topics in your prompt. I imagine they carefully scrubbed its training data for non-approved subject matter.
It'll be interesting to see what happens when China decides they need to ban something new, though, like Winnie the Pooh. You can't easily remove training data after the fact, so in that case they'd probably band-aid it and you could probably work around that.