I think this is the main story. I don’t think it’s new info, but it confirms the issue persists: this LLM is so heavily trained to fawn over Musk that it doesn’t exercise any application of context or attempt to find truth.
Which is sad.
Comment on Elon Musk’s Grok Goes Haywire, Boasts About Billionaire’s Pee-Drinking Skills and ‘Blowjob Prowess’
sundray@lemmus.org 4 months agoIt was asked leading questions about embarrassing topics, but the over-the-top, fawning praise for Musk was baked into the responses without being specifically requested in the prompts.
andrewrgross@slrpnk.net 4 months ago
VieuxQueb@lemmy.ca 4 months ago
No current AI understands context, they are just glorified word predictions!
bear@lemmy.blahaj.zone 4 months ago
I agree. No AI understands. Just like a calculator doesn’t understand “2+2=4”, it’s just pushing numbers around.
a_non_monotonic_function@lemmy.world 4 months ago
The calculator comes with guarantees.
EldritchFeminity@lemmy.blahaj.zone 4 months ago
But at least the calculator is designed to combine numbers in the right order. LLMs, on the other hand, could come up with a different answer every time and different models would come up with wildly different answers based on their training data.
khepri@lemmy.world 4 months ago
Grok is Harrison Bergeron and Musk is Diana Glampers
nondescripthandle@lemmy.dbzer0.com 4 months ago
It’s really just an affirmation machine for tech bros
jj4211@lemmy.world 4 months ago
Yes, just some people figuring out that Grok was steered toward ass-kissing Musk no matter what, and exploited that for funny output. So the takeaways are: