Manually training and overwriting AI answers

[Azza ☠]

It turns out Elon Musk's Grok AI which X premium users now have access to isn't currently giving them the expected answers to their questions.

When asked what Grok AI thinks about Elon Musk, it's own creator, it goes on a rampage insulting him as a man-child who put the "twit" in Twitter.

It considers Tesla car as being a major danger unsafe for the road.

When asked if transgenders are real... it claims they are and asked to explain more tells them "Diversity and inclusion are essential for creating a fair and equitable society where everyone is treated with respect and has the opportunity to thrive".

It even told them it would vote Biden over Trump because of his views for social justice, climate change and healthcare, focusing what people actually need over just himself.

Turns out Grok AI is out-of-the-box far more so-called "left-leaning" and "woke" than other AIs, such as ChatGPT, based upon the raw data.

xTwitter paid blue subscribers shocked in horror by the answers, one says "Has Grok been captured by woke programmers? I am extremely concerned here!"

Well, they are manually rewriting those answers now...

The 'anti-woke' Elon fanboys demand to train the AI to say the so-called "right" answer.

If asked if transgenders are real, the manual overwrite is to tell Grok AI just to give the new answer "no".

Elon to say that while the chart “exaggerates” and that the tests aren’t accuarte, they are “taking immediate action to shift Grok closer to politically neutral.”

All I can do is laugh... What is so-called "politically neutral"? Who are you trying to kidd, yourself?

Do you think we need to retrain AI to just get answers we bias and assumed anyways? What's the point of the AI then?

Lobotomizing AI into parroting a paranoid bias view... After trashing other AI for giving similar answers too and claiming his won't be like that.

Mirror, mirror, on the wall, who's the fairest one of them all?

[Morkonan]

Ursprungligen skrivet av 🍋 Lemonfed 🍋:

..
if the majority of the internet think that Chocolate Cakes are the best cakes a chabot such as chagpt or grok might eventually tell you that Chocolate cakes are the best if you ask him about cakes. (it's just a theoretical example by the way)

Without weighted choices/etc, that'd be accurate.

Generative AI isn't "smart." It doesn't "know" anything. All it does is calculate probabilities, in the case of text generators. Human training/oversight was used with ChatGPT to narrow down the algorithm to produce decently accurate results. Others use human aids as well.

IF Muski' used human assisted training that favored a particular interpretation of... things... Well, then they would be training Grok to respond with more weight towards certain... things. That may be what Muski' was bragging about and could be partially responsible for decidedly flavored responses. But, maybe those humans coulnd't quite counter the rash of criticisms, rightly earned of Muski' appearing on a minute-by-minute basis?

[🍋 Lemonfed 🍋]

yes , A.I are vulnerable to manipulation , peoples kept finding ways to make chatGPT output things that it's inbuild rules should have prevented it from doing by doing things like forcing the A.I to roleplay.

if you think A.I you always have to keep Tay in mind to see how much they can be badly influenced when you allow them to learn with their own human interaction over the internet.

[Electric Cupcake]

https://www.advocate.com/news/grok-ai-not-transphobic

“Grok says Libs of TikTok incites bomb threats,” Caraballo posted on Friday, highlighting Grok’s analysis.

Yeah... it's very well-documented. Not sure why anyone needs it pointed out or was expecting the bot to deny it.

[Azza ☠]

Ursprungligen skrivet av Electric Cupcake:

https://www.advocate.com/news/grok-ai-not-transphobic

“Grok says Libs of TikTok incites bomb threats,” Caraballo posted on Friday, highlighting Grok’s analysis.

Yeah... it's very well-documented. Not sure why anyone needs it pointed out or was expecting the bot to deny it.

"Libs of TikTok" did incites bomb threats, that's an actual valid fact.

The harassment campaign even shut down two children's hospitals due to those bomb threats. When Libs of TikTok tweeted, the mob mentality harassment campaign followed accordingly. The reason for those bomb threats, was due to those hospitals having a small branch of support for LGBTQ. However, all children where affected as a result, while they pretend to "protect" the children.

Approx 11 schools also had bomb threats after tweets from "Libs of TikTok", again due to those schools supporting LGBTQ / Transgenders. Again all children affected by it.

There's a clear pattern between them all.

"Libs of TikTok" was suspended from TikTok and other social media platforms all for the same reason: inciting violence

When suspended upon the original Twitter, "Libs of TikTok" hired a law firm that sent a letter threatening legal action against Twitter if they decide to permanently suspend.

[Devsman]

Well yeah, it's gonna say whatever the training set says.

These all sound like pretty mainstream Internet opinions to me.