🏄Snapchat AI went wild

Big tech pays up to $900,000 to AI experts, AI girlfriends might ruin human relationships

Author

Marc Csuzi
August 16, 2023

Happy Wednesday Surfers🏄! 

Snapchat’s My AI went rogue yesterday and posted a video on its story of a wall or ceiling. People went bananas when the AI ignored questions about it, something the chatbot had never done before.

So what do you think? Did the machine gain consciousness and tried to escape or was it just a bug as Snapchat later claimed?

I’ll stick with the bug explanation for now but will keep an eye on Snapchat AI.

ONE NEWS PIECE

🧙‍♂️ChatGPT tries hiding it was trained on Harry Potter and other copyrighted books

Researchers at ByteDance, TikTok’s Chinese parent company, found that ChatGPT now attempts to avoid responding to user prompts with exact phrasing from copyrighted works.

The creators of the most popular LLMs so far failed to disclose what data they trained their models on, but it’s pretty obvious it wasn’t just Wikipedia articles. With ongoing and lingering lawsuits from IP owners and artists OpenAI is apparently trying to mitigate the damage with this new tweak.

“It disrupts the outputs when one tries to continuously extract the next sentence… which did not happen in the previous version of ChatGPT” - the research paper states.

Despite this, the researchers were able to make ChatGPT as well as Google’s and Meta’s LLM respond to several prompts based on the Harry Potter books with phrases and work matching the original work exactly or almost exactly.

ONE IDEA

“So suppose you had LLMs in the 1600s and Galileo comes up for trial. You ask the LLM: Is Galileo right? What does it answer?”

“And one theory is he answers: No, because the overwhelming majority of human thought up until that point was that he was wrong. And so therefore that's what's in the training data.

Another way of thinking about it is a sufficiently advanced LLM will have evolved the ability to actually check the math. And will say he's right.

Now if the Church at that time owned the LLM, they would've given it human feedback to prohibit it from answering that question.

Same questions apply … with the human feedback training that's actually happening with these LLMs today.”

-Listen to internet OG Marc Andreesen discuss this on the Lex Fridman Podcast at 00:25:24

⌚ If you have one more minute:

  • Big tech companies are paying salaries as high as $900,000 to attract generative AI talent

  • AI girlfriends might ruin human relationships according to experts

  • Saudi Arabia and UAE invest in thousands of Nvidia chips, aiming to become global AI powerhouses

AI Art of the day 🎨

Happy Toys

This very creepy but convincing Happy Toys commercial by Youtube channel vvv—okay.

🌊🏄🌊🏄🌊🏄🌊🏄🌊🏄🌊🏄🌊🏄🌊🏄🌊🏄🌊🏄🌊🏄🌊🏄

That’s it folks!

If you liked it please share it with a friend and make this writer happy that ChatGPT didn’t replace him yet!

Or did it?