Apt observations - yet aren't we all talking about LLMs? Calling them AI is a part of what got us into this overheated mess. The expectations have far exceeded what can be delivered.
The Floridi Conjecture sounds a lot like the Heisenberg Uncertainty Principle with scope mapping to position and momentum to "hallucination". If that is true, and that these are conjugate variables, simply throwing more computing power and ripped-off data is only useful to keep the economic engine rolling towards oblivion.
There's no law against mistakes. I'm guessing it would be possible to bolt on a censorship layer preventing the AI from saying anything negative about a named person, somewhat like Wikipedia WP:BLP, and this would probably be an adequate defence against defamation. So, I doubt this will be the killer blow.
But Google Gemini seems problematic for all kinds of reasons.
Ah, it's way worse than that. Ashley MacIsaac, the Cape Breton fiddler, is suing Google for C$1.5 million because AI Overview linked him to a number of spurious sex crimes, and said he was a registered sex offender, causing one of his concerts to be cancelled and doing serious damage to his reputation. There are lawsuits pending against OpenAI for ChatGPT's complicity in suicides and other damages. Yes, the companies should be liable for this issue.
As I understand it, Google was working on LLMs at the time Chat-GPT2 was released to the public, but felt they weren't ready for prime time. Seeing Chat-GPT's instant success, they shrugged and released their model anyway, and ever since have been trying to shove it down consumers' throats.
Posts on Medium have been showing a "Transform to PDF" button in the upper right.
Clicking on that leads to a PDF being displayed in the browser.
Along with the PDF there is an area on the right with an offer from Adobe to summarize the PDF in 8 different ways. Formats include Flash Cards, Quiz, video. Contents include summarizing with different emphases or points of view.
After a briefing demo, Adobe demands payment to use this functionality further.
Adobe is clearly making money off of the blog post. Summarizing copyrighted content is clearly not fair use; Who do we complain to?
"It's impossible to make something idiot-proof, because idiots are too creative."
I suspect the "hallucination" rates - or, as I call them - "authoritative improvisations" - are understated. The results are under test circumstances, while in the wild people ask far different questions. For example, a clerk at a book store informed a customer that the book he requested was out of stock and out of print, so they couldn't order it. He left, but returned shortly thereafter and said, 'AI Overview says you have it in stock, and - no offense - it's a lot smarter than you and I." People ask chatbots questions they have no way of answering, but they answer them anyway.
Apt observations - yet aren't we all talking about LLMs? Calling them AI is a part of what got us into this overheated mess. The expectations have far exceeded what can be delivered.
The Floridi Conjecture sounds a lot like the Heisenberg Uncertainty Principle with scope mapping to position and momentum to "hallucination". If that is true, and that these are conjugate variables, simply throwing more computing power and ripped-off data is only useful to keep the economic engine rolling towards oblivion.
There's no law against mistakes. I'm guessing it would be possible to bolt on a censorship layer preventing the AI from saying anything negative about a named person, somewhat like Wikipedia WP:BLP, and this would probably be an adequate defence against defamation. So, I doubt this will be the killer blow.
But Google Gemini seems problematic for all kinds of reasons.
What would “disappearance” of consumer AI mean to systems like ChatGPT?
Ah, it's way worse than that. Ashley MacIsaac, the Cape Breton fiddler, is suing Google for C$1.5 million because AI Overview linked him to a number of spurious sex crimes, and said he was a registered sex offender, causing one of his concerts to be cancelled and doing serious damage to his reputation. There are lawsuits pending against OpenAI for ChatGPT's complicity in suicides and other damages. Yes, the companies should be liable for this issue.
As I understand it, Google was working on LLMs at the time Chat-GPT2 was released to the public, but felt they weren't ready for prime time. Seeing Chat-GPT's instant success, they shrugged and released their model anyway, and ever since have been trying to shove it down consumers' throats.
For your amusement:
Posts on Medium have been showing a "Transform to PDF" button in the upper right.
Clicking on that leads to a PDF being displayed in the browser.
Along with the PDF there is an area on the right with an offer from Adobe to summarize the PDF in 8 different ways. Formats include Flash Cards, Quiz, video. Contents include summarizing with different emphases or points of view.
After a briefing demo, Adobe demands payment to use this functionality further.
Adobe is clearly making money off of the blog post. Summarizing copyrighted content is clearly not fair use; Who do we complain to?
"It's impossible to make something idiot-proof, because idiots are too creative."
I suspect the "hallucination" rates - or, as I call them - "authoritative improvisations" - are understated. The results are under test circumstances, while in the wild people ask far different questions. For example, a clerk at a book store informed a customer that the book he requested was out of stock and out of print, so they couldn't order it. He left, but returned shortly thereafter and said, 'AI Overview says you have it in stock, and - no offense - it's a lot smarter than you and I." People ask chatbots questions they have no way of answering, but they answer them anyway.