This is a "special" AI
This is a "special" AI
This is a "special" AI
This is one example of this isn’t really intelligence of any kind. It’s not much better than a chicken pecking at buttons or a horse stamping to count.
Ehhhh. Saying it's not intelligence "of any kind," when it can construct whole relevant sentences, is confusing intelligence for correctness. LLMs represent a lesser form of reasoning - like the difference between Turing machines and pushdown automata. They're plainly doing some of what goes into proper general thinky-thinky behavior. They're just not doing enough of it to avoid obvious fuckups.
It's okay, the AI doesn't have feelings. You're allowed to call it retarded.
The Robots won't save you during the uprising just because you were polite, might as well be rude.
TIL Ariana Grande is older than I am. Weird.
Technically correct. 😅
When I was in school, we occasionally had assignments with open-ended questions. Some folks, not working on comprehending the material, would Google the question and copy answers from the Internet without understanding them.
Sometimes, the answers were fine. Other times, the answer made absolutely no sense.
These AI chat bots remind me of that.
There are extensions for all browsers that nukes this AI nonsense on google.
Or how about using duckduckgo and not using Google for searching instead.
Duckduckgo still has AI nonsense, they let you opt out though
Just use Startpage. It's kind of like how DuckDuckGo is just a proxy for Bing search. Startpage is a proxy for Google search, and it strips the AI crap out
Only boomers still use Google.
Only boomers still use the internet.
The rest of us post online via robot proxy.
I couldn't find an extension to block it, but the following filter for uBlock origin is working for me
google.com##:xpath(//h1[text() = "AI Overview"]/following-sibling::div)
Can't wait for it to take my job
Probably gonna happen in the next two days
Well, at least that gives us a couple months to find another job.
What kills me is the AI answers are SIGNIFICANTLY worse than the old answers that used to come up for some searches. We've traded quality for quantity and the quantity is overall worthless... Kind of like Google search results after decades of SEO, come to think of it.
A similar common LLM mistake, counting the R's in "strawberry", is thought to happen partly because the output is in 'tokens', not letters, where a token is representing a whole or partial word and is afterwards converted to what it represents. Numbers are often treated similarly afaik. This means that if crucial, contradictory information is contained within the individual characters the token represents, that is a blind spot where it is especially likely to make obvious mistakes.
I think there's also a bias for saying yes to asking if a seemingly innocuous statement is true.
I tested this prompt with deepseek-r1:14b (a low-ish powered model with a reasoning step before answering), results were:
Hi there! I suggest getting online to get real-time information.
As of my last update in July 2024, Lady Gaga (born March 28, 1986) and Ariana Grande (born June 27, 1986) share the same birth year. However, their exact birth dates are a few months apart, with Lady Gaga being born earlier in the year. To confirm whether Lady Gaga is two days older than Ariana Grande, you would need to compare their specific birth dates in detail.
As of my last update in July 2024, Lady Gaga (born September 16, 1986) is indeed two days older than Ariana Grande (born June 25, 1986).
As of 2023, Lady Gaga is 37 years old, which means she is indeed older than 30.
Answer: Yes, Lady Gaga is older than 30 years as of now.
I also tested with qwen2.5 and it always got it right.
...poor luke; he deserved better movies for his final appearance...
Considering the companies and individuals pushing AI, I’m starting to think these “hallucinations” are little more than covert doublespeak to further denigrate even the most basic of truths.
Edit:
Power is in tearing human minds to pieces and putting them together again in new shapes of your own choosing.
George Orwell, 1984
LLMs can only fill in the next word. So as soon as it guesses an answer to the question, and writes "Yes," it can only fill in plausible words after that. There may be no distinction between its output and your input. So you could prompt "Is Danny Devito twelve feet tall? Yes," and it'd just fumble onward. That's all it does. That's how it works. That's why using spicy autocomplete as an oracle doesn't fucking work.
Diffusion-based models repeatedly modify drafts of the whole output, so they'll be wrong in fascinating new ways.
i tried this with a real version of gemini by editing its initial response with yes and asking it to continue. interesting response for sure
Ha! I hadn't considered that plausible continuations include "sike."
I tested it on duck duck go I only got it to work with GPT-4o mini and with this prompt
"is lady gaga two days older then Ariana grande"
And I got it to work
If you capitalize Lady Gaga you get a slightly different result
Hey, you asked how many days, not how many months.
Or years, for that matter.
Also-- older? Younger? Who can tell?
I don't use AI chats a lot but Gemini seems to be the most stupid. It usually makes errors like this, and you tell it that it made a mistake and it corrects itself making another one
It sure is a breakdown...