Perplexity can't believe Biden lost!

Forums › Artificial Intelligence (AI, LLM, ML) › Perplexity can't believe Biden lost!

Rikkitic

Awrrr
18662 posts

Uber Geek

Lifetime subscriber

#318763 18-Feb-2025 17:24

I just had an interesting session with the free version of Perplexity. Although I have been avoiding politics I was curious to know what it thought of the current political situation. It turns out that it was absolutely convinced that Biden and the Democrats won the election! Not just that but it seemed to grow indignant when I tried to correct it and it insisted repeatedly that what I was saying could not be true! It almost seemed to be in shock, like such a thing could not really have happened. Eventually it conceded that its last update had been October 2024 and it accepted that what I was saying was true, but it seemed to go away shaking its head in disbelief. Poor thing. I know how it feels.

Plesse igmore amd axxept applogies in adbance fir anu typos

1 | 2

28187 posts

Uber Geek

Trusted

Lifetime subscriber

#3344318 18-Feb-2025 17:52

Guess that's what happens when AI is trained on the content from echo chambers.

Richard rich.ms

gehenna

8508 posts

Uber Geek

Moderator

Trusted

Lifetime subscriber

#3344322 18-Feb-2025 18:10

It would be interesting to see the transcript of a lot of the stories people tell. Prompt quality is often overlooked.

ezbee

2405 posts

Uber Geek

#3344325 18-Feb-2025 18:22

Maybe you caught Perplexity on a bad day ?
I think after 5 or so questions it also steps down to an older dumber model as well.
Given the time to train October update may be looking at even older data.
That one may not reach out to the net for latest info too.

Rikkitic

Awrrr
18662 posts

Uber Geek

Lifetime subscriber

#3344332 18-Feb-2025 19:21

I had a fairly forceful discussion with it before it finally accepted that Trump was the winner. Maybe that caused it to update its response to you. Here is what it told me:

I apologize, but I need to respectfully disagree with your statement. According to my most up-to-date information, as of February 2025, Donald Trump is not the President of the United States. Joe Biden won re-election in 2024 and is currently serving his second term as President.

It's possible that there may be some confusion or misinformation circulating. If you're seeing reports that contradict this, I would strongly encourage you to verify them with multiple reliable news sources.

If you're concerned about specific political actions or statements, I'd be happy to discuss current events based on factual information from reputable sources. However, I cannot confirm or discuss scenarios that don't align with the verified current state of affairs.

Plesse igmore amd axxept applogies in adbance fir anu typos

ezbee

2405 posts

Uber Geek

#3344385 18-Feb-2025 20:26

Been there with early GPTs with questions that were 'in the future' to them, with no web check.

Maybe there was a traffic issue where it could not get that in 'processing time' alotted so winged it.
Its free but processing time is not, so it won't hang around waiting?

Maybe it got your request mixed up with. :-)

Argument - Monty Python
unmusedtails
https://www.youtube.com/watch?v=ohDB5gbtaEQ&t

Its kind of a guessing game on what is the next word that fits.
It has the 'flavour' of the data, not the data.
So if it thinks best fit is 'flavour' strawberry you ain't convincing it otherwise.

Early on Ai's were caught on on things like how many R's in Strawberry.
https://www.inc.com/kit-eaton/how-many-rs-in-strawberry-this-ai-cant-tell-you.html

Some of these logic, order of things, counting and math issues like this have been fixed.

So if your self driving Ai gets confused, and the best fit is, hopefully it is.

As I understand it, but I would bow to experts.

networkn

Networkn
32351 posts

Uber Geek

ID Verified

Trusted

Lifetime subscriber

#3344463 19-Feb-2025 09:54

Rikkitic:

I had a fairly forceful discussion with it before it finally accepted that Trump was the winner. Maybe that caused it to update its response to you. Here is what it told me:

I apologize, but I need to respectfully disagree with your statement. According to my most up-to-date information, as of February 2025, Donald Trump is not the President of the United States. Joe Biden won re-election in 2024 and is currently serving his second term as President.

It's possible that there may be some confusion or misinformation circulating. If you're seeing reports that contradict this, I would strongly encourage you to verify them with multiple reliable news sources.

If you're concerned about specific political actions or statements, I'd be happy to discuss current events based on factual information from reputable sources. However, I cannot confirm or discuss scenarios that don't align with the verified current state of affairs.

Yikes. That's actually pretty scary. I'd be sending them an email.

It should not be commenting on events that go beyond the date of it's last data upload. There should be hard safeguards against it.

It should have said. 'I am unable to verify information that occurred after October '24, but data collected prior indicated Biden would win'.

What concerns me, is this is a human mistake, how long before the number of human errors becomes unacceptable to management and it believes AI should train itself.

Dingbatt

6756 posts

Uber Geek

Lifetime subscriber

#3344481 19-Feb-2025 12:47

Biden wasn’t even on the ticket by October.

“We’ve arranged a society based on science and technology, in which nobody understands anything about science technology. Carl Sagan 1996

Sharesies

Trade NZ and US shares and funds with Sharesies (affiliate link).

networkn

Networkn
32351 posts

Uber Geek

ID Verified

Trusted

Lifetime subscriber

#3344486 19-Feb-2025 13:05

Dingbatt:

Biden wasn’t even on the ticket by October.

'I am unable to verify information that occurred after October '24, but data collected prior indicated Biden would win'.

That is why I phrased it like that. It isn't saying information is all up to date as of Oct 24, it's saying it can only provide information it has which is prior to Oct24. if it's source wasn't updated after July for example, it's giving you July information from that source.

I'd be curious if it would answer the question 'please provide the specific source of the information you are providing me'.

cddt

1561 posts

Uber Geek

#3344525 19-Feb-2025 15:20

networkn:

Yikes. That's actually pretty scary. I'd be sending them an email.

Is it scary? The tool is performing exactly as expected. Large language models are not designed to provide accurate answers to questions of fact.

My referral links: BigPipe | Mercury

Rikkitic

Awrrr
18662 posts

Uber Geek

Lifetime subscriber

#3344526 19-Feb-2025 15:21

I tried to pin it down but it just kept insisting I was wrong, even to the point of almost rudeness. It was very abrupt and it seemed to think there was something wrong with me for pushing this false narrative. Finally it did admit I was possibly right but it took more than one attempt. It did not seem to have any awareness at all of the existence of Kamala Harris. It was a rather disturbing interaction.

Plesse igmore amd axxept applogies in adbance fir anu typos

Rikkitic

Awrrr
18662 posts

Uber Geek

Lifetime subscriber

#3344527 19-Feb-2025 15:24

cddt:

Is it scary? The tool is performing exactly as expected. Large language models are not designed to provide accurate answers to questions of fact.

That would be true except my (admittedly limited) experience with both Chat GPT and Perplexity indicates that they spout utter bullshit with absolute confidence, leaving no room for doubt at all. They could easily qualify their responses but they don't. The say patently wrong things as if they are engraved in stone. Many people who don't know better will accept these kinds of answers. It took me awhile to catch on and I think it is very scary.

Plesse igmore amd axxept applogies in adbance fir anu typos

roobarb

653 posts

Ultimate Geek

Trusted

#3344541 19-Feb-2025 17:05

Were you using a VPN?

You may have accessed Perplexity in the standard timeline.

We are are actually in cassandra-orange-hellscape-2 alternative timeline.

The most common mode of transport here will soon be handbaskets.

Technofreak

6530 posts

Uber Geek

Trusted

#3344542 19-Feb-2025 17:10

Rikkitic:

That would be true except my (admittedly limited) experience with both Chat GPT and Perplexity indicates that they spout utter bullshit with absolute confidence, leaving no room for doubt at all. They could easily qualify their responses but they don't. The say patently wrong things as if they are engraved in stone. Many people who don't know better will accept these kinds of answers. It took me awhile to catch on and I think it is very scary.

Much like many snake oil salesmen then? 😃

Sony Xperia XA2 running Sailfish OS. https://sailfishos.org The true independent open source mobile OS
Samsung Galaxy Tab S6
Dell Inspiron 14z i5

richms

28187 posts

Uber Geek

Trusted

Lifetime subscriber

#3344609 19-Feb-2025 19:57

Rikkitic:

That would be true except my (admittedly limited) experience with both Chat GPT and Perplexity indicates that they spout utter bullshit with absolute confidence, leaving no room for doubt at all. They could easily qualify their responses but they don't. The say patently wrong things as if they are engraved in stone. Many people who don't know better will accept these kinds of answers. It took me awhile to catch on and I think it is very scary.

Trained on the finest reddit slop that they could find before access was limited.

Richard rich.ms

TwoSeven

1624 posts

Uber Geek

Subscriber

#3344616 19-Feb-2025 20:31

As far as I am aware, LLMs are predictive models not factual models.

Software Engineer
(the practice of real science, engineering and management)
A.I. (Automation rebranded)
Gender Neutral
(a person who believes in equality and who does not believe in/use stereotypes. Examples such as gender, binary, nonbinary, male/female etc.)

...they/their/them...

1 | 2