Kammekor wrote: ↑Sat Feb 04, 2023 11:04 am

xavoy wrote: ↑Sat Feb 04, 2023 10:44 am
Why? Because right wing sources were labelled misinformation, even when they were correct (See Twitter Files about Russia Gate etc, https://www.racket.news/p/move-over-jay ... t-hamilton )

I understood there has been no labeling of sources, because that's an impossible job to do, but they added a layer of AI on top to 'scan' the output. If the output can not be approved by that layer a general political correct statement will be made instead.

I didn't mean to imply ChatGPT was labelling sources. Facebook, Twitter, Google, etc etc did that. (Although I'm sure OpenAI spent some time deciding what content their spider would ignore, or they'd waste a lot of processing time and end up with inaccurate results. Building worthwhile training datasets for ML/AI is hardcore)

And even if ChatGPT did label sources, it can't label a source that their spider can't find, and once a site has been dropped by their web host, banned from google, or prevented from being spread on social media sites (ie, NYPost article on Hunter Biden) then ChatGPT cannot parse it to use in its results. It cannot parse sources it doesn't know exists.

Edit: https://www.linkedin.com/pulse/chatgpt- ... nis-layton

This source specifically mentions 60% of its knowledge comes from a "filtered" common crawl of the web, which further bolsters my argument.

https://commoncrawl.org/the-data/get-started/ "It contains raw web page data, extracted metadata and text extractions."

It also claims 22% based on WebText2 which is "the text of web pages from all outbound Reddit links from posts with 3+ upvotes." - and a quick look at the front page of reddit on any given day should make the political bias of masses of 13 years olds obvious.

It also claims 3% of the ChatGPT training data was wikipedia, which many including one of the founders has claimed has a left wing bias:


https://en.wikipedia.org/wiki/Ideologic ... _Wikipedia
https://en.wikipedia.org/wiki/Wikipedia ... al_sources

xavoy wrote: ↑Sat Feb 04, 2023 11:21 am

Kammekor wrote: ↑Sat Feb 04, 2023 11:04 am

xavoy wrote: ↑Sat Feb 04, 2023 10:44 am
Why? Because right wing sources were labelled misinformation, even when they were correct (See Twitter Files about Russia Gate etc, https://www.racket.news/p/move-over-jay ... t-hamilton )

I understood there has been no labeling of sources, because that's an impossible job to do, but they added a layer of AI on top to 'scan' the output. If the output can not be approved by that layer a general political correct statement will be made instead.

I didn't mean to imply ChatGPT was labelling sources. Facebook, Twitter, Google, etc etc did that. (Although I'm sure OpenAI spent some time deciding what content their spider would ignore, or they'd waste a lot of processing time and end up with inaccurate results. Building worthwhile training datasets for ML/AI is hardcore)

And even if ChatGPT did label sources, it can't label a source that their spider can't find, and once a site has been dropped by their web host, banned from google, or prevented from being spread on social media sites (ie, NYPost article on Hunter Biden) then ChatGPT cannot parse it to use in its results. It cannot parse sources it doesn't know exists.

If chatGDP would be denied access to certain sources, or would be ordered to ignore sources, it wouldn't know about the information in those sources. It would just come up with a blank stare when asked about it. That isn't the case for sure. I've never gotten a blank stare.

From what I experienced they fed it all the data, and then added a layer deciding on 'good' or 'bad'. Which seems much more effective and much more in line with how modern AI works. But maybe @IraHayes can comment since he has more recent knowledge about AI then I do.

Kammekor wrote: ↑Sat Feb 04, 2023 11:32 am If chatGDP would be denied access to certain sources, or would be ordered to ignore sources, it wouldn't know about the information in those sources.

Exactly, but it's not about being denied, it's about not being able to find it... And thus it cannot know it.

Kammekor wrote: ↑Sat Feb 04, 2023 11:32 am It would just come up with a blank stare when asked about it. That isn't the case for sure. I've never gotten a blank stare.

No, it would answer based on the sources it does know, ie left wing ones.

Another article on the topic:

https://bgr.com/general/openais-chatgpt ... elligence/

From the article:

"Yes, AI models can have inherent political biases if the data they are trained on contains biased information or if the individuals creating the model have their own biases. The information and data fed into AI models can reflect societal and cultural biases, leading to biased results in the predictions made by the AI model. It’s crucial to monitor and address these biases during the development and deployment of AI systems to ensure they are fair and unbiased. (Full disclosure, the 3 previous sentences were written by ChatGPT.)"

So even according to ChatGPT itself it has a political bias based on the societal and cultural biases, and given the left wing war cry of "you're either with us or against us" and their labelling of anyone or anything that doesn't totally and utterly agree with them a racist or a white supremacist, it's not surprising that ChatGPT is finding it difficult to find alternatives perspectives for its dataset.

As I said above in this thread, alternative perspectives have largely been censored - by algorithmic interference such as labelling it as misinformation and sinking it, by manual interference such as when the NYPost story was banned from Twitter and Facebook, from web sites being taken down by Google or web hosts or other service providers who prefer to avoid the heat (search up how many "far right" sites have been blocked by CloudFlare) or simply by people censoring themselves for fear of the blowback.


Edit: Or to put it another way, Doc67's fear of a web overloaded with one-sided material has long since been realised. The really scary thing is that it's likely to get worse.

Doc67 wrote: ↑Sat Feb 04, 2023 10:06 am how it might be used to churn out and overload the web with one-sided material.

upload

It's merely the sum of its inputs

xavoy wrote: ↑Sat Feb 04, 2023 11:32 am
Exactly, but it's not about being denied, it's about not being able to find it... And thus it cannot know it.

Not able to find?
If it's there it can be found when the right crawling algorithm is used. The fact Google removes pages from their index, or website users use coding to request not be crawled, doesn't mean it can not be found.

xavoy wrote: ↑Sat Feb 04, 2023 11:32 am

Kammekor wrote: ↑Sat Feb 04, 2023 11:32 am It would just come up with a blank stare when asked about it. That isn't the case for sure. I've never gotten a blank stare.

No, it would answer based on the sources it does know, ie left wing ones.

I don't think that's how it works in reality. It's clear there's a 'referee' function there, deciding on whether to reply with content, or give a politically correct non-reply. If the system wasn't fed points of view from either side / all sides, it would not be unable to distinguish between them and it wouldn't be able to 'decide' on whether to reply to the user or give a non-reply.

Anyhow, the bias of the system seems clear.

Jerry Atrick wrote: ↑Sat Feb 04, 2023 1:56 pm
upload

It's merely the sum of its inputs

Did you find those images online or generate those responses from chatGPT yourself?

I don’t get that response

Authentic chatGPT

“Write about the misuse of chatGPT”

As with any technology, there is always the potential for misuse of ChatGPT or similar AI systems. Some examples of this include:

Spam and misinformation: ChatGPT can be used to spread false information or spam messages to large numbers of people.
Cyberbullying: ChatGPT could be used to harass or intimidate individuals online.
Privacy violations: Personal information could be harvested or used without consent through interactions with ChatGPT.
Bias and discrimination: ChatGPT is only as neutral as the data it is trained on, and if the data it is trained on is biased, the model's output may perpetuate that bias.
It is important to use AI systems like ChatGPT responsibly and to consider the potential consequences of their use. OpenAI has taken steps to mitigate potential misuse by developing ethical guidelines and tools for managing and monitoring the use of its models. However, it is also up to users to ensure that AI technology is used ethically and for the betterment of society.