5 edits
ChatGPT gave a response regarding nudity in paintings vs. nudity in other media. Its reply seemed like it could've been from a poster on this site.
The AI said:
In paintings, it’s often interpreted as symbolic, anatomical study, or storytelling.
In other media (like photography, film, or online content), it’s more often assumed to be sexual or commercial—even when that’s not always true.
What’s interesting is how automatic it is. With something like a Renaissance painting, viewers are almost “trained” to read it through a cultural lens—history, symbolism, technique—before jumping to sexuality. But with modern media, especially photography or anything online, the default lens flips: people often assume sexuality first, and only later (if at all) consider artistic intent.
@vivify saidIt will tell you bald faced lies with the same convincing tone. I spent hours trying to find a character in a game because ChatGPT told me he was in a certain town. I asked again and it told me a totally different town. It makes stuff up when it can’t find a good answer online.
This level of thoughtfulness from a computer is insane.
1 edit
@AThousandYoung saidMaybe in 2 or 3 years of adjustments and upgrades, that will no longer be an issue.
It will tell you bald faced lies with the same convincing tone. I spent hours trying to find a character in a game because ChatGPT told me he was in a certain town. I asked again and it told me a totally different town. It makes stuff up when it can’t find a good answer online.
I think ChatGPT Is the most revolutionary online tool since Google.
1 edit
@vivify saidWhen you ask about Renaissance paintings, it has hundreds of books and articles online to draw from. Of course it can deliver articulate responses.
Maybe in 2 or 3 years of adjustments and upgrades, that will no longer be an issue.
I think ChatGPT Is the most revolutionary online tool since Google.
But the hallucinations have gotten worse over time. This is because it predicts text, not facts, and the more steps involved in that prediction can lead to compounding errors on each step. One of the biggest issues I think is that it will always struggle with new information that has limited source content. It's not "thinking" but drawing on sentences in various different texts and reddit threads. If a crazy Reddit post says something completely wrong, but also sort of unique and quirky, I don't see how ChatGPT can purge that information selectively from its algorithmic output that a regular old human with a brain would obviously note as wrong.
Another issue is that it's easy to get into responses where it's just trying to output what it thinks you want to know. You can't get it to respond with "I don't know then answer". A lot depends on the prompts, and in many cases it doesn't matter, but for professionals using it has been quite problematic and appears unfixable.
2 edits
@wildgrass saidUsing ChatGPT should be like Googling something or using Wikipedia. They're good for pointing you in the right direction but you should check everything provided with legit sources. In fact, you can even ask ChatGPT for links to sources it pulls from.
When you ask about Renaissance paintings, it has hundreds of books and articles online to draw from. Of course it can deliver articulate responses.
But the hallucinations have gotten worse over time. This is because it predicts text, not facts, and the more steps involved in that prediction can lead to compounding errors on each step. One of the biggest issues I think is ...[text shortened]... it doesn't matter, but for professionals using it has been quite problematic and appears unfixable.
When's the last time you used it? If it's been a year or two, I'd recommend trying it out again. It's always been on point for me, and I expect it to only get better over time, as the technology improves and the team behind it continues tweaking and refining the tool
Mark my words: in the same way people now say "just Google it", people will start saying "just ChatGPT it".
@vivify saidGoogle's Gemini, and Anthropic's Claude are similarly amazing, and they are improving very rapidly. I would encourage anyone who can to pony up $20 a month to get access to the latest versions of these AI platforms and just give them a spin. Ask them to help you with something you might think is beyond their capability. I lean toward Claude because it's the first AI to tell me "I don't know." But if you want an AI that can access real time data, like "Tell me why ticker symbol XYZ is getting hammered in after market trading today," the free Google Gemini (the AI Mode button at google.com) is very handy. It's also useful to know that you can set parameters for how they interact with you. You can tell them to be more succinct, not try to validate you, not provide lengthy summaries, etc. which can overwhelm you with their output.
Maybe in 2 or 3 years of adjustments and upgrades, that will no longer be an issue.
I think ChatGPT Is the most revolutionary online tool since Google.
@Ponderable saidI doubt that. The response it gives, like in the OP, is far too nuanced. The AI seems programmed to sort information in a particular way for accuracy and context.
It is just giving you a stistical relevant mixture of human quotes...
@Sleepyguy saidI use my Alexa in the same way. I have one in my kitchen and it's more than just an assistant; you can actually tell Alexa if she misunderstood you and she will acknowledge and update her responses and processes in real time.
Google's Gemini, and Anthropic's Claude are similarly amazing, and they are improving very rapidly. I would encourage anyone who can to pony up $20 a month to get access to the latest versions of these AI platforms and just give them a spin. Ask them to help you with something you might think is beyond their capability. I lean toward Claude because it's the first AI to tell ...[text shortened]... try to validate you, not provide lengthy summaries, etc. which can overwhelm you with their output.
However: I don't think you should have to tell an AI try not to "validate you". This should be automatic. AI is worthless if just tells you want you want to hear.
The exception, of course, are for apps specifically designed to be a "friend" or whatever to the user. But for any app that's meant to provide useful information, it should not ever tailor facts to suit personal preferences.
1 edit
@Ponderable saidThis. It’s like a teachers pet that can repeat all the pet phrases on demand, that if you didn’t know might think the teachers pet actually had some insight.
It is just giving you a stistical relevant mixture of human quotes...
1 edit
@vivify saidPeople already say that. I think there's a mis-appropriated level of trust that we have already given to the AI algorithms that goes way beyond what we used to expect from simple Google-based searches. We expect it to always be correct, even with rather subjective questions. This confidence can be dangerous, for example when people ask leading questions and the chat bot is guessing what the user's intent was. With regular browser searches, we knew there was an additional step to check sources and verify accuracy.
Using ChatGPT should be like Googling something or using Wikipedia. They're good for pointing you in the right direction but you should check everything provided with legit sources. In fact, you can even ask ChatGPT for links to sources it pulls from.
When's the last time you used it? If it's been a year or two, I'd recommend trying it out again. It's always been on p ...[text shortened]... words: in the same way people now say "just Google it", people will start saying "just ChatGPT it".
I'm sure there's a trajectory for ChatGPT that has a net benefit on society and humanity. But as the algorithm gains power and speed, the larger risk is that humanity offloads cognitive workload forever. We won't have the critical thinking skills to know if it's wrong. And there's already evidence this is happening.
https://www.bbc.com/news/articles/cd6xz12j6pzo