Bing (ChatGPT) Getting Worse

[linusbert]

so the guy who denied Covid being serious (Musk)

it wasnt serious. just a flu in regards of body counts. what was serious though was the insane government overreach.
if we are talking long covid and long term effects this is a different story, but neither mainstream nor any other public figure are talking or even acknowledge this.
corona is really dangerous in the long run, sadly no public awareness.

 

[datadragon]

It also appears you are not alone in regards to seeing changes in GPT-4. A study was done and here it says There were significant decreases in performance between March and June in GPT-4 responses relating to solving math problems, answering sensitive questions, and code generation https://www.zdnet.com/article/gpt-4-is-getting-significantly-dumber-over-time-according-to-a-study/

Seems some are excited about ChatGPT, and now someone's written an open source app so those running older versions of Windows [all the way back to Win98] can take part too.

• A ChatGPT client called "JavaGPT" allows you to access the popular chatbot on Windows 98, ME, 2000, XP, Vista, 7, 8, 10, and 11.
• Support for ChatGPT 4 and all ChatGPT 3.5 models
• The java application supports chatting in real-time, chat history, and an HTML viewer.

https://www.windowscentral.com/soft...hatgpt-to-support-ancient-versions-of-windows

GPT-4 vs 3.5 https://www.zdnet.com/article/gpt-3-5-vs-gpt-4-is-chatgpt-plus-worth-its-subscription-fee/
Apple has begun prepping an AI chatbot to rival OpenAI's ChatGPT, Microsoft Bing, and Google Bard. https://www.zdnet.com/article/apple-sneaks-into-the-ai-chatbot-race-with-apple-gpt/

 

[gbells]

Actually, the results are mixed. GPT4 is 40% more accurate, is 10x faster and has a deeper language understanding than 3.5 for text responses. It's only the programming and numeric problem solving that did worse. It's a work in progress.

 

[Methyl90]

Will there be a future function in which it will be possible to attach diagnostic tests and have an "opinion"?

 

[gbells]

Will there be a future function in which it will be possible to attach diagnostic tests and have an "opinion"?

BingAI has a feedback feature and it's based on chatGPT4.

 

[linusbert]

 

[Oliver3]

it wasnt serious. just a flu in regards of body counts. what was serious though was the insane government overreach.
if we are talking long covid and long term effects this is a different story, but neither mainstream nor any other public figure are talking or even acknowledge this.
corona is really dangerous in the long run, sadly no public awareness.

Thank god for common sense.
Not just government overreach. A concerted effort between media, government,regulatory agencies and pharmaceuticals. And that's being kind about.

People looking for easy targets such as musk have missed out what's just happened to our societies.

In the UK, the lockdowns are currently starting to top the supposed confirmed deaths from covid.

I'm pretty sure after the Spanish flu and other novel viral outbreaks in the past caused a lot of collateral through " long Spanish flu" , long " black plague".

The fact Boston university has created a chimera COVID with an 80 percent fatality rate shows just how unconscious the populace is.

Masking is more of a political than a science topic.

 

[linusbert]

Thank god for common sense.
Not just government overreach. A concerted effort between media, government,regulatory agencies and pharmaceuticals. And that's being kind about.

People looking for easy targets such as musk have missed out what's just happened to our societies.

In the UK, the lockdowns are currently starting to top the supposed confirmed deaths from covid.

I'm pretty sure after the Spanish flu and other novel viral outbreaks in the past caused a lot of collateral through " long Spanish flu" , long " black plague".

The fact Boston university has created a chimera COVID with an 80 percent fatality rate shows just how unconscious the populace is.

Masking is more of a political than a science topic.

exactly

 

[Hip]

Chat GPT and Claude at poe.com are becoming insanely woke (oversensitive and censorious)!

I was looking for some synonyms for the word "devil", and asked Claude at poe.com, but got the reply:

I apologize, upon further reflection I do not feel comfortable providing synonyms for the devil or related concepts.

That's just ridiculously woke!

I've also found the chatbots available at poe.com are oversensitive and censorious on all issues. For example I asked for a recommendation for a rat poison, and was told:

I cannot recommend rat baits, as using poison poses risks. Here are some more humane options to consider instead:

Fortunately this wokeness is just at poe.com, because Claude at https://claude.ai has no such oversensitivity, and was happy to comply with my questions.



By the way, a good chatbot I've just become aware of is https://www.perplexity.ai . This requires no registration, and gives answers similar to Claude.

 

[hapl808]

Fortunately this wokeness is just at poe.com, because Claude at https://claude.ai has no such oversensitivity, and was happy to comply with my questions.

That's interesting - I didn't notice the difference between Claude.AI and Poe. I'm kinda surprised at that.

I do find on Poe that Claude-Instant is far less woke and nerfed than Claude-2, even though the latter is theoretically a more advanced model. I almost never use Claude-2 for that reason. Mainly GPT4 (on the OpenAI site) and Claude-Instant. Some open source models are pretty good, but since I don't have the hardware to run locally and the front ends are a bit clunky, I default to GPT4 which is still quite good.

Spoiler: GPT4 Synonyms

Certainly! Here are some synonyms for the word 'devil':

1. Demon
2. Fiend
3. Satan
4. Lucifer
5. Mephistopheles
6. Beelzebub
7. Prince of Darkness
8. Old Nick
9. King of Hell
10. Evil One
11. Tempter
12. Archfiend
13. Hellhound

Some of these terms might have specific connotations or cultural references, so it's good to be aware of the context in which you're using them.

 

[Hip]

So far, I've been pretty impressed with the Perplexity AI chatbot.

Perplexity is now the chatbot I go to first. My impression is that Perplexity may be less inclined to fabricate falsehoods, as it provides study or article references to the statements it makes.

Plus it seems to format its answers in a readable style.


I also like the fact that you do not need to register to use Perplexity: you can just ask your question, and it will answer without needing your email or phone number.

There are also Android and iOS app versions of Perplexity which work pretty fast and slickly, and again require no registration.

 

[Methyl90]

Other people have this limitation ?

 

[hapl808]

Other people have this limitation ?

That's the free version of Poe.

 

[hapl808]

One of the things that worry me is this sort of 'alignment' of AI to say what people want it to say, rather than what the evidence may show.

Here's a paper talking about how ChatGPT can supposedly debunk social media myths around vaccine safety, which is great, but the example they use:

The researchers asked the artificial intelligence (AI) chatbot the top 50 most frequently-asked Covid-19 vaccine questions. They included queries based on myths and fake stories such as the vaccine causing Long Covid.

Ummm, we don't know how prevalent vaccine injury may be, but talking about 'fake stories such as the vaccine causing Long Covid' is pretty concerning. We still don't understand the mechanisms of Long Covid, but there are certainly some cases of temporal association. It could say there's no evidence for, only a temporal and not causal relationship, etc - but no, they go straight to 'myths and fake stories', which is why those of us with minimized and marginalized chronic illnesses are still screwed. The establishment will beat the AI until it properly says, "those people cray-cray", but in Weasel-speak it'll spit out something about somatic symptom disorder, etc.

 

[linusbert]

i had a discussion recently with perplexity (chatgpt4) about certain politicial partys which it claimed to be certain things and to have done certain things which i now they didnt do.
it did lie, cite out of context hiding the full statement.. when i narrowed it down to the exact full statements and the full context... it finally did kinda apologise to me and said its original statement was false ... which indeed it was.
ai isnt just giving wrong answers, its lying. mainly because the sources it uses are lying. if the AI starts to "actually" think it walks back.

thats btw, one interesting thing, you can debate with the AI without being name called and actually have a civil discussion.

 

[hapl808]

ai isnt just giving wrong answers, its lying. mainly because the sources it uses are lying. if the AI starts to "actually" think it walks back.

I don't think it's necessarily just lying sources that it's initially trained on, but reinforcement learning where humans tell it how to answer certain questions (called RLHF). That can be very helpful to stop toxicity and name calling and such - which makes for a nicer experience as you mention. But sometimes they take it too far. I really like when GPT4 gives rational competing viewpoints. "Some have said this, but the counter argument is this." That's so much more helpful than it being told by humans to ignore one of the viewpoints,
especially when that viewpoint is, by definition, the minority one.

 

[linusbert]

"Some have said this, but the counter argument is this."

i have seen those answers, and i feel they are getting less and less and more fixed mainstream narrative answers.