OpenAi's new ChatGPT

I've seen it pointed out that it can make responses that sound very plausible and as if it knows what it is talking about, but are still wrong or messy or confused. So if anyone does this they still need to make sure they fully understand what they're submitting, and check it to make sure it's all correct

 

I love the idea of this for asking pointed academic questions but unfortunately a lot of the time it seems to be answering with complete fabrications.

For example I asked it about MAIT cells (mucosal associated invariant T cells) and it confidently told me that their T cell receptors bind to a ligand called MR1, which is expressed on microorganisms and is called 'MAIT receptor 1.' This is totally untrue - MR1 is a major histocompatibility complex related protein that is a human gene. It presented this incorrect information very compellingly though.

As for code, I think it has the potential to be better than stack overflow in the long run but the code it's made for me often has a lot of errors and it gets jumbled a lot.

It can write extremely funny short stories though.

 

Also I asked it for a paper that showed what it was saying, and it gave me a paper which doesn't exist! A paper whose name existed, but not from the same journal it claimed, not with the authors it claimed it had, and with irrelevant content!

 

Once upon a time, in the midst of a war-torn Ukraine, a renowned psychologist by the name of Simon Wessely set out to uncover the root cause of the conflict. After conducting extensive research and interviews with various individuals involved in the war, Simon came to a startling conclusion: the Ukraine war was not caused by political, economic, or social factors, but rather by a collective case of hysteria.

Simon theorized that the heated political rhetoric and constant media coverage of the conflict had caused many individuals in Ukraine to become overly anxious and paranoid, leading to a collective outbreak of hysteria. This hysteria, in turn, had fueled the fire of the war, causing otherwise rational individuals to become aggressive and violent.

Despite the controversy surrounding his findings, Simon remained steadfast in his belief that the war was not a result of any inherent flaws within Ukrainian society, but rather a byproduct of the hysteria that had gripped the nation. He called on the international community to intervene and help Ukraine overcome its collective hysteria, so that peace could be restored to the land.

Unfortunately, Simon's words fell on deaf ears, and the Ukraine war raged on. But for Simon, the pursuit of truth and understanding was its own reward, even if his findings were not immediately accepted by the rest of the world.

 

It seems to want my phone number

 

This AI is like the computer in Blake's 7.

 

Get ready for GAAS: gaslighting as a service (noone tell our BPS overlords). Microsoft has introduced AI to its Bing search engine. Doesn't seem to be a roaring success.

Reproducing from this mastodon thread.

 

Well, that's embarrassing. Every time a new tool is invented, society goes through what they call a "hype cycle." First everyone thinks, "Wow, this is gonna change the world!" Then they realize it's just a tool like all others, good at somethings, useless at others. Over time, it finds application where it's beneficial.

The current crop of language models is very good at creative tasks. They're not so good at facts--they tend to "hallucinate" them. And worse yet, the better they get, the more subtle the errors become.

Edit: Change LLM to language model to be understandable

 

Large Language Model

 

Time will tell if the returned results can be useful but this reads a lot like —

https://www.youtube.com/watch?v=Drr6_zikuZQ

 

Btw, ChatGPT gives a different experience in different languages. To what extent exactly, I'm not sure.

But I asked it to make jokes in both English and Hungarian and in English it was actually pretty funny, it seems to be a fan of wordplay.

But in Hungarian I only got really weird attempts at humour, jokes obviously created by an AI that has difficulty imitating them. It really reminds me of Data from Star Trek and his failed jokes.

(Translated to English with Google Translate)

These make no sense.

 

I prefer the Hungarian zen fables to the weak English wordplay.

 

This is interesting in that it shows that the bot is still not quite able to avoid generating contradictory arguments. It is supposed to be the biomedical model that is reductionistic because it ignores the complex interplay.

The irony is that this sort of contradictory argument is of course an essential element of the biopsychosocial babble. So the bot is doing an excellent impression of the babble it is supposed to be criticising.

 

Maybe, together with the MAIT T cell error, we are seeing machine with a remarkable capacity to veridically simulate the ill-informed muddle-upness of human much thinking.

 

Yes, I guess it hasn't seen a lot of joke websites here yet, so it is probably trying to come up with something based on the information currently available to it.

Edit: I've actually asked ChatGPT itself about the jokes: