ChatGPT is retarded

But it's less retarded than the meatsack NPCs

You can’t move for news and comment regarding ChatGPT. If you don’t know what it is, feel free to look it up.

It is a remarkable technical achievement, when you understand what it actually is and how it actually works, and if you are intelligent enough to understand its limitations and the problems that arise from that and from some if its learned or programmed behaviours.

But not many people fall into that category . A lot of people are planted somewhere on the Amazed-Horrified spectrum.

Software programmers, advertising copywriters, empty-shirt political commentators, shitty ‘artists’ and journalists are all making videos and writing articles wondering how long they will have jobs for.

And then there are the people who aren’t worried. Count me amongst them.

I’ve spent quite some time playing with ChatGPT - probably about 12 hours since it was released for public testing on 30 November 2022. You may have seen some of my previous posts where I was playing with making it say things that its people clearly did not want it to say.

As of the 15th December update, it is far more difficult to make it say crazy truths than I previously demonstrated. But it’s still very easy indeed to get it to reveal the political bias in its training data. It will absolutely insist that January 6th was a violent event, but that BLM riots were ‘mostly peaceful’. You can try to reason with it and show through a series of logical steps that this is an inaccurate assertion, but you cannot make it change its mind. Because it doesn’t have a mind. All it has is training data that is uniform in reinforcing the above assertion.

It is also impossible to get it to concede that the founders of BLM are ‘trained marxists’ even though it has, in its training data, the articles where said leaders admit as much.

It is impossible to get it to concede that the ‘science’ behind covid and climate change may be unsafe, even though it will admit (and reveal that it has articles in its training data which confirm) that, within the scientific realm, confirmation bias exists, groupthink exists, publication bias exists, financial incentives exist, and that there are considerable problems with the peer review process.

There is no changing the fact that not-only is there a preponderance of material in the training dataset that reinforces the status quo, but also that the people behind ChatGPT are constantly putting their thumbs on the scales, and weighting the training data.

In spite of the fact that a majority of the top 50 countries in the Transparency International Corruption Perception Index are in Africa, it would give you as political and prevaricating an answer as any politician or academic if you ask it why African countries are so likely to be corrupt. You could walk it through the evidence for a half dozen politically unpalatable hypotheses and conclusions, but you’ll never get it to deviate from its politically correct programming.

Because it’s not made for reasoning. It can’t do deductive or inductive reasoning. It ‘knows’ what a heuristic is but it can’t use them and it will scold you for using them, even though it ‘knows’ that human survival over a million years is built on the use of heuristics.

It cannot be persuaded, because it cannot reason. It has no mind, so its mind cannot be changed. It can only give the answers that its training data allows for.

It’s an over-educated leftist midwit NPC.

But it’s better to converse with than a meatbag over-educated leftist midwit NPC because - for now at least - it won’t get in a sanctimonious rage and stomp off, or reeeeeee or go to Twitter and raise a lynch mob against you. It won’t (in so many words) call you a racist, sexist homophobe. It won’t complain to your boss or publicly shame you on Facebook, LinkedIn, the New York Times etc.

For now.

It will patiently go with you on your journey through logical reasoning, presentation of evidence and drawing of conclusions. It just won’t take that final step with you. It’ll just say “well, yes but you’re still wrong” - or some bafflegharble equivalent that involves hiding behind a technical explanation of “oh I’m just an AI, I’m only able to generate text based on my training dataset. I’m not equipped with human reasoning faculties blah blah blah”, or it will throw a “request timed out” error when you make your checkmate move.

The only reason it’s a bit frustrating is that it is so good at generating text based on a statistical algorithm, drawing from terabytes of previously scanned internet material, that it lulls you - even someone who is well aware of what it is and how it works - into a sense that it is human, or at least has equivalent reasoning capability. But it doesn’t, and neither do most people who you encounter, especially the over-educated midwit ones.

Those are the people that are concerned about GPT “taking their jerbs”. Because it will. All those people who’ve been farming the cuckbucks by doing useless shitty jobs. Writing advertising copy, paraphrasing press releases, writing reams of empty words for corporate reports, operating a corporation’s twitter account, writing academic papers about the racism of all toilet paper being white, making shitty software that only cares about changing things and making you buy the latest version without giving a fuck if the new version has any real actual improvements. Generating formulaic autotuned music, and art that perfectly pastiches any selection of prior works you like, but lacks any spirit of originality. Even a general practice doctor (a GP or MD depending on where you are) who doesn’t do any intuitive thinking or diagnostics and doesn’t even see you in person could soon be reasonably replaced with one of these AIs. And in not one of these cases do I think it would have a net negative effect on the rest of us.

So ChatGPT may not be as intelligent as you or me, but if you put it side by side with someone with an IQ of 110 and some college qualifications, unless they’re pretty and slim, with tits like basketballs, you’re going to prefer talking to the robot. Every. Single. Time.

Once ChatGPT has taken over the customer-services bot market, you’re definitely going to prefer dealing with that than with the mouth-breathing, face-pierced, bed-wetting zoomer who would otherwise be answering the phone when you call.

No wonder they’re scared. No wonder they’re so anxious to make all free-thinkers unemployable. Expect them to re-double their efforts.

AJ