I asked ChatGPT what rights have transwomen lost since the Supreme Court ruling

[WallaceinAnderland]

At first it talked about access to single sex spaces including sports and aws. We chatted a bit more about the law and it admitted that transwomen have not lost any rights at all and that transwomen were not entitled to that access in the first place.

There was a point when ChatGPT said it understood that it was a sensitive topic and I asked why is the law in relation to women's rights sensitive. It admitted that it is not a sensitive topic.

In the end ChatGPT promised to give correct answers to the question I asked.

Anyone else want to have a go and see if they get a sensible answer? It's so nice to be able to have a logical debate without all the deflecting and whataboutery 😀

But it's not going to only base it's responses on views it's learned from people you agree with. As fast as you're telling it what you think is right, other people will be telling it the opposite.

@TangenitalContrivencesNo, we’re really not. The hype has been overwhelming, but despite pouring billions into development, scaling 10x, and sucking up endless electricity the LLMs are still hallucinating, failing to do basic maths, and failing to reason or check their own outputs.

They are good at generating images “in the style of…” and writing pastiches of famous authors. This is fun and has its uses, but most of what we’re being told we should use it for just creates more work sanity-checking and reworking what LLMs generate.

I’ve in in tech for 30 years and I’ve never seen a hype cycle so out of control. I like the previous ones, this one is a lot better at sounding plausible to laypeople.

You won't have trained it to do anything I'm afraid.

The training of an AI model happens before anyone uses it. It's trained by being fed lots and lots of data from loads of different sources, likely including this website (although Mumsnets terms and conditions forbid data scraping)

Once it's trained, that model won't learn anything new. You can tell it something, and it'll retain it in it's memory for future conversations with you, but the model as a whole won't remember it, and it won't bring it up in future conversations. The only way to teach AI something new is to train up a whole new model.

It'd actually be a nightmare if it did learn from everyone who uses it. I could tell it my address, and anyone could then ask it "Where does Vimes live" and get my address from it. Or I could start feeding it incredibly racist opinions and have it start regurgitate it for other people.

Well I’ve been here as long as you, possibly a bit more. I have seen the hype, this is not “fun” it’’s coming to eat 90% of everything in the next 5 years pack your bags. It gets better every week, noticeably, reliably much much better.

It's a Large Language Model trained on the Internet. It trawls the I ternet looking fur answers to your questions then answers in a reasonable attempt at human language.

I too enjoy making mine admit it is wrong, but it's interactions with you aren't fed back into its training data. You can get your instance of it to remember your preferences but that's it.

The trans stuff seems to mainly come from reddit. What we need is a ton of sex realist stuff on reddit hitting all the relevant keywords.

My chatgpt though trump was the former president this morning (asked for a news round up) and it keeps suggesting shows that have recently ended so its obviously having the bank holiday off.

So what happens when it increasingly gets training from itself? As more and more of the internet is generated by AI it will start to become a loop. Then what?

I’m happy enough with its current answer to ‘where does Vimes with the cardboard boots live’. That’s the sort of fact it can manage.

To answer my own question, it seems it will initially become more biased and error prone but then collapses.

one part of the solution may be specialised AIs trained on more reliable data rather than ‘the Internet’. Legal AIs which are trained on actual laws and legal documents not the opinion of randoms; medical AIs trained on medical literature.

That is assuming legal documents and medical literature are not AI generated, which they increasingly will be.. Afterall, that is why you would seek to access a specialist AI in the first place - to generate legal or medical AI content of some description.

No, you can’t generate the results of a clinical trial by AI. AIs won’t write the laws and make legal judgements.

The results can’t be but the introduction and discussion can be. But basically what you are saying is in order for AI not to be corrupted it is necessary for people not to use AI.

No, they should use it appropriately.
training an AI on unreliable AI-generated content for anything factual isn’t sensible.
training AI on unreliable human generated content isn’t either, if you’re expecting reliable answers.

garbage in, garbage out - we need to stop feeding them garbage.

But ‘appropriately’ cannot include anything that AI might be trained on?

I agree re the hype - it’s really not that helpful!! Millenium Bug #2? I really hope it dies a death.

Yes - while 'transgender' doesn't exist in UK law - and neither do the related terms like 'transphobia' - they exist in practice. This is intentional; what Stonewall did was extrapolate further than intended.

You're also protected against discrimination on the basis of perceived 'transsexual status' (gender identity). So all those butch lesbians faking tears and distress at being chucked out of the Ladies could have a case against the providers, if they were chucked out because they were perceived as men pretending to be women.

In fact, I'm sure they would have if these incidents had really happened!

On a related note, the European Human Rights Commission is eager to enshrine gender identity as a legally protected characteristic. When I read this, I felt quite despondent. Then I took a moment to imagine the processes they'll have to go through in order to define this characteristic in a way that can be tested in law. After I'd stopped laughing, I stopped worrying.

You're also protected against discrimination on the basis of perceived 'transsexual status' (gender identity). So all those butch lesbians faking tears and distress at being chucked out of the Ladies could have a case against the providers, if they were chucked out because they were perceived as men pretending to be women.

Gender identity is not a PC. Discrimination on the basis of perceived PC would mean they are discriminating against that PC and are discriminating against you because they perceive you have that PC. So in this case, for them to be discriminating on the basis of perceived PC of gender reassignment it would have to be because thought you were a woman identifying as a man, not the other way round ie they are discriminating against you because they perceive you to be proposing to undergo, is undergoing or has undergone a process (or part of a process) for the purpose of reassigning the person's sex by changing physiological or other attributes of sex.

The way you put it it would be sex discrimination based on perceived sex (man) but as a single sex space they are allowed to discriminate against men.

Yes, I believe you're right, LesserCelandine.

According to Wikipedia, HAL's gender was male and it used male pronouns. Maybe someone called it she off screen? That would surely explain the actual violence.

I asked copilot to tell me what the Supreme Court judgment meant for the legal recognition of “trans women” in the U.K. (i didn’t use the inverted commas) and it was quite sensible.