The AI Thread

[Gavster]

[Jobbo]

I'm very wary of AI in my work because every day there seems to be another case where someone has used it and things have gone badly - lawyers being censured by judges, investigated by the professional regulator etc. So I don't trust AI to do law.

However I was interested to watch this Matt Parker video yesterday:


AI it seems can do maths. If you watch the whole video it seems pretty adept at it - at least, at providing proofs or refutations of theorems - by virtue of not giving up like humans, not having biases like humans and being able to explore multiple strands or branches at the same time. Whether it is an efficient use of energy is a separate point (the theorems which have been proved or refuted aren't terribly useful) and I'd like a bit more detail on how much the human checking involved refining or correcting the output but overall I'm pretty impressed.

But: "computer good at doing maths tasks" isn't a particularly novel headline. Charles Babbage knew that over 200 years ago.

[Gavster]

I've not watched it but wondering if it's a similar reason why AI became really good at chess early on. It's an fixed environment, a rigid set of rules, and they key variables never change - 32 pieces and a chequered board. After that, it's simply a case of learning the vast amount of potential moves and trying them all out, which an AI can do far better and quicker than a human ever could.

When it comes to AI replacing people, who can do totally weird, abstract and unexpected things, the fixed environment totally disappears, hence why AI is still crap at writing or making creative judgements.

[RobYob]

I watched it mainly because I really enjoyed Matt's book "Humble Pi", the math's is far far above anything I want to put the effort in to trying to understand but I think Matt's conclusion that beyond being an exceptional search engine LLMs do seem to be doing something interesting and maybe "creative".

[Gavster]

Hey Gemini, how should I share all the images, data and videos I want you to analyse?



<Uploads it to Google Drive>

[Jimexpl]

One of the parents at school has just got a job using Claude to find it for him.

He set up a process for Claude to interview him for an hour, give it a copy of his CV, search once a day on linkedin and the top 20 companies that he wanted to work for, work out if he was a decent fit for the job, create a cover letter matching his qualifications to their requirements, send the cover letter to him for approval, submit the application, and give him a set of bullet ponts for the interview.

About two days to set it all up and test it, then six weeks later he's had an interview and an offer for almost the ideal job.

Now, he works in the gaming industry, and has a decent understanding of AI, but I wondered how many people are doing this, and how it will change recuitment and the need to recruitment companies.

[Beany]

Well, recruitment companies weaponised AI against candidates, so it's only fair candidates do the same.

Also I'd recommend he builds transferable skills up, the gaming industry ain't the most stable these days - even making good games isn't a guarantee of not being bought up by a bigger company, and unceremoniously shitcanned to make that quarters shareholder meeting sound better

[jamcg]

What could possibly go wrong with this?

https://www.bbc.co.uk/news/articles/c5y3j3ngevmo

[duncs500]

What could possibly go wrong with this?

https://www.bbc.co.uk/news/articles/c5y3j3ngevmo

Saw that, indeed, nothing to be concerned about at all.

[GG.]

Yes these things are extremely worrying - you would think after Covid 19 and the controversy about WIV and gain of function testing that efforts would have been made to try and put some guardrails about what kind of virology testing and development can be done.

I'm not generally pro-regulation but AI is something that is an existential threat on multiple levels. Its far from clear to me that the sum of its benefits will outweigh its harms. It may well be something in future years we wish we could uninvent.

[Mito Man]

Damned if you do, damned if you don’t. The risk of not doing it is greater for us I think. Just think of the bad actors doing the same thing, who will never be regulated.

[IanF]

Yeah, I’d say that MAD is the only way with these kind of threats… I’ve seen a few comments/social media posts recently (although nothing official obviously) that Covid 19 was more likely from a laboratory than any other source. Its release being accidental or not is moot as the threat was/is still there

[GG.]

I don't think it was deliberate (why would you release it on your own territory) but 100% it came from WIV.

MAD doesn't work so well for something you can generate at home and let out of a vial in a shopping centre versus needing a bomber or ICBM to carry a warhead.

[Mito Man]

I assume the hope is if you understand the viruses enough to make them then you also have the knowledge to make the cure and then branch out to existing diseases but that’s my optimistic take on things.

Exciting times ahead though. Once nuclear fusion is cracked it will be possible to make enough weapons grade plutonium to destroy civilisation on a weekly basis, with no further refining required.

[IanF]

I don't think it was deliberate (why would you release it on your own territory) but 100% it came from WIV.

MAD doesn't work so well for something you can generate at home and let out of a vial in a shopping centre versus needing a bomber or ICBM to carry a warhead.

Could say the Chinese can easily lose a few tens of millions without damaging their country too much, but bad actors aren’t always at state level, and don’t forget the BC of NBC. Dirty bombs or biological attacks have long been possible with just a few (highly motivated) individuals. Maybe we’ve just been lucky to survived this far.