ChatGPT

[Slim]

I suspect most people will have heard of ChatGPT by now. If not - I'm not sure how to describe it really. It's like a chat bot in that it converses with you in English, but it doesn't do anything frivolous like roleplaying as your online girlfriend or anything like that - rather, it's like a sort of helper for information and tasks across a wide range of topics.

https://chat.openai.com/

Last night, I asked it to analyse a BASH script (simple computer program) that I'd written. It explained how the program worked, very clearly. So I gave it a Python 2 script that I wrote a few years ago, and asked it to convert it to Python 3. It did. And it worked perfectly.

I was very, very impressed by that. And actually it was highly useful because Python 2 is no longer supported on modern Linux distros, and I don't know Python 3. Artificial Intelligence got my code working again.

So I thought I'd try a different topic this morning.







That it has this information available to it is impressive, but what really surprises me is the way it hangs onto the context of a conversation and responds naturally.

This is the future. It's in its infancy now, but in 20 years time with more developed methods and faster hardware, you won't be able to tell the difference between talking to an AI program and a human being.

[The Picnic Wasp]

Truly frightening.

Magnificent.......but frightening!

[Slim]

[The Picnic Wasp]

My goodness! There's a bit of my mind checking it's not April 1st with this. So many scenarios I can think of to interrogate this. I probably wouldn't enjoy the answers. Is there a possibility machines might gain awareness? Perhaps they would make better decisions or perhaps they might just play the odds in way that so called humanity might discount.

[Slim]

AI has been mind-boggling to me for years. It was one of my final year courses on my CS degree; things that seemed like sci-fi to us back then like machine reading and image recognition are commonplace now. For example every multi-storey car park has a camera that will glance at the front of your car and record your registration number.

Happens in a second, but there are a huge amount of steps that go into that - scanning through the pixels in the image and recognising high-contrast regions, then working out if they constitute a line or a boundary between a letter and a background, then compiling them into shapes, then matching them with pre-defined shapes. All these things are complex problems in their own right.

[The Picnic Wasp]

I wish my education or career had headed in that direction. I might have the wrong kind of brain for it though. Despite my fascination for the wonders of software, I don't seem to have the focus or patience to sit down and just work things out. I thought that early Cubase would be the outlet for creativity that I dreamed of, but even my puny Atari 1040 proved an obstacle too far back then. A few years later I stood in a queue in a blizzard outside SoundControl in Glasgow's West End as they were selling Cubans VST dirt cheap in their sale to the first few customers. I was bitterly disappointed as even with my new Gateway PC (with internet) I made very little progress. Geddy's interview quote of "stupid computer" came to mind regularly when he described his experience with MOTU software for the first time. I now use Logic Pro X. My MacBook has given up the ghost so I'll buy the new Mac Mini in a couple of weeks and hope that I can develop one or two ideas I've written in the past. Thinking about it, Geddy's interest in music tech has cost me a fortune over the years. No real regrets though.

[dom]

Fascinating, mind blowing stuff.

I read about the Google AI device named AlphaZero in Yuval Noah Harari's "21 Lessons for the 21st Century". In 2017 it was given the basic rules of Chess and after 4 hours of analysing them it was put against Stockfish 8, a computer programme that was designed solely for Chess and had a database of years and years of Grand Master winning strategies and moves  .  Stockfish 8 could process 70 million moves per second compared to the AI's 80,000. 

In a hundred games between the 2 Alpha Zero remained undefeated winning 28 and drawing 72.

It  also managed moves never seen before, able to think/process differently to the human brain.  That was 6 years ago and I imagine AI has moved on significantly since then

[Slim]

Just one more of these. I wanted to see if the poetry capability was just digging and adapting from a library of predefined ideas - after all I only gave it a name, a gender and a location. So I gave it a specific circumstance, where it would have to apply some creativity.



Now let's face it - as poetry, it's crap. But as a demonstration of the ability of artificial intelligence to perform an abstract, specific task, I find it astonishing.


All characters represented in this production are fictitious. No identification with actual persons (living or deceased) is intended or should be inferred.

[pxr5]

Please write my thesis <on insert subject> for me.
Please do my homework <on insert subject> for me.

Scary stuff indeed.

[Slim]

Interesting exchange with it this morning - it got some maths wrong, twice, but both times acknowledged its mistake. However in terms of understanding what I typed and responding with the appropriate use of language, phenomenal. I do find it bizarre that it changed its mind on the principle though.

[The Picnic Wasp]

Incredible stuff. I wonder if a day will arrive when a court stenographer (or machine derivative thereof) will ever relay data directly to a program such as this. Fraud and financial irregularity trials do appear to be an ordeal for human beings. Also, can it consider the possibility of unfairness of one party to another and give a balanced opinion, or would the algorithms be impossible to achieve objectively as things stand today?

[Slim]

In my opinion - if there is an objective judgement to be drawn, and if the criteria are sufficiently deterministic, then it should be no problem.

I just asked it:

[pxr5]

And now you may have to pay for the privilege https://www.bbc.co.uk/news/technology-64492750

[Slim]

That's reasonable I think. At the moment it's available as a "free preview". The article says the free version will still be available, but I wonder for how long?

[Slim]

Linked from the BBC article, a transcript of an "interview" that an engineer at Google claims to have had with their own, unreleased, AI product. But if I had to bet the house, I'd say this was fake.

https://cajundiscordian.medium.com/is-lamda-sentient-an-interview-ea64d916d917