The new GPT-4 AI arrived today.

[clv101]

This long interview with Sam Altman (OpenAI CEO) is useful to frame what GPT4 is in the context of AGI:
https://www.youtube.com/watch?v=L_Guz73 ... LexFridman

[BritDownUnder]

I like those Lex Fridman interviews but with him being a bit of Putin fanboy (being Russian born) I find him a bit irritating at times. However watch the one with Stephen Kotkin where he got put right by Stephen more than once on a few Russian history matters.

[clv101]

Lex has another great interview this week with Eliezer Yudkowsky. This, and last week's are long (best sped up a bit), deep dives into large language models, artificial general intelligence and the alignment challenge.

Also see this for a hint at the rate of improvement:
https://www.youtube.com/watch?v=5SgJKZL ... IExplained

[PS_RalphW]

It is reported that chat gpt widely invents and quotes fake references to non existent articles by real reporters in real newspapers when inventing it's own articles on a topic. These will mean that it will be necessary to follow and check every reference quoted on every Web page to have any confidence at all on anything you read on the Web. If it does this with scientific publications as well, it will give a huge boost to every conspiracy theory and pseudoscience theory on the planet.

[Mark]

It is reported that chat gpt widely invents and quotes fake references to non existent articles by real reporters in real newspapers when inventing it's own articles on a topic. These will mean that it will be necessary to follow and check every reference quoted on every Web page to have any confidence at all on anything you read on the Web. If it does this with scientific publications as well, it will give a huge boost to every conspiracy theory and pseudoscience theory on the planet.

Very worrying - we already have enough problems with people twisting things in this 'post truth' world....
Suspect the climate change deniers and their big oil backers are already onto it...

[Mark]

ChatGPT: Mayor starts legal bid over false bribery claim:
https://www.bbc.co.uk/news/technology-65202597

An Australian mayor said he may take legal action over false information shared by advanced chatbot ChatGPT. Brian Hood, Mayor of Hepburn Shire Council, says the OpenAI-owned tool falsely claimed he was imprisoned for bribery while working for a subsidiary of Australia's national bank. In fact, Mr Hood was a whistleblower and was never charged with a crime.

His lawyers have sent a concerns notice to OpenAI - the first formal step in defamation action in Australia. OpenAI has 28 days to respond to the concerns notice, after which time Mr Hood would be able to take the company to court under Australian law. If he pursues the legal claim, it would be the first time OpenAI has publicly faced a defamation suit over the content created by ChatGPT. OpenAI has not responded to a BBC request for comment.

[Vortex2]

Going from the sublime to the ridiculous ...

After a lot of work, I managed to get a baby AI chatbot running standalone on a fastish Linux PC with 8GB of RAM.

(It wouldn't run under my 8GB Windows PC as the OS has stolen much of the RAM)

It is similar to - but much, much less capable than - GPT-4.

It's a tad slow .. but no worse that GPT-4 when it is overloaded.

It does however show that we will soon see AIs all over the place

They don't have to be mega smart like GPT-4 ... standalone 'thick' AIs focussing on a niche topic such as car servicing will do very well.

[johnny]

It is reported that chat gpt widely invents and quotes fake references to non existent articles by real reporters in real newspapers when inventing it's own articles on a topic. These will mean that it will be necessary to follow and check every reference quoted on every Web page to have any confidence at all on anything you read on the Web. If it does this with scientific publications as well, it will give a huge boost to every conspiracy theory and pseudoscience theory on the planet.

Well, didn't Vortex ask it a peak oil question, and it didn't seem to answer with LATOC quotes, doomer garbage, Ruppert or Heinberg videos, and all the other nonsense that was so common and easy to spot in the moment, even all those years ago before it was revealed as a crock. Do we have any other examples from chat gpt on a topic we know the facts of that it generated a garbage response so we can examine it in light of the known?

[clv101]

The pace of AI development is quite remarkable. This week, it's worth watching this lecture from Sebastien Bubeck:
https://www.youtube.com/watch?v=qbIk7-J ... tienBubeck

And AI Explained is always worth watching:
https://www.youtube.com/watch?v=6NoTuqD ... IExplained

And this from Tristan Harris (Social Dilemma) and Aza Raskin is also excellent, I saw this a few weeks ago but it's on YouTube now.
https://www.youtube.com/watch?v=xoVJKj8 ... Technology