The new ChatGPT is a polyglot, apparently

[Cainntear]

The AI revolution is leaving Arabic speakers behind

...

That's pretty weak, poorly informed stuff. The idea of Arabic being complex is just diverting from the fact that LLMs rely on phenomenally huge datasets. We have stupendous amounts of English text -- more than any other language, and more than could be read by a human in many lifetimes.

What she's talking about is not about actually proposing changes that make Arabic specifically better -- she's talking about a fundamental change in the entire architecture of systems... and in fact the abandonment of AI technology almost entirely.

[Le Baron]

...and in fact the abandonment of AI technology almost entirely.

A great idea. I could do with a rest from hearing about it. I hope it is banned soon until it can be properly regulated and implemented and in the meantime the authorities could work on improving the education of the great mass of almost-cretins the internet world has birthed.

[alaart]

Interesting, the Japanese one sounds actually less like an AI and more like a middle aged women with a hinch of a locality. The slighlty nasal pronunciation, that really has some character.

In contrast all the other languages sound very generic and ai-like.

[Cainntear]

...and in fact the abandonment of AI technology almost entirely.

A great idea. I could do with a rest from hearing about it. I hope it is banned soon until it can be properly regulated and implemented and in the meantime the authorities could work on improving the education of the great mass of almost-cretins the internet world has birthed.

Well the biggest problem is only one of perception. Sadly, if people who don't understand it continue to rattle off tripe about it (see previous link) people who actually know what the hell it is will have to keep talking about it to get people to actually understand it. And no, I'm not claiming to fully understand it myself, but I do know a tiny fraction of it, which seems to be more than most of the people writing about it do. It's all very depressing really.

[Kraut]

My chatbot tells me this:


But give me more information about what you said before about"via fine tuned GPT models". In what fields will you be "fine-tuned"?

Fine-Tuning aims at specializing a language model, which was originally trained to solve tasks as diversified as code generation, cloze deletion, reading comprehension, summarization etc., into a narrower field. By selecting examples from a particular corpus belonging to the desired domain and guiding the model to minimize the negative log likelihood of correctly predicting the next word given the preceding ones in context, e.g. news articles, scientific papers, movie scripts, instructions, recipes etc., the obtained specialized model can outperform its original versions. So, there might be thousands, probably millions or billions (!), specializations each dedicated to serving a certain aspect.
As long as this project follows open source principles, you will always find yourself surrounded by a vast community contributing to improving every submodule.

On your knees, doubters!

[Le Baron]

Excuse me whilst I go and eat some glass. There's also a local competition to watch paint drying, I'm considering entering.

[tungemål]

Excuse me whilst I go and eat some glass.

When the Swedes eat ice cream, they say they eat "glass", which I always found funny.

[Nogon]

Excuse me whilst I go and eat some glass.

When the Swedes eat ice cream, they say they eat "glass", which I always found funny.

First time I was in Norway, I was very surprised about a shop selling "icecream and iron".

[Cainntear]

Excuse me whilst I go and eat some glass.

When the Swedes eat ice cream, they say they eat "glass", which I always found funny.

First time I was in Norway, I was very surprised about a shop selling "icecream and iron".

And I got confused as the first thing "iron" conjured into my head was an electric clothes smoothing device. Cue image of trying to get the wrinkles out of a scoop of vanilla...!

[Kraut]

ChatGPT and the Dawn of Computerized Hyper-Intelligence
https://www.youtube.com/watch?v=S_E4t7tWHUY

Dr. Jordan B. Peterson and Brian Roemmele discuss the future of human civilization: a world of human androids operating alongside artificial intelligence with applications that George Orwell could not have imagined in his wildest stories. Whether the future will be a dystopian nightmare devoid of art or a hyper-charged intellectual utopia is yet to be seen, but the markers are clear … everything is already changing.

Brian Roemmele is a scientist, researcher, analyst, entrepreneur, and tech expert on the forefront of artificial intelligence. His current publication, Multiplex, offers itself as an experiment in journalism as he and his team give live updates on the empirical research they conduct in the field and advocate for the positive emergence and acceptance of AI in much the same way as personal computers.