@AnthonyPym A cool experiment you can do with ChatGPT that perhaps illustrates what it does with language: 1. Upload a simple image, one color on the left side and a different color on the right side. 2. Ask it to change the color on the left side, for example, to red, by generating an image. 3. Tell it to do not make any other changes. What it will generate will surprise you. And it's not that it does not recognize the colors (I have asked it and it correctly named them), it seems to interprete and understand the request, but the generative part sort of takes different pieces and puts them together. That may well be what it does with language (lumping different sources and words together in a seemingly coherent manner to accomplish what was asked, so that it can simulate human language, but just like the image experiment it's just weaving everything together)

I love your thoughts, really I just love it, thank you

Funny moment, just now watching your video and while you had the screenshot up about the Netflix situation with The Squid Game, I had a notification on my iPhone about Netflix releasing Squid Game: The Challenge, coming November...

It’s tempting to buy quantum AI stocks but as someone who’s trying to practice positive investing I’m not sure this suits my portfolio or the future of humanity. It’s a very important issue and we need to talk about it more. Thank you for addressing these important aspects.

I've been attending a lot of language service industry round tables on AI, and despite the hype and the threats, etc., the resounding opinion is "it is still not possible or advisable to use these systems with client data (i.e. client's source files) due to confidentiality concerns" so for the time being, none of these systems are ready for production at the LSP level

@AnthonyPym

Жыл бұрын

Many thanks for pointing that out. If everybody says it, it must be true.

@InsideInterpreting

Жыл бұрын

@@AnthonyPym Do I detect some Aussie sarcasm?

How much is all of this just a case of certain tech companies hyping up their own technologies and translators and interpreters seeing certain behaviours among their clients and linking the two? When it comes to research, it does appear that few researchers (myself sadly among them) have the time, patience, or training to continually wade through the mound of scientific literature and vague claims put out by MT and GPT companies so it is easier to accept it at face value. I remember wading into that world for my 2019 book and it was strange to find that, once you learn to wade through the acronyms, metrics, and algorithms, the underlying thinking and approach is actually pretty clear and hasn't substantially changed since SMT. It's still a case of assuming (rightly or wrongly) that past usage predicts future solutions and that accuracy is primarily mathematical. We are seeing some incredible results from those assumptions but they don't seem to be the exact assumptions that expert human translators use, and certainly not when they specialise. I would caution the use of court cases as evidence of IP theft. Most of the smaller producers of IP would struggle to afford to bring a case against the big guns and any settlements simply won't make the court records. We do know that some artists are suing some of the big neural art generators and that GPT models suck in huge swathes of the internet. Maybe the best indicator of IP theft is output or the declared methods used by the companies. Although how far that constitutes IP theft is one for the courts to decide.

@AnthonyPym

Жыл бұрын

What's new since SMT? Deep learning, basically.

@tatjanamiladinovic4977

5 ай бұрын

​@@AnthonyPym Serbian subtitles please

@AnthonyPym

5 ай бұрын

@@tatjanamiladinovic4977 You're welcome to add subtitles in whatever language you like!