'I'll make £12.24 an hour in my new job - I live payday to payday'
As the chancellor prepares to update her plans for the economy we hear how ordinary people are tackling their own financial challenges.
www.bbc.co.uk
Prices of the electricity we use to charge
- Thread starter flecc
- Start date
Free breakfast club roll out: everything you need to know
educationhub.blog.gov.uk
Free breakfast club roll out: everything you need to know – The Education Hub
The Education Hub is a site for parents, pupils, education professionals and the media that captures all you need to know about the education system. You’ll find accessible, straightforward information on popular topics, Q&As, interviews, case studies, and more.
educationhub.blog.gov.uk
Strangely enough I saw one of those out on the roads last week. Years since I last saw one. Can`t say what we used to call them.
Let’s Buy California from Trump – Denmark’s Next Big Adventure
denmarkification.com
Help Denmark Buy California – Because Why Not?
Buy it from Trump, the bigliest crowdfunding ever
denmarkification.com
Forget about rare earth minerals. We need more copper
Writers think AI models have stolen their copyright and should pay?
That's because they don't have a clue how the AI models work. They don't store the stuff they have been trained on. They read it and remember how words and ideas work, just exactly like we do. When they use their 'knowledge', they don't just vomit up the same text they read. They use their knowledge mixed up with all the other stuff they know from reading millions of documents and they come up with an entirely unique explanation or report, or story, if you ask them to write one.
If you ask an LLM the same question three times, you will not see the same answer again. Every time you ask, the answer is different, pretty much like if you ask a person the same question repeatedly.
The 'They stole my creation' argument is full of holes. You put your work online as a journalist or writer and the LLMs read it and internalised it and integrated its meaning into a massive multi-dimensional vector database of ideas and words. What you wrote does not exist inside the LLM, only its meanings do, and they are linked with multi dimensional vectors ina database of billions of other concepts from other stuff it has absorbed.
The parallels between modern machine learning and human learning are very close. If I talk about a theme that I read online or in a book, my talk is NOT a direct copy of anything I read, it is an amalgam of what I read over my lifetime and my own appreciation of the issues.
So it is with a machine, be that ChatGPT or Google running on thousands of GPUs in a giant electricity sucking beast of a data-centre, and dealing with 50,000, or 5 million concurrent users, or on my little £350 home based machine on my desk.
French publishers and authors sue Meta over copyright works used in AI training
French publishers and authors are taking Meta to court, accusing the social media company of using their works without permission to train its artificial intelligence model.
apnews.com
That's because they don't have a clue how the AI models work. They don't store the stuff they have been trained on. They read it and remember how words and ideas work, just exactly like we do. When they use their 'knowledge', they don't just vomit up the same text they read. They use their knowledge mixed up with all the other stuff they know from reading millions of documents and they come up with an entirely unique explanation or report, or story, if you ask them to write one.
If you ask an LLM the same question three times, you will not see the same answer again. Every time you ask, the answer is different, pretty much like if you ask a person the same question repeatedly.
The 'They stole my creation' argument is full of holes. You put your work online as a journalist or writer and the LLMs read it and internalised it and integrated its meaning into a massive multi-dimensional vector database of ideas and words. What you wrote does not exist inside the LLM, only its meanings do, and they are linked with multi dimensional vectors ina database of billions of other concepts from other stuff it has absorbed.
The parallels between modern machine learning and human learning are very close. If I talk about a theme that I read online or in a book, my talk is NOT a direct copy of anything I read, it is an amalgam of what I read over my lifetime and my own appreciation of the issues.
So it is with a machine, be that ChatGPT or Google running on thousands of GPUs in a giant electricity sucking beast of a data-centre, and dealing with 50,000, or 5 million concurrent users, or on my little £350 home based machine on my desk.
My brother has just received our £27.30 from ALCS for us to share for library loans of our father's books (H.E.Todd). Way back when while he was actively writing and storytelling (mainly schools and libraries) his library loan income was probably close to his book sales and storytelling income. Something similar for AI use of data seems a good compromise.
A few months ago I got a few AI systems to write a short version 'in the style of'. One declined, one produced some poor text not at all in the style, and one made quite a passable go at it (I forget which was which). The real challenge for AI would be to keep a couple of hundred infant school children enchanted for an hour.
I reckon the future of AI is really in robotics. Some new robots don't need pretraining, they learn from watching you doing things.
ChatGPT use case is a good entry point but you can't leave it to make any decision or create anything worthwhile.
I can't see why.
IF a particular book is used, it is read once and digested. It is never stored in the form the author wrote it. The content is fed into an analysing level in the model and its semantic content is used only once and its conceptual content is remembered only in so far as it is compared and linked to the semantic content of huge amounts of other content.
The words of the individual author never come out of the model, they were simply analysed and their similarity and difference to other meaning and content was noted and stored as numbers in a massive vector database.
Libraries may pay - but that is a historical anomaly, and anyway they repeatedly lend out the actual book written by the author. The LLM when it outputs anything, outputs a digest of what it has read from tens of thousands of authors - never the actual words of a particular author.
On creative writing courses, one of the things they always tell those who attend, is this:'If you want to learn how to write; read, read and read more and more authors. Read their books, digest them and you will learn how to do it yourself.' That is EXACTLY what the LLM training does.
So - when JK Rowling wrote her blockbusting series on the young boy at wizard school, she did so on the back if all the reading she had done in her life. She didn't copy any particular author or style, she internalised tens of thousands of hours or reading, university education and schooling and then she started outputting her stuff on the back of all of it. Should she pay too?
She did pay, directly or indirectly, by buying or borrowing the the books she read and learned on.
The advice I would have expected would be 'go back to a normal diet'. How long before AI develops succinctness?
Related Articles
-
MTF Enterprises announces acquisition of EMU Electric Bikes- Started by: Pedelecs
-
Wisper 806T folding bike wins Which? ‘Best Buy’- Started by: Pedelecs
-
Sustrans calls for protected cycle lanes- Started by: Pedelecs
-
Amazon launch their first UK e-cargo micromobility hub- Started by: Pedelecs
