[LINK] OpenAI just admitted it can't identify AI-generated text.

[Kim Holburn]

AI tools have been and can be very useful tools but the main reason that not being able to identify LLM generated text is a problem, 
is that as more text is LLM generated when they train new LLM systems, if they train them on LLM generated text, the systems will 
get further and further from useful and maybe even understandable output.  GIGO.

LLM generated stories and books may be interesting for a time but they often do have clear artefacts that will make them less 
interesting.

Here's a good if long example why:

https://www.youtube.com/watch?v=aC99lNQdNmA

Quote: AI is improving all the time but at its very  best you will only ever get serviceable imitations of mediocre products. But 
plenty of successful mainstream movies are merely mediochre recycled products.


On 2023/07/29 8:19 pm, Stephen Loosley wrote:
> On 29/07/2023 6:32 pm, Kim Holburn wrote:
>
>> What is this "admitted"?  AI can't identify anything.  It's a machine learning system that strings words together.  It chooses 
>> the next word based on the one before it.  That's it. There is no "identify", "admit",
> "truth", meaning, "understanding" here.
>
>
> Seems little of true value in AI socially expensive false-flag eye-candy?
>
> AU Edition | 29 July 2023
>
> The Conversation
>
> This week, the Australian Society of Authors sounded an alert over the risks generative AI technology poses to authors and 
> illustrators, in a submission to an Australian government inquiry. The inquiry, which closes next week, will consider what the 
> government can do to mitigate risks and support “safe and responsible AI practices”.
>
> “We consider the large-scale scraping and exploitation of works without regard to authors and illustrators rights to be 
> outrageously unfair,” the society wrote. They’re concerned about the risk of copyright infringement and degradation of author 
> rights, the risk to incentives to create and the risk to integrity in publishing.
>
> Publishing academic Millicent Weber surveys the worldwide scene in an article this week. In the United States, the Authors Guild 
> last week submitted an open letter to the chief executives of AI companies, asking their developers to obtain consent from, credit 
> and fairly compensate authors. Some of the world’s best-known authors were among the more than 10,000 signatories, including 
> Jonathan Franzen, Margaret Atwood, Geraldine Brooks and Linda Jaivin.
>
> In the world’s first copyright-related ChatGPT lawsuit, authors Mona Awad and Paul Tremblay are suing OpenAI, claiming their books 
> were used to train the AI software without their consent. Science-fiction magazine Clarkesworld temporarily closed its submissions 
> earlier this year, after receiving hundreds of AI-created stories. As of this month, 984 books for sale on Amazon list ChatGPT as 
> a coauthor.
>
> AI now shadows publishing contract negotiations, with some authors reporting stalled contracts as a result of AI uncertainty. The 
> Australian Society of Authors reports actively working on, among other things, a model clause for publishing agreements 
> specifically relating to AI.
>
> Every new technology brings concerns about how old media might be superseded, and the social and cultural implications, says 
> Weber. Unpacking these concerns can reveal as much about existing practices as it does about new technology.
>
> It prompts us to pause and ask – why do we read? Relationships with human authors are central, says Weber. This is proved by the 
> fact bestsellers are created as much by author-focused promotion as by a book’s contents.
>
> Concerns about the impact of generative AI on creators – and their livelihoods – are also at the heart of the current Hollywood 
> writers’ and actors’ strikes, writes Jasmin Pfefferkorn. “It is crucial,” she says, “that an equilibrium is reached between 
> protections for creative professionals, and the application of generative AI as a useful tool.”
>
> If you’re interested in AI, be sure to subscribe to our free weekly newsletter Science Wrap, covering the latest in science and tech.
>
>
> Jo Case
>
> Deputy Books
> + Ideas Editor
> The Conversations

-- 
Kim Holburn
IT Network & Security Consultant
+61 404072753
mailto:kim at holburn.net  aim://kimholburn
skype://kholburn - PGP Public Key on request