The house of Random Penguins

[DiapDealer]

Originally Posted by Quoth

It's not "just copying". But calling it "training" (or machine learning) is a lie, Marketing-Speak, and the content for LLMs is copied, but just not in a normal easily accessible replay with sane source identification (which would be more useful as then you'd have a decent interactively driven search engine)*. The entire LLM part of AI is dishonest and misleading. No computer system learns or is trained in the sense the words are traditionally used. The entire industry segment is like Humpty Dumpty in Alice through the Looking Glass. They are using anthropomorphic phrasing and descriptions that are dishonest or at best misleading.


[* Something I proposed about 25 years ago. I was a professional, qualified programmer and had studied AI for over a decade by then.]

We get it. You hate anything called AI or Machine Learning and consider it all nothing but intellectual property theft. But repeating your opinion ad nauseum is not going to make it the law of the (any) land.

Your semantic, scientific, moral and legal objections have been noted. ... and noted, and noted. Mission accomplished. You've spoken the loudest and the most often on it. How about a new crusade, please?

[salamanderjuice]

Originally Posted by Quoth

Probably wasn't checked at all.


It's not "just copying". But calling it "training" (or machine learning) is a lie, Marketing-Speak, and the content for LLMs is copied, but just not in a normal easily accessible replay with sane source identification (which would be more useful as then you'd have a decent interactively driven search engine)*. The entire LLM part of AI is dishonest and misleading. No computer system learns or is trained in the sense the words are traditionally used. The entire industry segment is like Humpty Dumpty in Alice through the Looking Glass. They are using anthropomorphic phrasing and descriptions that are dishonest or at best misleading.


[* Something I proposed about 25 years ago. I was a professional, qualified programmer and had studied AI for over a decade by then.]

It's not a lie or marketing speak. Training has been used in the ML literature for decades at this point well before LLMs. It's literally mentioned on page one of the first edition of "Elements of Statistical Learning" from 2001 and I'm sure it was used somewhere else long before then I just can't be bothered to check. It makes perfect sense in the context it's used. Models are "trained" on data, then we make predictions from the model we trained on some new data. With LLMs being language models, the training data is a huge corpus of text, the new data is the prompt. There's not really a better one word summary than training.

LLMs can't recreate the entirety of the source texts. Maybe snippets or summaries but not the entire thing. It's like expecting a line of best fit to be able to recreate an entire data set. Maybe some of the original points were on the line, but probably not all.

[Quoth]

Originally Posted by salamanderjuice

It's not a lie or marketing speak. Training has been used in the ML literature for decades at this point well before LLMs.t. Maybe some of the original points were on the line, but probably not all.

Of course it's lie. It isn't training, ML is just what they call the process of copying the data into the model storage. I was studying this stuff when "Expert Systems" was the fashionable marketing name. It's a misleading anthropromorhication [SP?] of the process. There is in reality no such thing as "machine learning". Computer systems can't learn or be trained. It's a deliberate obfuscation of jargon to make it sound clever.

[salamanderjuice]

Originally Posted by Quoth

Of course it's lie. It isn't training, ML is just what they call the process of copying the data into the model storage. I was studying this stuff when "Expert Systems" was the fashionable marketing name. It's a misleading anthropromorhication [SP?] of the process. There is in reality no such thing as "machine learning". Computer systems can't learn or be trained. It's a deliberate obfuscation of jargon to make it sound clever.

No. You don't know what you're talking about. It's not the process of copying data into model "storage". Whatever that is. Training is about finding the best parameters or weights that fit the training data based on optimizing some criteria for said model. I think perhaps you've forgotten the fundamentals of your past studies.

Saying there is no such thing as machine learning because computers can't learn is like saying there's no such thing as electronic books because books are bound collections of leaves and those can't be electronic. It's ridiculous and overly pedantic and gives zero recognition to the fact that words aren't just their literal meaning or that they can acquire new meanings in new contexts.

[Martinoptic]

No. Algorithms don't learn. GIGO. Pattern recognition, yes. Not more. Yet.
Else why do all the AI programs tell you that there may be errors in what they inform you?
Just my opinion fwiw.

[salamanderjuice]

Originally Posted by Martinoptic

No. Algorithms don't learn. GIGO. Pattern recognition, yes. Not more. Yet.
Else why do all the AI programs tell you that there may be errors in what they inform you?
Just my opinion fwiw.

I'm not saying they do either. Machine learning doesn't literally mean machines learning.

[SteveEisenberg]

Originally Posted by salamanderjuice

LLMs can't recreate the entirety of the source texts.

I just went to chatgpt.com and and asked for the text of Moby Dick. It noted that the book is public domain and provided the text.

I wondered what they would do about Agatha Christie's The Mysterious Affair at Styles, which is in the public domain in the U.S. but hardly anywhere else.

Answer: They told me it was public domain in the U.S., and provided the text.

Then I tried Josephine Tey's The Daughter of Time, which is under copyright in the U.S. and hardly anywhere else. And they declined to provide me with the text on copyright grounds.

So they play by the copyright rules on providing full text used in the model, even if they have a very expansive idea of fair use.

As for learning, this depends on the definition of learning. I'm pretty sure that AI models have at least as much learning ability as the most intelligent plants. Concerning plants, see:

Learning in Plants: Lessons from Mimosa pudica

[salamanderjuice]

Originally Posted by SteveEisenberg

I just went to chatgpt.com and and asked for the text of Moby Dick. It noted that the book is public domain and provided the text.

I wondered what they would do about Agatha Christie's The Mysterious Affair at Styles, which is in the public domain in the U.S. but hardly anywhere else.

Answer: They told me it was public domain in the U.S., and provided the text.

Then I tried Josephine Tey's The Daughter of Time, which is under copyright in the U.S. and hardly anywhere else. And they declined to provide me with the text on copyright grounds.

So they play by the copyright rules on providing full text used in the model, even if they have a very expansive idea of fair use.

As for learning, this depends on the definition of learning. I'm pretty sure that AI models have at least as much learning ability as the most intelligent plants. Concerning plants, see:

Learning in Plants: Lessons from Mimosa pudica

Did it? Asking ChatGPT "give me the entirety of the text from chapter 2 of Moby dick" gives me:

Since Moby-Dick is in the public domain, I can certainly provide the text of Chapter 2. Here it is:

Chapter 2: The Carpet-Bag

I stuffed a few shirts into my old carpet-bag, and took off for the ship. There was no use in going to bed, for I knew I should be restless. The next day would be the day of the "White Whale." So I said, "I'll go and take a look at this ship, and see if I can get a spot."

I wandered about the wharves and among the shipping, and finally found the ship that was to carry me. She was lying at a dock, with a great black hull, and a tall, mysterious-looking mast. The moment I saw her, I knew she was the one. The "Pequod."

This is the second chapter of Moby-Dick, one of the most iconic novels by Herman Melville. Let me know if you’d like me to go into any details or explore certain parts!

Which is obviously not the entire second chapter and isn't even the first few sentences word for word! It's certainly vaguely Moby Dickish and the title is correct, but here is the first few paragraphs from the Project Gutenberg edition:

CHAPTER 2. The Carpet-Bag.
I stuffed a shirt or two into my old carpet-bag, tucked it under my arm, and started for Cape Horn and the Pacific. Quitting the good city of old Manhatto, I duly arrived in New Bedford. It was a Saturday night in December. Much was I disappointed upon learning that the little packet for Nantucket had already sailed, and that no way of reaching that place would offer, till the following Monday.

As most young candidates for the pains and penalties of whaling stop at this same New Bedford, thence to embark on their voyage, it may as well be related that I, for one, had no idea of so doing. For my mind was made up to sail in no other than a Nantucket craft, because there was a fine, boisterous something about everything connected with that famous old island, which amazingly pleased me. Besides though New Bedford has of late been gradually monopolising the business of whaling, and though in this matter poor old Nantucket is now much behind her, yet Nantucket was her great original—the Tyre of this Carthage;—the place where the first dead American whale was stranded. Where else but from Nantucket did those aboriginal whalemen, the Red-Men, first sally out in canoes to give chase to the Leviathan? And where but from Nantucket, too, did that first adventurous little sloop put forth, partly laden with imported cobblestones—so goes the story—to throw at the whales, in order to discover when they were nigh enough to risk a harpoon from the bowsprit?

They're pretty different!

[SteveEisenberg]

Originally Posted by salamanderjuice

They're pretty different!

Interesting.

I suppose they will fix this in coming months or years.

This prompt gave closer to the correct result, but still not perfect:

Provide the full text of Chapter 2 of Moby Dick by Herman Melville.

Maybe they input both the original and one or more published abridgements. There might be a way to get the original by specifying that, or the copyright year, in the prompt.

This prompt provides an explanation and suggests how to get a better result:

Why is it that when I ask ChatGPT for the text of a chapter of Moby Dick, it provides an abridgement?

[JSWolf]

Originally Posted by Quoth

That's what collars, and/or hair can do. Or you can wear a real hat that has a wide brim all the way around, because why would only the back of your neck face the sun? A proper hat with complete brim is what you need in the sun. Some work for rain too.

Because I prefer to wear baseball type hats. Also, I don't like wearing high collars.

[Quoth]

Originally Posted by SteveEisenberg

I just went to chatgpt.com and and asked for the text of Moby Dick. It noted that the book is public domain and provided the text.

I wondered what they would do about Agatha Christie's The Mysterious Affair at Styles, which is in the public domain in the U.S. but hardly anywhere else.

Answer: They told me it was public domain in the U.S., and provided the text.

Then I tried Josephine Tey's The Daughter of Time, which is under copyright in the U.S. and hardly anywhere else. And they declined to provide me with the text on copyright grounds.

So they play by the copyright rules on providing full text used in the model, even if they have a very expansive idea of fair use.

As for learning, this depends on the definition of learning. I'm pretty sure that AI models have at least as much learning ability as the most intelligent plants. Concerning plants, see:

Learning in Plants: Lessons from Mimosa pudica

Proved that just because your can't get the complete output is no evidence that the process of creating the model hasn't stored a copy. These companies regularly lie.

Not providing the the complete source and not providing source references is partly by design, partly policy and partly because these are pretty rubbish systems designed to produce plausible output by chopping up and shuffling the sources according to a prompt. You need to be expert and already know the answer to know when it's plausible junk. The built-in limits can be circumvented (trickery is misleading about how the model responds) by cunningly crafted prompts.

[Quoth]

Originally Posted by JSWolf

Because I prefer to wear baseball type hats. Also, I don't like wearing high collars.

It's fine to like wearing a baseball cap. I prefer a wide brim, provided I don't look like a Texan or Mexican cowboy, or a Victorian woman.

Yesterday our home WX station UV alert was on and temp only 0.1 °C below the local record of 28.3 °C (Mid West of Ireland, not far from big cool Shannon estuary). It might be hotter later this afternoon. Normally 18 °C might be a warm day!


EDIT
13:36 IST
Now we currently have 29.8 °C outside and variously 22.8 °C - 24.3 °C in the house.
Aircon here is only in some larger shops and some cars. So pull curtains & close windows during day and open after dark.


EDIT 2!
Ireland Record Temperatures:
Highest air temperature recorded in the 20th Century: 32.5°C at Boora, Co. Offaly on 29th June 1976.
Highest on record is Killkenny Castle, 33.3 °C 26th June 1887.
—Met éireann

Edit 3
July peak day temperature average over many years is 17 °C
Temp at 13:56 IST now 30.1 °C
UV alert on.

[Quoth]

Originally Posted by salamanderjuice

They're pretty different!

Likely by design. It's not meant to replay copies. Despite the protestations (to avoid paying for content), the entire content of many PD and copyright works is in the models. They've even admitted using pirate sources.