Generating an e-book with all your ChatGPT conversations - Desktop walkthrough

[Comfy.n]

Prerequisite: any Firefox-based browser with the add-on Save As eBook installed:

https://addons.mozilla.org/en-US/fir...n/saveasebook/
or
https://chromewebstore.google.com/de...hnegfmpejdnffj - I've tested and got this working on FF120 and WF Classic!

Steps:

1) On ChatGPT, export chat data by clicking the User Profile icon then Settings, Data Control -> Export data

2) Download file from the link sent in the email and extract.

3) Open the file chat.html in your browser.

4) Click the add-on button, then on Save Page. (or keyboard shortcut Alt + Shift + 1)

5) Drag the resulting EPUB file into Calibre, select it and press K (Edit ToC). Then click Generate ToC from all headings and OK. Done!

(not sure where to post this, but since it involves Calibre I thought I'd share here)

[Quoth]

Or just copy/paste stuff on the Internet into MS Word/LO Writer, format/style/layout, add headings and then use Sigil or Calibre to convert to epub.

Save-As-eBook is a blunt tool that doen't work on some browsers.

Also ChatGPT is practically spyware and not much more useful than the free Emacs which has Eliza in it.
https://www.emacswiki.org/emacs/EmacsDoctor

I actually added chats from multiple bots years ago in to an ePub.

Eliza 1960s. Other early ones were Parry and Racter. ALICE was first well known Web ChatBot (1990s). Mitsuku was 21st C.

Mitsuku, or Kuki to her close friends, is a record-breaking, five-time winner of the Loebner Prize Turing Test and the world's best conversational chatbot (according to folks like Google AI Research). As featured in theNew York Times, Wall Street Journal, BBC, Guardian, Wired, and Radiolab.

Alexa, Cortana, Siri and Go Google are Chatbot front ends to search.
The biggest difference between early ones and current so called LLM is the ability to add to the database without editing the code (so called 'Learning', which is not learning). ChatGPT and competitors use an unprecedentedly large database (the "Model") built from scraping the internet and no payment for copyright content. The makers are IP pirates and it's almost a con.
Human like descriptions are deliberately used, such as learning, training, hallucinating. None of those happen and "hallucination" is simply because it fails easily and mostly produces plausible nonsense. There in no AI in original 1950 to 1980s meaning. Due to cost constraints the input data isn't properly categorised.

As per the Wizard of Oz, look behind the curtain.

[pdurrant]

Originally Posted by Quoth

Mitsuku was 21st C.

I just gave Kuki (current name for Mitsuki) a go. A failure in its third message.

Kuki: Hi there, I'm Kuki. I'm a friendly AI, here to chat with you 24/7. We could start by getting to know each other if you like. What is your name?
Me: Hi Kiki, I'm Paul.
Kuki: Who is Kiki? My name is Kuki.
Me: Whoops - sorry, Kuki, just a typo.
Kuki: There is no need to apologise. No, you are a typo.

[Quoth]

Originally Posted by pdurrant

I just gave Kuki (current name for Mitsuki) a go. A failure in its third message.
…

Not much advance on the Eliza version inside Emacs.

@thomasfuchs@hachyderm.io
There's a few different camps on generative "AI":

- "It's our lord and savior" (mostly AI companies pushing this)
- "It will kill us all" doomerism (it won't, it has no intelligence or sentience or agenda)
- "It has some use cases!" centrism (problematic because it enables the lord and savior types & it's also wrong as work products are highly unreliable, even for basic things like summaries and tabulation)
- "It's a bullshit tech hype, paired with theft and environmental waste" (correct)

Obviously I think he's right.

Not actual property theft (criminal & fixed maximum tariffs paid to state, but IP "theft", a civil offence in most countries, so ChatGPT and others have to be sued. However the tariff is unlimited and based on assessed hypothetical damages, not actual accounts.

Which is why ChatGPT owners and others of LLMs are claiming they will indemnify users (i.e. they'll pay the fine if they lose). Suits are only starting.

They create plausible rubbish.
https://www.theregister.com/2024/03/...ages/?td=rt-4a
Someone I know in a senior role in an International Tech Company is trying to explain to Manglement why it's a bad idea.