chatGPT

[jacob]

Am I missing something here? I had some paperback instruction manuals printed up a decade ago. It wasn’t magic. Just color laser jet. Plastic coated paper covers. 80 sheets or so of B&W laser printed glossy paper. Thermal bonded spines. I didn’t think of it as book printing but for all intents and purposes that’s what it was.

I think the difference is that print-on-demand machines can make "perfect bound" copies using your own cover (the ERE book is an example of this) whereas thermal binding requires using a plasticy cover or spine. As far as I know print-on-demand machines have no upper limit on the number of pages (if they do, it's very high), whereas a "thermally spined" book can't get too thick before it risks coming apart. In terms of durability rankings, thermal spines are between spiral bound and perfect bound which in turn sit below stitched books. All this comes down to how many times the book is going to be read before pages start falling out.

For a limited run, it's possible to make your own press with a few pieces of wood and some wingnut bolts. When I looked into it, the estimated production time per book would have been 10-15 minutes.

[Sclass]

Ahhh. The photos on Amazon are misleading. They have these thermal binding machines that look like they do perfect binding. Apparently these cheap machines need a special spine or covers + spine to work in their press. The press is slow. Looks like it takes a few minutes of heating and cooling. The results look like conference proceedings. Not really paperback quality.

The shop that did my manuals served the tech companies around Great America Parkway. They did instruction manuals, marketing materials, data books, annual reports and employee conduct manuals. The quality was like the typical manuals that come with software products. Come to think of it those pages do fall apart when you thumb through them too much.

[CS]

@jennypenny

Np. There are lots of places (besides Amazon) you can set up a paperback (or hardback, whatever) for print on Demand and only sell to yourself. A print run is cheaper, yes, but once the book is set-up, you can order more copies later. You can even update, etc. Maybe a combo of the two vendors would serve your purposes.

Ingramspark is one. New ones are popping up all the time.

I’d second the Kickstarter idea. They have the infrastructure and you could get it out to your most favored nation status people. You could also do all kinds of supplemental items if that sounds like fun. I think I’ve bought a dozen books or more on Kickstarter. It is one way to get the upfront funding for the print run.

I think a publishing deal would have made your crazy anyhow. “Here, take my IP, give me squat, and then mismanage it into the next decade” doesn’t sound like your style. Trad publishing is demanding all rights forever these days. It’s a raw deal.

[jacob]

Back on Chat GPT. I was trying to figure out the derivative of (X - X^1/2)/X^2 last night and I decided to check my work using Chat GPT. It got it completely wrong. I typed in where the error was and it apologized and corrected the error which was misquoting the quotient rule of differentiation and then it made an algebraic mistake. I corrected it three times and it gave three different answers and they all were wrong. I really have no idea how this machine works but I was originally under the impression it could solve simple math problems. Maybe it’s not good in this way.

It's a GIGO problem due to the input being based on "the average online enthusiast". It's more appropriate to think of it as regenerative AI than generative AI. Any errors are more blatant in specific problems like mathematics and less blatant (subject to a social version of Dunning Kruger) in more general and opinionated fields. Indeed, [regenerative] AI may even provide better answers than the random human in those cases.

For some questions, the best answer is given by the "wisdom of the crowd". For other questions, the best answer is given by "the smartest person in the room".

For math mechanics, I recommend: https://www.wolframalpha.com/input?i=d% ... 29%2FX%5E2

[fiby41]

It gives you the right formulae and substitutes the values also but gives something else for the result calculation. You can just copy paste the equation in excel and get the calculation done.

[Sclass]

Hey thanks for the Wolfram link. That is really useful.

[jennypenny]

I've decided to DIY it completely, similar to sclass's suggestion. I'm sticking to my decision to steer clear of social media and feeding the superweb. I can barely bring myself to return emails and post here. Boyle got under my skin more than I realized.

[xmj]

@xmj:

Very interesting. As you noted third time was the charm. I think because you prompted with the word "exotic." By morphing ERE with WSP and Millerd, you were pushing the recommendations more towards "productivity" and "conventional success." So, the first two lists of books for ERE/WSP/Millerd were more "boring" than the list generated when you simply asked for higher Wheaton Level suggestions based on ERE. ChatGPT uses the words "advanced" and "alternative" to describe that first more interesting list, and uses the words "complement", "alternative", and "holistic" to describe the second more interesting list. Both of the more interesting lists of recommendations contained titles that are on the "challenging of basic assumptions" edge of being complementary to original theme (temperament) of either ERE (INTj) or ERE/WSP/MIllerd (eNTJ).

Maybe I should have read that comment more carefully, because as a surprise to absolutely no one (except myself), the few MBTI tests I did over the weekend came out ENTJ. Seems I've moved from INTP to its shadow over the course of a dozen years.

Well done 7w5

[ertyu]

Interesting experiments xmj! I enjoyed reading your conversations.

Pre-print estimating the economic impact of chatGPT by OpenAI researchers: https://arxiv.org/abs/2303.10130

They may be a bit biased because the larger the disruption the more valuable the company... however these estimates seem reasonable. GPT-4 has passing grades for many college entrance exams, grad-level entry exams and also the Uniform Bar Exam (Figure 1). They try to quantify exposure* for various jobs. Lowest being dishwasher. Higher exposure for jobs like writing and translating.

Interestingly, one of my acquaintances who is a content writer for the daily dot got first let go in favor of chatgpt, and then rehired (she is a freelancer, so when they contacted her again, she went, "since the last time we worked together, my rates have changed" and it worked). There seems to be something about the human element that cannot be replaced.

[mountainFrugal]

That is interesting. Did they hire back the entire team? Was she the only one to get hired back? I still think for average internet content this is still going to replace (or already has) a bunch of writers. Just like coding with AI assistants, it is going to make those with good taste/baseline ability better able to compete and do more. Perhaps she has an editorial eye as well?

[ertyu]

Could be, it's not info I or she has. I do know of two such people who have been rehired but both of them also have other writing hobbies or income streams -- the first one writes fanfic and ttrpgs which regularly get funded on kickstarter, and the second one also gets commissioned for fringe BDSM stories and is now working on a BDSM porn novel in addition to her content writing gig. They're also both in their late 30s. It's very plausible to me that they would be some of the better content writers out there.

[jacob]

https://arxiv.org/abs/2503.23674

Large Language Models Pass the Turing Test
Cameron R. Jones, Benjamin K. Bergen

We evaluated 4 systems (ELIZA, GPT-4o, LLaMa-3.1-405B, and GPT-4.5) in two randomised, controlled, and pre-registered Turing tests on independent populations. Participants had 5 minute conversations simultaneously with another human participant and one of these systems before judging which conversational partner they thought was human. When prompted to adopt a humanlike persona, GPT-4.5 was judged to be the human 73% of the time: significantly more often than interrogators selected the real human participant. LLaMa-3.1, with the same prompt, was judged to be the human 56% of the time -- not significantly more or less often than the humans they were being compared to -- while baseline models (ELIZA and GPT-4o) achieved win rates significantly below chance (23% and 21% respectively). The results constitute the first empirical evidence that any artificial system passes a standard three-party Turing test. The results have implications for debates about what kind of intelligence is exhibited by Large Language Models (LLMs), and the social and economic impacts these systems are likely to have.

It's also interesting in terms of what is says about humans because now we have an intelligence (or the appearance of intelligence---is there a functional difference) that is not human. For example, here is the ranked list of subjects of most useful to ferret out whether the agent is NOT HUMAN.

Strange
Jailbreak
Scenario/Game
Logic Math
General Knowledge
Humor
Rude
Current Events
Are you a bot/human?
Surroundings
Opinion
Time
Daily Activities
Personal details
Weather
Human experience

In the dating threads, humor has been suggested as a way to estimate people's intelligence. What this shows is that a more effective way to ask your prospective date how they'd commit a crime, respond to something strange, partake in some kind of strategy planning, or run mental puzzle games like what kind of bird is ARTPOR? I'm looking forward to your dating reports

This of course also implies that humans in their typical social state (talking about their experiences, the weather, personal details, how their day was, and what their opinion on this or that is) are NOT demonstrating signs of human intelligence---these conversations are basically just being "generative" rituals that could be replaced by an app at this point.

[7Wannabe5]

One of my dating filters used to be "What would you do if we were camping together and a raccoon was breaking into our food supply?" Another was "What would you do if our flight was canceled/grounded and we were going to be stuck overnight in an airport together?" I will go see what my favorite AI will do with those.

[7Wannabe5]

The results were such that I would be able to tell the difference between AI and human male I might date, because the AI did a much better job at coming up with the ideal response I would prefer if I wrote an answer key beforehand, while the actual human males veered more towards "worse" and/or "surprising." This points to the reality that the problem with any standard is that the unexpected forces recalibration.