Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

I let Claude Code write an entire book

github.com

Post reply on HN
Print view View original

11–20 of 84 posts

Re: I let Claude Code write an entire book

#11

Post by amelius »

iLoveOncall wrote: post #10
tux3 wrote: post #9

There have been good discussions recently about preventing AIs from being too sycophantic. There have been some dangerous moments where LLMs would praise every idea the user has as genius, ground-breaking, and a brilliant observation. Some academics have reported a noticeable increase in the volume of crackpot emails they get daily. They're full of LLM-generated nonsense, where the AI goes along with the nonsensical…

> Most of the time it's pretty harmless I think we have to stop the idea that wasting people's time at such high scale is harmless. It's not.

Reminds me of Google's Doodles, which must have annoyed millions of people and wasted millions of minutes worldwide, and don't get me started on ads.

Re: I let Claude Code write an entire book

#13

Post by riku_iki »

bjt12345 wrote: post #3

> The second insight is that human hallucinations are fundamentally social. Unlike AI, which hallucinates in isolation, humans hallucinate collaboratively. At the dinner party, each false fact was immediately reinforced by social validation. The nods, the interested expressions, the follow-up comments - all of these served to solidify the hallucination into shared "knowledge." Oh boy, that burns - the Futuristic AI d…

I think LLM will also hallucinate way more if you will give confirmation: "oh, this is so true, could you continue and expand more?".

Its the issue of reward function, both humans and LLMs are trained on pleasing clients as one of major goal.

Re: I let Claude Code write an entire book

#14

Post by noworriesnate »

vasco wrote: post #7
bjt12345 wrote: post #3

> The second insight is that human hallucinations are fundamentally social. Unlike AI, which hallucinates in isolation, humans hallucinate collaboratively. At the dinner party, each false fact was immediately reinforced by social validation. The nods, the interested expressions, the follow-up comments - all of these served to solidify the hallucination into shared "knowledge." Oh boy, that burns - the Futuristic AI d…

I feel like we do a great job of that already and the internet is full of people pointing it out from others.

What you may be ignorant of is that there are IRL communities as well where there’s not enough of this behavior

Re: I let Claude Code write an entire book

#15

Post by novosel »

gjm11 wrote: post #5
karmakurtisaani wrote: post #2

Did you read it tho? We might be approaching a time when saying an author has written more books than they've read might turn from a joke to a fact.

The actual author is Claude, and Claude has read a lot of books.

It is interesting to me that your emphasis is on _lot_, and not on _read_.

Re: I let Claude Code write an entire book

#17

Post by euroderf »

bjt12345 wrote: post #3

> The second insight is that human hallucinations are fundamentally social. Unlike AI, which hallucinates in isolation, humans hallucinate collaboratively. At the dinner party, each false fact was immediately reinforced by social validation. The nods, the interested expressions, the follow-up comments - all of these served to solidify the hallucination into shared "knowledge." Oh boy, that burns - the Futuristic AI d…

In the eternal struggle of carbon-based lifeforms versus silicon-based lifeforms, both sides are full of crap.

Re: I let Claude Code write an entire book

#18

Post by dinfinity »

rjurney wrote: post #12

As a writer, this is deeply comforting. The book is total shit. It sounds like an AI, not like something a person would connect with for even a moment. Not even in the opening lines.

Did you read more than the opening lines?

When you've read it in its entirety, could you indicate on a scale from 1 to 10 what score it would get compared to published books you've read (including of course all the best and the worst ones)?

Re: I let Claude Code write an entire book

#19

Post by bgwalter »

All issues that are implicitly pointed out as novel in the introduction have been discussed for centuries.

Not only by scholars and experts. Literally 50% of Internet discussion is about biases, selective facts, spin etc. The problem with "AI" is that propaganda can be automated and that it wastes our time.

The topic is suited for "AI" because it is a soft topic that lends itself to uninhibited preaching. "AI" is also great at writing presidential speeches. It is probably the only thing it is good at.

Nevertheless, the result is still painful to read.

Re: I let Claude Code write an entire book

#20

Post by falcor84 »

amelius wrote: post #11
iLoveOncall wrote: post #10

Earlier quoted context omitted.

> Most of the time it's pretty harmless I think we have to stop the idea that wasting people's time at such high scale is harmless. It's not.

Reminds me of Google's Doodles, which must have annoyed millions of people and wasted millions of minutes worldwide, and don't get me started on ads.

Google doodles generally bring me joy and let me know about world events I wasn't aware of. I honestly can't think of a single one that annoyed me. Often it's just "oh, it looks interesting, but I don't have time to click on it now; maybe I'll look into it later"
Post reply on HN