Hacker News Reader

Comment by SchemaLoad | Hacker News Reader

GenAI largely seems like a DDoS on free resources. The effort to review this stuff is now massively more than the effort to "create" it, so really what is the point of even submitting it, the reviewer could have generated it themself. Seeing it in software development where coworkers are submitting massive PRs they generated but hardly read or tested. Shifting the real work to the PR review.

I'm not sure what the final state would be here but it seems we are going to find it increasingly difficult to find any real factual information on the internet going forward. Particularly as AI starts ingesting it's own generated fake content.

View on HN

More relevant than ever:

> The amount of energy needed to refute bullshit is an order of magnitude bigger than that needed to produce it.

https://en.wikipedia.org/wiki/Brandolini%27s_law

Wow the 3 comments from OC to here are all bangers, they combine into a really nice argument against these toys

The P≠NP conjecture in CS says checking a solution is easier than finding one. Verifying a Sudoku is fast; solving it from scratch is hard. But Brandolini's Law says the opposite: refuting bullshit costs way more than producing it.

Not actually contradictory. Verification is cheap when there's a spec to check against. 'Valid Sudoku?' is mechanical. But 'good paper?' has no spec. That's judgment, not verification.

producing BS can be equated to generating statements without caring for their truth value. Generating them is easy. Refuting them requires one to find a proof or a contradiction which is a lot of work, and is equal to "solving" the statement. As an analogy, refuting BS is like solving satisfiability, whereas generating BS is like generating propositions.

> The P≠NP conjecture in CS says checking a solution is easier than finding one...

... for NP-hard problems.

It says nothing about the difficulty of finding or checking solutions of polynomial ("P") or exponential ("EXPTIME") problems.

It's not contradictory because solving and producing bullshit are very different things. Generating less than 81 random numbers between 1 and 9 is probably also cheaper than verifying correctness of a sudoku.

> The effort to review this stuff is now massively more than the effort to "create" it

I don't doubt the AI companies will soon announce products that will claim to solve this very problem, generating turnkey submission reviews. Double-dipping is very profitable.

It appears LLM-parasitism isn't close to being done, and keeps finding new commons to spoil.

There are a dozen startups that do this.

> Seeing it in software development where coworkers are submitting massive PRs they generated but hardly read or tested. Shifting the real work to the PR review.

I've seen this complaint a lot of places, but the solution to me seems obvious. Massive PRs should be rejected. This was true before AI was a thing.

In some ways it might be a good thing that shorthand signals of quality are being destroyed because it forces all of us to meaningfully engage with the work. No more LGTM +1 when every PR looks good.

AI slop security reports submitted to curl - https://gist.github.com/bagder/07f7581f6e3d78ef37dfbfc81fd1d...

HN Search: curl AI slop - https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...

This one is hilarious. https://hackerone.com/reports/3516186

If I submitted this, I'd have to punch myself in the face repeatedly.

The great disappointment is that the humans submitting these just don’t care it’s slop and they’re wasting another human’s time. To them, it’s a slot machine you just keep cranking the arm of until coins come out. “Prompt until payout.”

Reply on news.ycombinator.com