[HN Gopher] Brandolini's Law
___________________________________________________________________
Brandolini's Law
Author : xrd
Score : 67 points
Date : 2023-04-14 10:47 UTC (12 hours ago)
(HTM) web link (en.wikipedia.org)
(TXT) w3m dump (en.wikipedia.org)
| dang wrote:
| Related:
|
| _Brandolini 's Law_ -
| https://news.ycombinator.com/item?id=25956905 - Jan 2021 (59
| comments)
| rpastuszak wrote:
| Aka Gish Gallop (if you do it really, really fast)
| jongjong wrote:
| That's why I mostly look at incentives.
|
| Scientists funded by special interest groups could easily design
| studies in a way to show the results desired by their
| benefactors. In fact, that's exactly what happened with Big
| Tobacco to 'disprove' the link between smoking and cancer. It's
| easier to introduce bias into a study than to prove that bias was
| introduced. It's like with complex computer simulations, if you
| change one seemingly insignificant variable, the simulation could
| give you a completely different result.
|
| Looking at incentives can lead you astray in the short term (due
| to complexity of incentives or hidden incentives), but in the
| long run, incentives provide the most predictable framework for
| figure out the behavior of people and animals.
| breck wrote:
| I agree with this.
|
| Especially with multi-agent systems (anything to do with people
| or biology), which are inherently vastly more complex and
| harder to predict than mechanical systems, it's very important
| to know the incentives of the people making the statements.
|
| Most often they are overestimating their confidence with the
| bias toward their own incentives.
| Paul-Craft wrote:
| [flagged]
| sorokod wrote:
| That one order of magnitude number should be revisited given the
| recent advances in AI.
| Paul-Craft wrote:
| Hah. In which direction?
| 3pm wrote:
| I think it can also be called 'Burden of proof' inversion.
| https://en.wikipedia.org/wiki/Burden_of_proof_(philosophy)
| renewiltord wrote:
| This is one of those things that didn't exist before LLMs. Unlike
| AI agents, people usually form an informed opinion of things and
| don't hallucinate facts with great confidence.
| jen20 wrote:
| Alberto tweeted this about 10 years ago, and was discussing it
| before then. I heard it directly 'out of the horses mouth' in
| London around that time.
| gowld wrote:
| Check the date. And it wasn't new then.
|
| See also:
|
| "Gish Gallop."
|
| "A lie gets halfway around the world before the truth puts it's
| shows on."
|
| The parable of the pillow feathers
| https://www.chabad.org/library/article_cdo/aid/812861/jewish...
| renewiltord wrote:
| Pretty astounding that all these people had LLMs back then
| and didn't release them. It's also crazy that the rabbi was
| talking with an LLM back then. The folk story precedes modern
| computers. How did he run inference? There are probably many
| such secrets in our ancient ancestors' pasts. Who knows how
| many h100s they had back then?
| jimhefferon wrote:
| Try cable sometime.
| renewiltord wrote:
| I watched some cable news channels once at a friend's. It
| appears they have already embraced the AI revolution since
| they had newscasters reading things that seemed unlikely and
| which I later confirmed to be untrue. The only conclusion is
| that LLMs hallucinated the text and AI-generated newscasters
| read it.
|
| Humans would never make the mistakes these guys did. I think
| we should regulate this news that isn't factual.
| fdhfdjkfhdkj wrote:
| [dead]
| waynesonfire wrote:
| same thing applies to fixing tech debt.
| DiscourseFan wrote:
| Its usually not just random bullshit that gets spread around, but
| believable stories that follow a similar sort of logic to those
| we generally agree are true. Though, what might seem to you
| unbelievable may be dramatically different from what others
| consider so, and some things you think are true others might call
| bullshit. Though the material application of knowledge, in the
| end, always demonstrates its useful validity.
| naikrovek wrote:
| yeah. people never (or almost never) make an effort to verify
| things that they believe are true.
|
| therefore, if it is believable, it is believed.
| airstrike wrote:
| because learning something that contradicts your beliefs is
| actually stressful
|
| https://en.wikipedia.org/wiki/Cognitive_dissonance
| tedunangst wrote:
| I'll see it when I believe it.
| jihadjihad wrote:
| Kind of like the Mark Twain quote... "A lie can travel halfway
| around the world while the truth is still putting on its shoes."
| User23 wrote:
| Adding to the confusion is that now "debunking" is often a
| euphemism for a kind of bullshitting.
| dtolnay wrote:
| I think about pull requests the same way:
|
| As a library maintainer, closing and empathetically conveying why
| a pull request is not a net benefit to the project is an order of
| magnitude more effort than what it takes to throw up not-well-
| motivated pull requests on someone else's project.
| pachico wrote:
| I met him last year when he was invited by the company to talk
| about Event Storming.
|
| He's a very smart and nice guy.
| davidw wrote:
| Yeah I met him at a conference in Firenze a few years back.
| Really nice, bright person.
| paganel wrote:
| > talk about Event Storming.
|
| Didn't know about it, so I had to check it out [1]. Looks like
| another corporate bleak thing where they have to game-fy and
| infantilise a so called "process" in order to fool some execs
| into paying money to newly found experts in this new business
| technique. Capitalism is really doomed with these kind of
| people running the show, Schumpeter was right.
|
| [1] https://en.wikipedia.org/wiki/Event_storming
| jen20 wrote:
| If you've ever worked with the kind of organisation where
| this technique is valuable, you'll understand it's exactly
| that: fooling people into telling you what they actually need
| to build, instead of what their Serious Businessperson
| Cosplay persona tells them they need.
| cs702 wrote:
| Interestingly, the amount of effort needed to get large language
| models (LLMs) to generate trustworthy information in a reliable
| manner often seems to be an order of magnitude bigger than the
| amount of effort needed to get them to generate bulls#!t.
| rukuu001 wrote:
| Brandolini is a great guy! He's also behind 'event storming'
| which can be a pretty nice way of getting shared understanding
| and design of a system.
| rpastuszak wrote:
| I've been running event storming sessions for years and just
| realised that I had two different Brandolinis in my head.
| ucirello wrote:
| I wonder if Brandolini was referring to
| https://www.youtube.com/watch?v=ly2BaYeej2Q
| santoshalper wrote:
| GPT makes this 100x or maybe even 1000x. On the other hand, can
| we potentially train generative AI to detect and refute BS as
| well? It may be our only hope.
| rpastuszak wrote:
| Bonus: a quick tutorial on how to use GPT to scale up
| attribution bias: https://sonnet.io/posts/emotive-conjugation/
| wolfram74 wrote:
| Neal Stephenson's Anathem[0] which revolves around epistemology
| a lot coined the term Artificial Inanity for AI
|
| [0]https://englishwotd.wordpress.com/2014/02/17/artificial-
| inan...
| nerpderp82 wrote:
| GPT is also pretty good at cutting through BS. It can detect
| logical fallacies for instance or explain a lack of rigor in a
| discussion. Depends on how you fine tune it, couple it with an
| external fact database and you could get it to cite its
| sources. Couple it with a prolog engine AND a fact database and
| it could modus pwnens ur ass.
| xedrac wrote:
| That's funny, because ChatGPT feeds me BS quite often. It's
| only when I call it out that it corrects itself.
| vkou wrote:
| > On the other hand, can we potentially train generative AI to
| detect and refute BS as well? It may be our only hope.
|
| LLMs store their training information in an incredibly lossy
| format. You're going to need some kind of different approach if
| you want one to tell the difference between plausible-sounding
| bullshit and implausible-sounding truth.
___________________________________________________________________
(page generated 2023-04-14 23:00 UTC)