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

ChatGPT is now finding bugs in databases

celerdata.com

Post reply on HN
Print view View original

181–190 of 204 posts

Re: ChatGPT is now finding bugs in databases

#181

Post by wholinator2 »

kristopolous wrote: post #98
pfannkuchen wrote: post #88

Earlier quoted context omitted.

I’m no fan of Peterson, but I’m curious— how are intentionally altering the gene pool of a group and cleaning your room colinear?

Look at the table of contents of this book titled "Eugenics" https://archive.org/details/naturessecretsre00shan/page/6/mo... There were multiple points of interest for the Eugenicists. One of them was for the general public and how to get them to have a proscribed conduct and behavior. Maybe you've seen those diagrams where they have 2 life paths depicted from the Victorian era. That's from Eugenics. Example: https:/…

This doesn't make sense to me. But isn't there a difference between senior engineer and jail? Are you saying we should have no input on how children grow up or society behaves because some eugenics people picked the idea up once? Why must we be afraid of _everything_ the eugenicists _ever said_. Can't people sometimes have even a single good idea? I'm not in favor of eugenics at all. But social conduct and manners? That sociology, social psychology, we should grow away anything from any field that has ever been touched by _bad people_?

When i was younger the things Jordan Peterson said actually helped me. This whole "Reverse Midas Touch" thing is getting really out of hand. Just because a person has had bad ideas, doesn't make the person bad or require destruction of every idea they've ever held or spoken. Just because a group of racist and ignorant people held bad beliefs, doesn't mean every possible belief they held was bad and must be destroyed. This whole argument is just, "but the bad people thought that so it must be bad too", without ever establishing what's actually bad about it. Tell me what's wrong with the ideas without mentioning a group of bad people, tell me about why the ideas are bad. Tell me about the people who "manners and proper behavior" has hurt, and then explain to me why it can't possibly be different. Speak about the ideas, not the people. Everyone knows why those people were bad. Now tell me about the ideas

Re: ChatGPT is now finding bugs in databases

#182

Post by SamPatt »

paxys wrote: post #62
danielvaughn wrote: post #56

Earlier quoted context omitted.

I think a lot of overly-eager and under-informed people are jumping on the bandwagon and shouting half-baked proclamations from the rooftops to get clicks and views. But underneath all the bullshit is something truly useful, so I wouldn't necessarily say that Confidently Incorrect™ is ChatGPT's primary ability. Coincidentally, over the past few weeks of this ChatGPT craze, I've needed to create a lot of fake data to…

While I agree with your premise, the "toil" that you mention is a very small use case for a tiny minority of people. And automating it away has always been trivial. What would have taken 30 minutes to write a Python script now takes 5 minutes to come up with the perfect prompt. And then you have to test and be sure that it does what you want. Neat, but not world-changing. Meanwhile ChatGPT is already being hailed as…

Increasing software development speed 6x would be fairly world changing already, wouldn't it?

Add in all the automation of customer service via actually good Chatbots and lots of very cheap (and fairly high quality) copywriting and it seems world changing to me.

Re: ChatGPT is now finding bugs in databases

#183

Post by micromacrofoot »

avgcorrection wrote: post #119
taberiand wrote: post #26

Earlier quoted context omitted.

My hypothesis is that we will soon see the emergence of a super-human artificial intelligence, not because of some astounding breakthroughs in our understanding of intelligence, but because of irrefutable proof that humans just aren't all that complicated to surpass.

Right. But that tech will be created by humans, somehow. These misanthropist slogans run up to contradictions so easily.

I don't think it's so contradictory, we have a long history of stumbling into technological advancement. We have a tendency to get ahead of ourselves before we even understand what we're doing.

Re: ChatGPT is now finding bugs in databases

#184

Post by kragen »

zoklet-enjoyer wrote: post #168
kragen wrote: post #114

Earlier quoted context omitted.

Wow, so jenkem was largely a TOTSE creation? Do you have any idea how the original stories from Africa got started — were there TOTSE members in Zambia?

There's a BBC article from 1999 about jenkem in Africa. It's messed up and sad. Someone saw that and turned it into a meme. Jenkem in America was a totse creation http://news.bbc.co.uk/2/hi/africa/406067.stm

I suspect jenkem in Zambia was also a hoax

Re: ChatGPT is now finding bugs in databases

#185

Post by zoklet-enjoyer »

kragen wrote: post #184
zoklet-enjoyer wrote: post #168

Earlier quoted context omitted.

There's a BBC article from 1999 about jenkem in Africa. It's messed up and sad. Someone saw that and turned it into a meme. Jenkem in America was a totse creation http://news.bbc.co.uk/2/hi/africa/406067.stm

I suspect jenkem in Zambia was also a hoax

I hope so. I found the author's email address. I'll ask her

Re: ChatGPT is now finding bugs in databases

#186

Post by IIAOPSW »

stevenhuang wrote: post #144
IIAOPSW wrote: post #139

Earlier quoted context omitted.

This is at best going to slightly improve the output. I asked about something hyper specific and also extremely trivial: transit wayfinding type queries specific to the NYC subway. All the semantics are services and stations. A service is nothing more than a list of stations it stops at and a station is nothing more than a list of services which stop at it. Eg if I tell it "I'm on the F, how do I get to Time Sq", it…

You're right, it seems particularly bad in navigation. I lead with some general queries on the NYC system, asked it for the common stops for service lines F and A, and it also hallucinated. The wiki for the service lines have pretty complete representations here, so this data should be suitably represented in its training corpus. Would be interested in how a multimodal LLM such as PaLM-E (trained on maps, etc) fair i…

I'm not hopeful that "seeing the map" (multimodal training) will make the difference everyone is hoping for. The transit map is going to be the exact same information as the lists, but coded in the learned design language of colored lines and dots. The words-only version should work just as well or better because it shortcuts the implicit OCR problem of trying to make it learn off the map. Indeed transit maps other than NY are often abstracted and have nothing to do with the underlying geography. So abstract representations (such as lists of words) should be fit for purpose.

Here's another one that fails spectacularly. The digits 0-9 drawn as an ASCII 7 segment display. It gets it mostly correct, but it throws in a few erroneous non-numbers and repeated/disordered/forgotten numbers. Asking it for ASCII drawings of simple objects can really go off the rails quickly.

The fault mode is very consistent. When a prompt forces it to be specific and accurate on an unambiguous topic for 10 or more line items, it will virtually always hallucinate at least one or two. Especially if the topic is too simple to hide behind a complex answer. Even if its learned not to hallucinate 90% of the time, and even if that's good enough to pass at first glance, within a list of 10 things it only has a 35% chance of not hallucinating any of them.

For what its worth, it did very well on law questions. Try as I might, it refused to accept that there's a legal category known as "praiseworthy homicide". Though, I suspect this has less to do with the underlying model and more to do with openAI paying special attention to profitable classes of queries.

I'm sorry to say, I think the problem maybe more intrinsic to the current approach to AI. Frankly, LLM works unreasonably beyond my expectations, but its making up for a lack of a real first principles theory of cognition with absurd amounts of parameters and training.

Re: ChatGPT is now finding bugs in databases

#187

Post by kristopolous »

wholinator2 wrote: post #181
kristopolous wrote: post #98

Earlier quoted context omitted.

Look at the table of contents of this book titled "Eugenics" https://archive.org/details/naturessecretsre00shan/page/6/mo... There were multiple points of interest for the Eugenicists. One of them was for the general public and how to get them to have a proscribed conduct and behavior. Maybe you've seen those diagrams where they have 2 life paths depicted from the Victorian era. That's from Eugenics. Example: https:/…

This doesn't make sense to me. But isn't there a difference between senior engineer and jail? Are you saying we should have no input on how children grow up or society behaves because some eugenics people picked the idea up once? Why must we be afraid of _everything_ the eugenicists _ever said_. Can't people sometimes have even a single good idea? I'm not in favor of eugenics at all. But social conduct and manners? T…

I genuinely appreciate your engagement but this exact thing is what I was trying to avoid as a time suck.

This type of literature has been a hobby of mine for years and recently I've noticed that I've given myself some expertise.

Hopefully I'll be motivated to write a more complete volume in the future. It takes many pages to build proper context here.

Thanks. I've read your reply but I'm trying to be judicious with my time these days.

Re: ChatGPT is now finding bugs in databases

#188

Post by kristopolous »

barrysteve wrote: post #150
kristopolous wrote: post #98

Earlier quoted context omitted.

Look at the table of contents of this book titled "Eugenics" https://archive.org/details/naturessecretsre00shan/page/6/mo... There were multiple points of interest for the Eugenicists. One of them was for the general public and how to get them to have a proscribed conduct and behavior. Maybe you've seen those diagrams where they have 2 life paths depicted from the Victorian era. That's from Eugenics. Example: https:/…

Manners and conduct existed before eugenics and continues to exist after eugenics. You can't pull one blighted moment in the 1920s as proof the whole idea is faulty. There is little accuracy in your description of events. Peterson said we share serotonin with lobsters and it fuels a victory/defeat set of behaviours. If you drop the serotonin from his idea, it's no longer what he was talking about. Your description fa…

I genuinely appreciate your engagement but this exact thing is what I was trying to avoid as a time suck.

This type of literature has been a hobby of mine for years and recently I've noticed that I've given myself some expertise.

Hopefully I'll be motivated to write a more complete volume in the future. It takes many pages to build proper context here.

Thanks. I've read your reply but I'm trying to be judicious with my time these days.

Re: ChatGPT is now finding bugs in databases

#189

Post by kristopolous »

nqrac wrote: post #161
kristopolous wrote: post #82

Earlier quoted context omitted.

correct! Btw, I removed them from my examples, didn't want to "discuss" that here. I'm kind of a history nerd and it's like a fishing hook with delicious bait when I see stuff like that. "Oh wait! You think it was just the south that argued states rights?! Actually (brings up fugitive slave clause jurisprudence)" Or how manner and conduct books of the 1800s, eugenics texts, and jordan peterson's 12 rules for life are…

Peterson an academic who is clueless about the real world. His best work was discussing politics in connection with liberty, especially the Trucker protests. His biggest flaw is that (like the other Daily Wire pundits) he never goes into economics and has absurd ideas that we live in a competence hierarchy. The whole "intellectual dark web" imploded after the start of the Ukraine war, when they first went full neocon…

I'd like to recommend "decoding the gurus" podcast which is probably the best treatment of this material I've seen.

It's still lacking but I should "put up or shut up". They're doing a fine job

Re: ChatGPT is now finding bugs in databases

#190

Post by hungryforcodes »

theonlybutlet wrote: post #148
soco wrote: post #137

Earlier quoted context omitted.

Like said somewhere on the internet, we don't need AI to create art, we need AI to do the dishes and free our time to create art.

But art's the easy part, I wonder if that's why it's fun...

...not done well though. Dishes are much easier to do well.
Post reply on HN