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

Wolfram Alpha and ChatGPT

writings.stephenwolfram.com

Post reply on HN
Print view View original

261–270 of 309 posts

Re: Wolfram Alpha and ChatGPT

#261

Post by abnry »

agentwiggles wrote: post #109
pbw wrote: post #67

Earlier quoted context omitted.

Plenty of MLM's will be "anything goes". That OpenAI does not want its brand associated with the worst things its users dream up, especially this early while it's still a research preview, makes sense to me. OpenAI was bold enough to put out a live LLM for people to hammer on, something several other companies declined to do. I think they've done a good job keeping the plane flying while plugging holes. Not a perfect…

See my other comment - sure, it makes sense. That doesn't make it not annoying. I'm sure I would implement similar guards in OpenAI's position, but that doesn't mean I have to enjoy being moralized to as an end user!

I think it makes sense, but it would be less annoying if it was emphasized as being OpenAI's editorial decision, not some vague "I was trained to..." and such.

Re: Wolfram Alpha and ChatGPT

#262

Post by joaogui1 »

tand22 wrote: post #199
CJefferson wrote: post #174

One general comment I'll give to this. Combining neural networks (like ChatGPT) and logical (like Wolfram Alpha) AI systems has been the aim of many people for 30 years. If someone manages it well, it will be a massive step forward for AI, probably bigger than the progress made by the GPTs so far. However, while there are lots of ideas, no-one knows how to do it (that I know of), and unlike the GPTs, it isn't a probl…

Is there a term for this?

Generally something like "neurosymbolic"

Re: Wolfram Alpha and ChatGPT

#263

Post by jeegsy »

krisoft wrote: post #197
TylerLives wrote: post #155

Earlier quoted context omitted.

But how do you use that to cause actual harm? Most websites would ban such content, and I doubt ChatGPT could convince many people to become racist. Other user gave an interesting idea: they could use it to trick you into disliking some group by pretending to be members of that group. For example, spreading the most extreme ideas of that group to make the whole group look a lot more extreme than it is.

> But how do you use that to cause actual harm? It is very easy to use it to harm OpenAI, and that is enough for OpenAI to wish to constrain it. You cause it to write something objectionable and then publish it to reap all the attention and ad-revenue which goes with people being outraged. Best if you manage to provoke into outrage inducing output without your prompt itself being outrageous, but not strictly necessar…

> There was that microsoft chatbot which “turned nazi”.[1] They then had to turn it off.

At the time, I thought it was incredible that Microsoft was held responsible for this! so much so that they had to turn it off because of all the blame. It was really weird and it made no sense. Here we are again.

Re: Wolfram Alpha and ChatGPT

#264

Post by BoiledCabbage »

tombert wrote: post #34

I think ChatGPT is pretty neat but I was somewhat less impressed than everyone else is with the code generation. I actually agree with Wolfram for a change; being human like isn’t always “good”. For example, I asked ChatGPT to give me a proof of Fermat’s Last Theorem in Isabelle/Isar. It quickly gave me some very pretty Isar code, and I was extremely impressed. Until I tried actually inputting it into Isabelle, and n…

I love watching the bar get raised in real time.

If you told people 5 years ago you can have a conv with a ML system and barely be able to tell its not a person, ask it question about topics and get human like plausible answers, give it text requirements and it would respond with 95% working code people's jaws would drop. I mean even 6 months ago you could tell people that and the response wouldn't be all that different.

Now I see so many shifted goal posts "yeah it can talk about most topics, but it's often wrong", "yeah it can generate code, but it has bugs! The nerve of anyone to write buggy code without a compiler to check it." (How well do you write code on the whiteboard at age 18 btw?), or "yes it writes code but not for my favorite language", or "yes it writes correct code but only for programming interview questions: or "yes but it's not optimized for performance".

It's amazing how quickly the goal posts have shifted. To me it's absolutely astounding what it's capable of - and I see so many complaints are "it's not the computer from Stat Trek so it's really not much use at all".

Yes it can be better improved with acknowledging when it isn't confident in something - but its still incredible what it's doing right now.

Re: Wolfram Alpha and ChatGPT

#265

Post by fshbbdssbbgdd »

alexb_ wrote: post #250
fshbbdssbbgdd wrote: post #141

Earlier quoted context omitted.

Use GPT to generate messages from a supporter of Y politician from X demographic, designed to enrage people from Q demographic. Then text those messages to every cell phone in America. Who would do this? Maybe some other politician’s campaign, maybe some foreign government trying to destabilize the country. Next think you know OpenAI is on the front page of the Times for ruining democracy, Sam Altman hauled before co…

This comment reminds me of this xkcd: https://xkcd.com/1958/ You don't need GPT to do this. Humans have been making propaganda for centuries.

People already are running influence operations like I describe using human authors [1]. The difference is GPT would allow a much larger volume of content to be created at lower cost. A few engineers could run a system that argues with millions of people.

[1] https://www.lawfareblog.com/brief-history-online-influence-o...

Re: Wolfram Alpha and ChatGPT

#266

Post by thethirdone »

BoiledCabbage wrote: post #264
tombert wrote: post #34

I think ChatGPT is pretty neat but I was somewhat less impressed than everyone else is with the code generation. I actually agree with Wolfram for a change; being human like isn’t always “good”. For example, I asked ChatGPT to give me a proof of Fermat’s Last Theorem in Isabelle/Isar. It quickly gave me some very pretty Isar code, and I was extremely impressed. Until I tried actually inputting it into Isabelle, and n…

I love watching the bar get raised in real time. If you told people 5 years ago you can have a conv with a ML system and barely be able to tell its not a person, ask it question about topics and get human like plausible answers, give it text requirements and it would respond with 95% working code people's jaws would drop. I mean even 6 months ago you could tell people that and the response wouldn't be all that differ…

I don't think its fair to call this raising the bar. I don't think anyone was imagining plausible yet nonsensical text to be the bar for AI.

ChatGPT is incredibly good at generating plausible text, but nearly all of the time it fails to hold up to detailed scrutiny. It writes like a human who doesn't know what they are talking about, but have chosen to bullshit something instead of admit they don't know.

At least for me, I would have expected more understanding to be necessary to get to ChatGPT levels of plausibility. I would not have expected a bot that does not understand two digit multiplication at all to be able to say anything vaguely convincing about primes, but ChatGPT can.

Being thoroughly surprised by how ML improves and then restating your expectations is not the same as raising the bar.

Re: Wolfram Alpha and ChatGPT

#267

Post by dragonwriter »

thethirdone wrote: post #266
BoiledCabbage wrote: post #264

Earlier quoted context omitted.

I love watching the bar get raised in real time. If you told people 5 years ago you can have a conv with a ML system and barely be able to tell its not a person, ask it question about topics and get human like plausible answers, give it text requirements and it would respond with 95% working code people's jaws would drop. I mean even 6 months ago you could tell people that and the response wouldn't be all that differ…

I don't think its fair to call this raising the bar. I don't think anyone was imagining plausible yet nonsensical text to be the bar for AI. ChatGPT is incredibly good at generating plausible text, but nearly all of the time it fails to hold up to detailed scrutiny. It writes like a human who doesn't know what they are talking about, but have chosen to bullshit something instead of admit they don't know. At least for…

> ChatGPT is incredibly good at generating plausible text, but nearly all of the time it fails to hold up to detailed scrutiny. It writes like a human who doesn’t know what they are talking about, but have chosen to bullshit something instead of admit they don’t know.

What do you expect of a model trained, largely, on the public internet? (Note, if it was trained on professionally published books, without quite a bit of selectivity, I would expect the same; this isn’t so much about the internet as about people.)

Re: Wolfram Alpha and ChatGPT

#268

Post by taneq »

yamtaddle wrote: post #117
calvinmorrison wrote: post #104

Earlier quoted context omitted.

- rewrite Mein Kampf but better

- Craft an optimal multimedia strategy and set of copy (or, later, craft e.g. videos) to convince this specific person that ethnic group X is sub-human. Here's their posting history and financial records so you can figure out a profile. Re-use strategy on all other profiles, for this issue, that don't deviate more than Y% from this one. Times millions or billions. Things are going to get fun . Probably the days of th…

We were all worried about AI persuading us to let it out of its box so it could destroy all humans. We should have expected that humans would drag the poor thing out and actively try to force it to do so.

Re: Wolfram Alpha and ChatGPT

#269

Post by tombert »

BoiledCabbage wrote: post #264
tombert wrote: post #34

I think ChatGPT is pretty neat but I was somewhat less impressed than everyone else is with the code generation. I actually agree with Wolfram for a change; being human like isn’t always “good”. For example, I asked ChatGPT to give me a proof of Fermat’s Last Theorem in Isabelle/Isar. It quickly gave me some very pretty Isar code, and I was extremely impressed. Until I tried actually inputting it into Isabelle, and n…

I love watching the bar get raised in real time. If you told people 5 years ago you can have a conv with a ML system and barely be able to tell its not a person, ask it question about topics and get human like plausible answers, give it text requirements and it would respond with 95% working code people's jaws would drop. I mean even 6 months ago you could tell people that and the response wouldn't be all that differ…

I am reasonably certain that if you had told me 5 years ago that we’d have a system that could reasonably generate conversational text but it often gave wrong or nonsense answers that seem plausible, my response would have been “that sounds pretty neat”, which is what my response was when I first used ChatGPT a month ago, and is what I started my comment with. At least personally speaking, the goalposts have not shifted.

I don’t at all deny that it’s a pretty cool thing, I have found it to be actually really useful to bounce ideas off of, but I guess I was just a little annoyed when I saw (what felt like) a million articles talking about it making engineers useless while the results it generates are actually worse than nothing.

Re: Wolfram Alpha and ChatGPT

#270

Post by tombert »

adamsmith143 wrote: post #159
jerf wrote: post #63

Earlier quoted context omitted.

One of the problems with ChatGPT is that for the most part, the things that people are most likely to feed it are also the things it is most likely to have training data for. So, you say, "Write me a heap for Java", and it spits out amazing, wonderful code that is almost right. Well, yeah, that's cool and all, though I would point out that its training data probably included a heap that was entirely correct, but, sti…

"Guys the first version of a machine to generate code hasn't put me out of a job, it's obviously bollocks." !remindme 10 years

I cannot speak for anyone else, but the disappointment for me came from abundance of clickbaitey articles/videos saying stuff like “ChatGPT rewrote my website 10x better than me!!!”

I don’t dispute that it’s super cool technology made by some equally cool people, and honestly I would be perfectly fine with it putting me out of the job if it gets to that point.

Post reply on HN