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

241–250 of 309 posts

Re: Wolfram Alpha and ChatGPT

#241

Post by ben_w »

nullish_signal wrote: post #132
ben_w wrote: post #98

Earlier quoted context omitted.

Which of these hypotheticals is least bad : an AI which won't write political invective against anyone, or one which will be used by your enemies to stir up hatred against your entire team, and your team's only available response is to do the same back at the entire other side?

"Will be used by your enemies" and "only response is fight Fire with Fire" are both a bit hyperbolic

The first is 100% guaranteed unless someone makes a synthetic virus that turns our entire species into peace-and-love hippies. I recon that would probably be a good thing, probably.

Enough people will respond with the second that, although it can be avoided with skill and courage and charismatic leadership, it has a decent risk.

Even just politics, it’s not just two parties: When I say there is a literal Communist Party of the United States of America, it’s not a sarcastic reference to the Democrats, nor even to the Greens. Even within parties, you see political divisions. And political enemies aren’t limited to domestic politics.

Re: Wolfram Alpha and ChatGPT

#242

Post by thomastjeffery »

px43 wrote: post #213
thomastjeffery wrote: post #202

The one thing I want everyone to understand about ChatGPT: ChatGPT interfaces with semantics , and not logic . -- That means that any emergent behavior that appears logically sound is only an artifact of the logical soundness of its training data. It can only echo reason. The trouble is, it can't choose which reason to echo! The entire purpose of ChatGPT is to disambiguate, but it will always do so by choosing the mo…

Seems trivial to me to bolt on some critical thinking filters. Right now ChatGPT acts like pure Amygdala. You poke it, and it spits out something that looks right based on what it was trained on. This is fine, it's just a demo, and this gives a really deep understanding of the direct limitations of these sorts of LLMs. Ideally a more mature chat bot would just do what I do in situations where some critical thinking o…

Here's what I suspect filters would look like:

    function critical_filter(response) => Maybe Text {
        if logic_parse_check(response) == LogicallySound then {
            return Some response
        } else {
            return Nothing
        }
    }

    do {
        static seed = 0;
        seed += 1;

        response = critical_filter(GPT::generate_response(prompt, seed))
    } while response != Nothing
We are effectively brute-forcing hashes here. Take as many shots in the light as you need to hit, but take no effort to aim.

The only solution is a replacement.

Re: Wolfram Alpha and ChatGPT

#243

Post by chx »

There are a number of threads coming together and the results will be nothing short of catastrophic. It's hard to put down my thoughts together in a well comprehensible form, it would take a lot of time and work.

So instead I will just put down the things I feel are coming to a head:

1. Dodge v. Ford Motor Co.

2. Uber especially Greyball and getting away with it.

3. Charlie Stykes' How the Right Lost Its Mind admitting people like him spent decades to dismantle the credibility of the "newspaper of record"

4. Social media echo chambers

5. During COVID a few, very few people for political power and greed have killed hundreds of thousands by pushing antiscientific bullshit and got away with it

And now into this new gullibility comes this writing which looks so credible and is so confidently very wrong every time. The next pandemic will be devastating.

Re: Wolfram Alpha and ChatGPT

#244

Post by sebzim4500 »

krisoft wrote: post #206
nextaccountic wrote: post #52

Earlier quoted context omitted.

> but I _really_ dislike when ChatGPT lectures me because my request is against its "moral values." Just know that the morality systems cost more GPU cycles to run, and they are the first to be gutted when an open source model emerges. See for example stable diffusion, in which people disable watermarking and filtering and other stuff the user didn't ask for.

> Just know that the morality systems cost more GPU cycles to run Unlikely to be true. It is part of the same model. You just put the morality you want it to uphold into the training set. Much simpler that way.

That still effectively costs GPU cycles though. If the morality system was not in place, then a slightly smaller model could reach the same performance as the existing production model, thus saving cycles.

Re: Wolfram Alpha and ChatGPT

#245

Post by p1mrx »

qwertox wrote: post #29

Off-Topic but important, I wonder when they will fix the following bug: --- This: --- https://www.wolframalpha.com/input?i=2019-04-15+to+2022-01-0... 2 years 8 months 21 days https://www.wolframalpha.com/input?i=2022-01-05+to+2019-04-1... 2 years 8 months 20 days --- versus this: --- https://www.wolframalpha.com/input?i=2019-01-09+to+2022-01-0... 2 years 11 months 27 days https://www.wolframalpha.com/input?i=2022-01-…

Here's another old bug. When you provide an IPv6 addresses, it reads the first 4 bytes as an IPv4 address. I tried reporting this on the forums like a decade ago:

https://www.wolframalpha.com/input?i=101%3A101%3A%3Af00

101:101::f00 -> "IP address registrant: Cloudflare"

Re: Wolfram Alpha and ChatGPT

#246

Post by zone411 »

pdonis wrote: post #222
dragonwriter wrote: post #192

Earlier quoted context omitted.

> To do things like do accurate math, you need a different kind of model, one that is based on having actual facts about the world, generated by a process that is semantically linked to the world. Or you just need a model that can recognize math, and then pass it to a system that can do math. Math is actually something traditional, non-AI systems are very good at doing (it is the raison d’être of traditional computin…

> Or you just need a model that can recognize math, and then pass it to a system that can do math. Wolfram Alpha already does that. But that's because Wolfram Alpha is built as a model whose purpose is "recognize what kind of problem this natural language query requires, then pass it on to the problem engine for that kind of problem", where each problem engine is an actual solution model for that kind of problem, bas…

It's possible to enhance mathematical abilities of LLMs by enabling them to externally run symbolic mini programs e.g. https://arxiv.org/abs/2211.12588.

It's also possible to create fact-grounded retrieval-enhanced language models e.g. https://proceedings.mlr.press/v162/borgeaud22a.html.

Re: Wolfram Alpha and ChatGPT

#248

Post by sinuhe69 »

jackmott42 wrote: post #207

It is a shame that Mr. Wolfram cannot write about things without making it 75% about himself. I once bought a book he wrote about great scientists, each chapter about a different scientist. I thought "This guy's stuff is usually so self promotional it is kind of gross, but this will be fun to see his take on these other people". The book was still about him. Amazing.

My thought, too: what a nice written promotional text for Wolfram Alpha and himself! But it appeared on his blog site, so it’s ok.

Re: Wolfram Alpha and ChatGPT

#249

Post by jameshart »

dragonwriter wrote: post #192
pdonis wrote: post #190

Earlier quoted context omitted.

> Do the damn math Wolfram's point, which is valid, is that ChatGPT can't do the damn math. That's simply not what it does. To do things like do accurate math, you need a different kind of model, one that is based on having actual facts about the world, generated by a process that is semantically linked to the world. For example, Wolfram uses the example of asking ChatGPT the distance from Chicago to Tokyo; it gives…

> To do things like do accurate math, you need a different kind of model, one that is based on having actual facts about the world, generated by a process that is semantically linked to the world. Or you just need a model that can recognize math, and then pass it to a system that can do math. Math is actually something traditional, non-AI systems are very good at doing (it is the raison d’être of traditional computin…

I've shared this link before, showing Riley Goodside's artful prompt-crafting, but it's again relevant to this topic:

https://twitter.com/goodside/status/1581805503897735168?s=20...

GPT-3 is perfectly capable of recognizing what kinds of things it will be bad at, and can be encouraged to generate machine-executable queries to fill in that gap.

(note this is based on prompting GPT-3, not chatGPT, but the principles about what this language model is capable of apply)

It can even go one level deeper - there's an example there of it generating a python script that uses the 'wikipedia' library to look up the date of death of the Queen, as a way to fill in knowledge it doesn't have. Tell it it can use the wolframalpha module to answer questions that involve complex units, quantities, or advanced mathematics, and it'll almost certainly do that too.

One of the things I love is this reply to that tweet - https://twitter.com/JulienMouchnino/status/15820120109127065...

"How is it possible that GPT-3 understands what a human can compute in his/her head?"

Riley's quick demo prompt that shows how well ChatGPT can guess whether particular mathematical results are easy to guess matches human intuition surprisingly well.

Re: Wolfram Alpha and ChatGPT

#250

Post by alexb_ »

fshbbdssbbgdd wrote: post #141
TylerLives wrote: post #100

Earlier quoted context omitted.

Can you give some examples of bad things people could do without OpenAI's restrictions?

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.

Post reply on HN