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

OpenAI Trains Language Model, Mass Hysteria Ensues

approximatelycorrect.com

Post reply on HN
Print view View original

101–110 of 119 posts

Re: OpenAI Trains Language Model, Mass Hysteria Ensues

#101

Post by kirillzubovsky »

gfodor wrote: post #81
kirillzubovsky wrote: post #49

What if OpenAI didn’t write the piece? What if the research was announced by the machine, and the folks at OpenAI are all dead?

You joke, but there's a real point here -- many commenters in this thread are complaining that OpenAI's position on this is a marketing stunt. Presumably, if this stuff gets commercialized, it will probably be adept at a few domains first, and I feel like writing good marketing copy will be one of them. So perhaps the bot itself didn't do so here, but it wouldn't surprise me if a self-marketed bot exists in the near…

What if I am the machine and you the last human left alive?

Re: OpenAI Trains Language Model, Mass Hysteria Ensues

#102

Post by kirillzubovsky »

gfodor wrote: post #81
kirillzubovsky wrote: post #49

What if OpenAI didn’t write the piece? What if the research was announced by the machine, and the folks at OpenAI are all dead?

You joke, but there's a real point here -- many commenters in this thread are complaining that OpenAI's position on this is a marketing stunt. Presumably, if this stuff gets commercialized, it will probably be adept at a few domains first, and I feel like writing good marketing copy will be one of them. So perhaps the bot itself didn't do so here, but it wouldn't surprise me if a self-marketed bot exists in the near…

p.s. I was kidding, but I was completely serious. If they can train a machine to write good copy, they can train the best Russian bots to troll people on Facebook, write New York Times pieces, and fake and influence pretty much anything done through a written text. Heck, they could write a business book and get it into the top-10 that year. Actually, that last part, they should, it would be amazing!

Re: OpenAI Trains Language Model, Mass Hysteria Ensues

#103

Post by cs702 »

ilyasut wrote: post #23

Ilya from OpenAI here. Here's our thinking: - ML is getting more powerful and will continue to do so as time goes by. While this point of view is not unanimously held by the AI community, it is also not particularly controversial. - If you accept the above, then the current AI norm of "publish everything always" will have to change - The _whole point_ is that our model is not special and that other people can reprodu…

This seems very reasonable to me. All the outcry seems... disproportionate.

That said, withholding the pretrained models probably won't make much difference, because bad actors with resources (e.g., certain governments) will be able to produce similar or better results relatively quickly.

All it will take is (1) one or two knowledgeable people with the willingness to tinker, (2) a budget in the hundreds of thousands to a few millions of dollars at most, and (3) a few months to a year. Nowadays a lot of people are familiar with Transformers and constructing and training models across multiple GPUs.

Re: OpenAI Trains Language Model, Mass Hysteria Ensues

#104

Post by kirillzubovsky »

kirillzubovsky wrote: post #101
gfodor wrote: post #81

Earlier quoted context omitted.

You joke, but there's a real point here -- many commenters in this thread are complaining that OpenAI's position on this is a marketing stunt. Presumably, if this stuff gets commercialized, it will probably be adept at a few domains first, and I feel like writing good marketing copy will be one of them. So perhaps the bot itself didn't do so here, but it wouldn't surprise me if a self-marketed bot exists in the near…

What if I am the machine and you the last human left alive?

You know how when we broke the Enigma we couldn't really let Germans catch onto it, so we had to mask our knowledge of their positions by maintaining statistically insignificant number of accidental wins? Much the same, a good AI should make deliberate typos.

Re: OpenAI Trains Language Model, Mass Hysteria Ensues

#105

Post by smsm42 »

ilyasut wrote: post #23

Ilya from OpenAI here. Here's our thinking: - ML is getting more powerful and will continue to do so as time goes by. While this point of view is not unanimously held by the AI community, it is also not particularly controversial. - If you accept the above, then the current AI norm of "publish everything always" will have to change - The _whole point_ is that our model is not special and that other people can reprodu…

> some of the samples generated by the model

Mostly it's scary not because it's good - as writing goes, it's quite bad. It forms coherent sentences, but otherwise it's nonsense. I've seen similar nonsense producers in early 90s on basis of Markov chains and what not.

No, the scary part is how much it reminds me of what I am reading in the media all the time. My current pet concern is that AIs will start passing the Turing test not because AIs are getting so good but because humans are getting so bad. A bunch of nonsensical drivel can easily be passed as a thoughtful analysis or a deep critical think-piece - and that's not my conjecture, have been repeatedly proven by submitting such drivel to various academic journals and it being accepted and published. I'm not saying people are losing critical thinking skills - but they are definitely losing (or maybe never even had?) the habit of consistently applying them.

Re: OpenAI Trains Language Model, Mass Hysteria Ensues

#106

Post by xiphias2 »

chrinic83 wrote: post #53
xiphias2 wrote: post #12

Elon Musk was kicked out because he poached Andrej Karpathy from OpenAI to lead Autopilot. Anyways, it was worth it, Andrej is doing an amazing job, and OpenAI is still alive :)

> Anyways, it was worth it, Andrej is doing an amazing job, and OpenAI is still alive :) Tesla does not even offer their full self driving package anymore. No coast to coast drive yet. Hard to say that's an amazing job. OpenAI abandons their open source GitHub repos after a year, is now not releasing code, and is always in DeepMind's shadow. Alive, yes. Successful, no.

Did you really expect Tesla to launch full driving? They started with being 5 years behind Waymo and without lidars or high resolution mapping, precise GPS sysyem that Waymo has...basically Elon wanted the impossible.

At the same time Andrej dropped out the idea of a fully learned end-to-end model (that's just impossible with the current deep learning technology), and started replacing the somewhat working heuristics with machine learning methodically one-by-one. Also he ramped up the data gathering pipeline.

He needs to build the full simulation, agent systems that can simulate other drivers/humans, implement reverse reinforcement learning...there's so much to do where Waymo is far ahead (but Tesla is ahead in data gathering).

Re: OpenAI Trains Language Model, Mass Hysteria Ensues

#107

Post by smsm42 »

avip wrote: post #35
ilyasut wrote: post #23

Ilya from OpenAI here. Here's our thinking: - ML is getting more powerful and will continue to do so as time goes by. While this point of view is not unanimously held by the AI community, it is also not particularly controversial. - If you accept the above, then the current AI norm of "publish everything always" will have to change - The _whole point_ is that our model is not special and that other people can reprodu…

I've just read i.e https://twitter.com/gdb/status/1096098366545522688 and even though it's "best of 25" (I guess cherry-picked by a human) - this is mind-blowing. I am actually having a very hard time believing this is legit generated text.

I read it and found it to be a bunch of walking in circles and repetitive baloney. It starts with a bunch of claims that is just the reversal of a pro-recycling poster and then goes into a repetitive meandering exploration about paper being made from materials, which is made from another materials. Probably something a model would regurgitate if fed with some popular literature about recycling. The most astonishing fact for me is that people actually think it's somehow surprisingly good.

Re: OpenAI Trains Language Model, Mass Hysteria Ensues

#108

Post by smsm42 »

anjc wrote: post #93
imtringued wrote: post #91

Earlier quoted context omitted.

I couldn't be more disappointed with this bullshit honestly. The texts have almost zero coherence and keep repeating the same patterns (which they presumably learned from the data set) over and over again. If this is their best out of 25 samples then they aren't going to fool anyone. >Recycling is NOT good for the world. >It is bad for the environment, >it is bad for our health, >and it is bad for our economy. >Recyc…

The parts you criticise are the parts I was most impressed with. These sorts of repetitions can be persuasive in writing/arguments, and it's impressive to me that a model learned this type of writing.

> These sorts of repetitions can be persuasive in writing/arguments

That is the saddest part. It's not because AI is good, it's because we count saying "X is good/bad" 3 times as a persuasive argument. It won't be hard to learn this kind of "arguing", it's just sad that's what we're teaching our AIs to do and get excited when they do it.

Re: OpenAI Trains Language Model, Mass Hysteria Ensues

#109

Post by esjeon »

smsm42 wrote: post #105
ilyasut wrote: post #23

Ilya from OpenAI here. Here's our thinking: - ML is getting more powerful and will continue to do so as time goes by. While this point of view is not unanimously held by the AI community, it is also not particularly controversial. - If you accept the above, then the current AI norm of "publish everything always" will have to change - The _whole point_ is that our model is not special and that other people can reprodu…

> some of the samples generated by the model Mostly it's scary not because it's good - as writing goes, it's quite bad. It forms coherent sentences, but otherwise it's nonsense. I've seen similar nonsense producers in early 90s on basis of Markov chains and what not. No, the scary part is how much it reminds me of what I am reading in the media all the time. My current pet concern is that AIs will start passing the T…

> I've seen similar nonsense producers in early 90s on basis of Markov chains and what not.

Exactly. When it comes to generating a large volume of apparently-good sentences, non-AI (or classical) approaches are still better than good. Those will be equally disruptive, since the defending side is yet to develop a proper countermeasure based on the "sensible"-ness of content. Plus, they will be much easier to customize and adapt to the situation, while ML-based solutions often need remodeling and retraining when repurposed.

> My current pet concern is that AIs will start passing the Turing test not because AIs are getting so good but because humans are getting so bad

AI will start deceiving the public even before it pass Turing test. It's much harder to spot bots amidst people than in a 1vs1 chatroom.

Re: OpenAI Trains Language Model, Mass Hysteria Ensues

#110

Post by iamcreasy »

ilyasut wrote: post #23

Ilya from OpenAI here. Here's our thinking: - ML is getting more powerful and will continue to do so as time goes by. While this point of view is not unanimously held by the AI community, it is also not particularly controversial. - If you accept the above, then the current AI norm of "publish everything always" will have to change - The _whole point_ is that our model is not special and that other people can reprodu…

Hello Ilya. Great work.

One of the reason Elon distanced himself because of what OpenAI team wanted to do. I am wondering if this new paper has anything to do with that? Or what it is in general that Elon doesn't agree with what OpenAI is doing?

Thanks!

Post reply on HN