News for hackers. Stuff that matters.
Re:Spinning Up in Deep RL by colordrops on Monday August 17, 2020 @02:28AM (#24184549)
I assume this comment is generated? The link is a standard Sphinx doc.
Re:Spinning Up in Deep RL by Davidzheng on Monday August 17, 2020 @03:05AM (#24184729)
If that comment is generated, I will quit my current job and work full time on AI. I don't believe it.
Re:Spinning Up in Deep RL by 317070 on Monday August 17, 2020 @04:04AM (#24185069)
All the blogs posted by e.g. this user [0] were generated by GPT-3. [1] Some of those reached the top of HN.

That comment indeed looks a lot like it is generated. It has correlated a bunch of words, but it did not understand that the link between UI and AI is tenuous. It is probably one of the few comments where it is so glaringly obvious. There are likely a lot more comments around which are generated but which went unnoticed.

This comment is not generated, as the links below are dated after the GPT-3 dataset was scraped.

[0] https://news.ycombinator.com/submitted?id=adolos

[1] https://adolos.substack.com/p/what-i-would-do-with-gpt-3-if-...

Re:Spinning Up in Deep RL by dang on Monday August 17, 2020 @01:34PM (#24189528)
That story is bogus. See https://news.ycombinator.com/item?id=24165040 and https://news.ycombinator.com/item?id=24063832.

In a way the real story is that people are so eager to believe it that it didn't matter that it was untrue. Like Voltaire's God, if it didn't exist it was necessary to invent it.

Re:Spinning Up in Deep RL by amelius on Monday August 17, 2020 @06:22AM (#24185803)
Isn't gpt3-generated prose always seeded by some words or sentences? I'm curious what the poster used as seed, and why/if they chose to focus on UI?

Would it be possible to use gpt3 to "beautify" existing prose without changing its meaning? Now that would be useful!

Re:Spinning Up in Deep RL by rytill on Monday August 17, 2020 @10:23AM (#24187518)
The prompt design likely has as input the HN title, and potentially some other metadata, but not necessarily the first few words of the comment.
Re:Spinning Up in Deep RL by shawnz on Monday August 17, 2020 @08:07AM (#24186423)
Incredible. People in those threads are accusing the "author" of being GPT-3, and other commenters are chastising them for being so disrespectful as to suggest such a thing.
Re:Spinning Up in Deep RL by dang on Monday August 17, 2020 @01:34PM (#24189521)
Re:Spinning Up in Deep RL by shawnz on Monday August 17, 2020 @01:43PM (#24189653)
To be clear, I'm not saying that person was wrong to criticize the other post necessarily. I just think it's a testament to the abilities of GPT-3 that it could fool a reasonable person.
Re:Spinning Up in Deep RL by dang on Monday August 17, 2020 @02:34PM (#24190276)
That would be true if it had actually happened, but it didn't. What happened is that someone faked it to exploit people's desire for such a thing to happen, and then used the claim to get a lot of attention, including getting some journalists to write about it—whether because they actually believed it, or simply because it was such an attention-getter.
Re:Spinning Up in Deep RL by arthurcolle on Monday August 17, 2020 @11:39AM (#24188253)
Re:Spinning Up in Deep RL by thewarrior on Monday August 17, 2020 @09:57AM (#24187239)
AI is now good enough to generate comments like that so you probably should :)
Re:Spinning Up in Deep RL by lazzlazzlazz on Monday August 17, 2020 @04:52AM (#24185342)
It's definitely generated. But excellent nonetheless.
Re:Spinning Up in Deep RL by Findeton on Monday August 17, 2020 @02:44AM (#24184624)
And now we're going to need a turing test captcha.
Re:Spinning Up in Deep RL by slavik81 on Monday August 17, 2020 @03:19AM (#24184799)
I'd be more worried about bots gaming the voting. I'm perfectly happy to share this website with intelligent machines if they make insightful comments.

Whether odomojuli's post was written by a human, robot or dog is rather immaterial. It's the content of the post that makes it good or bad. It can be evaluated without knowing the author.

Re:Spinning Up in Deep RL by jimmySixDOF on Monday August 17, 2020 @08:21AM (#24186503)
>I'm perfectly happy to share this website with intelligent machines if they make insightful comments.

Not me. Not without total attribution so I know it's a bot and whose. There is no AI generated text etc without an agenda - implicit or explicit; benign or sinister.

Advancing the conversation is one thing but when u can't tell between a Russian GRU bot let loose to promote trigger words and some 4chan teenager who had a bad day at school then the online forum as a mode of expression is dead.

A 4channer is entitled to their opinion however wrong. A bot acting like it's human needs to be hunted, killed, and erased.

Re:Spinning Up in Deep RL by odyssey7 on Monday August 17, 2020 @03:43AM (#24184950)
Perhaps this will keep us on our toes. I tend to find something more believable if it's presented in nice prose (the AI's forte) -- which is really not how I should be evaluating information.
Re:Spinning Up in Deep RL by stareatgoats on Monday August 17, 2020 @03:04AM (#24184726)
And/or add a new bullet to the guidelines: "Please don't assume that a comment is auto generated without solid proof".
Re:Spinning Up in Deep RL by polytely on Monday August 17, 2020 @05:36AM (#24185567)
Add another bullet next to that: don't submit generated comments.
Re:Spinning Up in Deep RL by StavrosK on Monday August 17, 2020 @05:49AM (#24185635)
I wonder if the fact that we now can't trust anything we read is good or bad. Maybe it'll lead to critically thinking about everything.
Re:Spinning Up in Deep RL by 6gvONxR4sf7o on Monday August 17, 2020 @11:38AM (#24188249)
I suspect it's bad. Any cognitive dissonance and disagreement can be hand waved away as bots, rather than approached critically.

Ever since the 2016 foreign-meddling-in-the-election news, I've see people commenting that there must be tons of russian bots/shills/astroturfers/etc in comments where I see genuine disagreement. I'm sure there are both, but I suspect the dismissal of "'people/opinions' that disagree with me aren't real people/opinions" is more common than the actual act of fake commenting.

Re:Spinning Up in Deep RL by comboy on Monday August 17, 2020 @06:04AM (#24185702)
Could you ever?
Re:Spinning Up in Deep RL by StavrosK on Monday August 17, 2020 @06:08AM (#24185732)
I meant in the specific sense of trusting the basic humanity of HN, not the general veracity of the internet at large.
Re:Spinning Up in Deep RL by comboy on Monday August 17, 2020 @06:14AM (#24185769)
I would want to say that comments should be judged solely on their content, but yes, "redis works for us" is a different message from an account that you know is CTO at some big corp instead of some random one.

Web of trust is inevitable either way. I just hope it won't be owned by any huge company. Most likely it will be practically shared by a few of them, like the Internet or web standards.

After playing some with GPT-3 though I noticed I work a lot of like it though. Unrelated comments about web of trust are a great example.

Also, dude, I see you in almost every comment thread, how do you consume HN? Some clever scripts or lots of tabs and lots of refreshing?

Re:Spinning Up in Deep RL by StavrosK on Monday August 17, 2020 @06:21AM (#24185797)
Haha, I think I'm just more likely to comment than anyone else. I just read HN a few times during the day, nothing extravagant.

It might be confirmation bias (or maybe we just like the same things), if you look at my history I only leave a few comments a day.

Re:Spinning Up in Deep RL by yamrzou on Monday August 17, 2020 @05:32AM (#24185535)
An AI having their own personal qualms on OpenAI's ethics and virtues? I doubt it. It would be hilarious if this was generated.
Re:Spinning Up in Deep RL by ajfjrbfbf on Monday August 17, 2020 @02:57AM (#24184696)
> feel comfortable saying that the biggest obstacle in progress for AI is UI

Yep.

Re:Spinning Up in Deep RL by 4k05 on Monday August 17, 2020 @03:05AM (#24184734)
Do you have a source for this? Not nitpicking, i actually need it for a project...
Re:Spinning Up in Deep RL by BiasRegularizer on Monday August 17, 2020 @04:00AM (#24185046)
I believe he's referring to the absurdity of this statement, especially in relation to SpinningUp (it doesn't have any UI as it's a RL library, and the docs site are generated using standard Sphinx doc generator)
Re:Spinning Up in Deep RL by hobofan on Monday August 17, 2020 @08:08AM (#24186429)
OP might have just clicked through to the main OpenAI site and mixed the UI part in. The bulk of the comment is about documentation and presentation in general and viewing that in the context of the general great UX of OpenAI.

I know that GPT3 is impressive, but I'm not as convinced as some of the other commenters here that it's a generated comment. If a similar comment were posted on a non-AI related post, nobody would bat an eye.

Is there an equivalent to Hanlon's razor for AI? "Don't attribute to an text generation AI, which could be adquately explained by slightly nonsensical speech".

Re:Spinning Up in Deep RL by _5659 on Monday August 17, 2020 @02:31PM (#24190243)
That's exactly right, I should've been more astute to point out that it was Sphinx in my comment and dwell more on the general demeanor of the documentation with more examples.