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

OpenAI Releases Largest GPT-2 Text Generation Model

openai.com

Post reply on HN
Print view View original

131–140 of 166 posts

Re: OpenAI Releases Largest GPT-2 Text Generation Model

#131

Post by pure-awesome »

krick wrote: post #91

Wow, some samples are frighteningly good. I was impressed by previous models and I don't know if I'm just lucky this time, but... wow. Can anybody who is not into climbing even tell this is all fake? Jain Kim is an experienced climber. In 2006, she became the first woman from Korea to climb all five 8,000 meters (24,064 ft) peaks in the Swiss alpine ski run Alps in 24 hours. In 2009, she made history again by setting…

> Can anybody who is not into climbing even tell this is all fake?

Yes, quite clearly from the following:

> She made the first ascent of 8,832-meter K2 in China, the second highest mountain in the world, in 2009 and the third highest mountain in Europe.

Firstly, this sentence scans poorly. I'm guessing it should be:

> In 2009, she made the first ascent of 8,832-meter K2 in China, the second highest mountain in the world, and the third highest mountain in Europe.

Second, how can a mountain in China be the third highest mountain in Europe? How can the second highest mountain the world be the third highest in Europe?

If I came across this in the wild, then even if I didn't think it was fake, I'd definitely think it was poorly proofread.

Re: OpenAI Releases Largest GPT-2 Text Generation Model

#132

Post by csomar »

vicaya wrote: post #115

Just like various chat bot incarnations, it's reasonably OK at generating generic sentences without real understanding. I tried several attempts with this leading fragment: "Is borrow checking unique to Rust? Probably not. But". The results are not pretty (the best one is as follows). Not impressed at all with this giant model :| "Is borrow checking unique to Rust? Probably not. But the Rust library is written by and…

Am I the only one impressed by the text generated? Sure, it doesn't have any understanding but are you factoring that 1. most people on the world do not know that Rust is a programming language and 2. a single person can not have that much general knowledge. Sure they can know about Rust borrow checker but will not be able to expand that much on another subject.

Re: OpenAI Releases Largest GPT-2 Text Generation Model

#133

Post by genidoi »

ssivark wrote: post #122
vicaya wrote: post #115

Just like various chat bot incarnations, it's reasonably OK at generating generic sentences without real understanding. I tried several attempts with this leading fragment: "Is borrow checking unique to Rust? Probably not. But". The results are not pretty (the best one is as follows). Not impressed at all with this giant model :| "Is borrow checking unique to Rust? Probably not. But the Rust library is written by and…

Wow! That is meaningless but difficult to distinguish at a glance, especially when read by someone not familiar with the subject. Are we going to see such auto generated content take over as the primary pillar of the SEO content farms? ... Kinda like an automated (text) version of Siraj Raval videos :-P

It's interesting how the first paragraph in particular reads like a deflection of the original question, somewhat akin to a middle manager avoiding exposing their lack of domain expertise in response to a technical question by giving a vague answer.

There were heaps of semantic pathways the network could have chosen such as guessing what a borrower is, or maybe even weighing the significance of "borrower" more than rust in the original question because it has no clue what rust+borrower is; but it instead picked the safer option of deflection. Fascinating stuff.

Re: OpenAI Releases Largest GPT-2 Text Generation Model

#134

Post by lifthrasiir »

mrob wrote: post #125
krick wrote: post #96

Earlier quoted context omitted.

Yeah-yeah, and "Lomonosov Ridge" is a weird name for a peak in China (and you might even happen to know the Lomonosov Ridge is underwater). Please don't be distracted by details. Maybe the author meant 24 h per peak, maybe his typing is really spurious. Is it really obvious to somebody who doesn't know anything about climbing that this is not true? Come on. What's really fascinating is how truth is mixed up with made…

>8,000 meters is 24,064 ft It's 26,247 ft.

I'm kinda surprised that, 8,000 m = 24,064 ft implies the conversion factor of 1 ft = 30.08 cm instead of the correct 30.48 cm (note a single digit change). It... it might be capable of making a humane calculation mistake. (I think it is entirely possible, but still surprising though.)

Re: OpenAI Releases Largest GPT-2 Text Generation Model

#135

Post by pengstrom »

buboard wrote: post #11

> (CTEC) found that extremist groups can use GPT-2 for misuse, specifically by fine-tuning GPT-2 models on four ideological positions: white supremacy, Marxism, jihadist Islamism, and anarchism. CTEC demonstrated that it’s possible to create models that can generate synthetic propaganda for these ideologies I wonder how they tested that

In what world are "Marxists" considered extremists comparable to supremacists and jihadists?

Re: OpenAI Releases Largest GPT-2 Text Generation Model

#136

Post by stared »

vicaya wrote: post #115

Just like various chat bot incarnations, it's reasonably OK at generating generic sentences without real understanding. I tried several attempts with this leading fragment: "Is borrow checking unique to Rust? Probably not. But". The results are not pretty (the best one is as follows). Not impressed at all with this giant model :| "Is borrow checking unique to Rust? Probably not. But the Rust library is written by and…

Well, for detailed knowledge it is not enough.

But for Internet ramblings about anything (politics, religion, capitalism vs socialism), I bet it is well beyond the average human level. (If you want to protest, go to some random Facebook, YouTube, Reddit or Twitter thread. No, not HN, or specialized groups of interests, or anything dominated by academics or IT specialist.)

Also, somewhat related: https://news.ycombinator.com/item?id=21438318 "Undercover reporter reveals life in a Polish troll farm"

A few friends of mine became parents and started participating in some parenting FB groups. It was from them a shocking contact with getting outside of the intellectual bubble.

I would be really interested in judging the quality of GPT-2-generated tests against human texts. Questions like "does the person know what they talk about?", "are they smart?" with control on the knowledge of a particular subject (e.g. do they know Rust?) would give some insight into the effective level of AI for text generation.

Re: OpenAI Releases Largest GPT-2 Text Generation Model

#137

Post by speedplane »

stared wrote: post #136
vicaya wrote: post #115

Just like various chat bot incarnations, it's reasonably OK at generating generic sentences without real understanding. I tried several attempts with this leading fragment: "Is borrow checking unique to Rust? Probably not. But". The results are not pretty (the best one is as follows). Not impressed at all with this giant model :| "Is borrow checking unique to Rust? Probably not. But the Rust library is written by and…

Well, for detailed knowledge it is not enough. But for Internet ramblings about anything (politics, religion, capitalism vs socialism), I bet it is well beyond the average human level. (If you want to protest, go to some random Facebook, YouTube, Reddit or Twitter thread. No, not HN, or specialized groups of interests, or anything dominated by academics or IT specialist.) Also, somewhat related: https://news.ycombina…

> Well, for detailed knowledge it is not enough. .. But for Internet ramblings about anything (politics, religion, capitalism vs socialism), I bet it is well beyond the average human level.

The real question is, how far apart are these? It certainly seems that there is a wide gulf, but the technology is moving incredibly rapidly, unclear if it's 3 years, 5 years, or 100 years away.

Re: OpenAI Releases Largest GPT-2 Text Generation Model

#138

Post by jpdus »

krick wrote: post #96
bonoboTP wrote: post #93

Earlier quoted context omitted.

> Can anybody who is not into climbing even tell this is all fake? Hm, I'm pretty sure it's hard to climb five 8000 meter peaks in 24 hours :)

Yeah-yeah, and "Lomonosov Ridge" is a weird name for a peak in China (and you might even happen to know the Lomonosov Ridge is underwater). Please don't be distracted by details. Maybe the author meant 24 h per peak, maybe his typing is really spurious. Is it really obvious to somebody who doesn't know anything about climbing that this is not true? Come on. What's really fascinating is how truth is mixed up with made…

K2 is not in China ;-)

Re: OpenAI Releases Largest GPT-2 Text Generation Model

#140

Post by thesz »

csomar wrote: post #132
vicaya wrote: post #115

Just like various chat bot incarnations, it's reasonably OK at generating generic sentences without real understanding. I tried several attempts with this leading fragment: "Is borrow checking unique to Rust? Probably not. But". The results are not pretty (the best one is as follows). Not impressed at all with this giant model :| "Is borrow checking unique to Rust? Probably not. But the Rust library is written by and…

Am I the only one impressed by the text generated? Sure, it doesn't have any understanding but are you factoring that 1. most people on the world do not know that Rust is a programming language and 2. a single person can not have that much general knowledge. Sure they can know about Rust borrow checker but will not be able to expand that much on another subject.

This looks like it can copy large chunks of text. That's it.

It is hard task in itself ("copy" is a standard test for memory-enhanced neural systems like RNN and such) but here it needs to "understand" things. For example, what makes borrow checker in Rust unique which would be valid continuation.

Post reply on HN