OpenAI GPT-2: An Almost Too Good Text Generator!

📝 The paper "Better Language Models and Their Implications" is available here:

GPT-2 Reddit bot:

Criticism:

The Bitter Lesson video:

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
313V, Alex Haro, Andrew Melnychuk, Angelos Evripiotis, Anthony Vdovitchenko, Brian Gilman, Bruno Brito, Bryan Learn, Christian Ahlin, Christoph Jadanowski, Claudio Fernandes, Dennis Abts, Eric Haddad, Eric Martel, Evan Breznyik, Geronimo Moralez, Javier Bustamante, John De Witt, Kaiesh Vohra, Kasia Hayden, Kjartan Olason, Levente Szabo, Lorin Atzberger, Marcin Dukaczewski, Marten Rauschenberg, Maurits van Mastrigt, Michael Albrecht, Michael Jensen, Morten Punnerud Engelstad, Nader Shakerin, Owen Campbell-Moore, Owen Skarpness, Raul Araújo da Silva, Richard Reis, Rob Rowe, Robin Graham, Ryan Monsurate, Shawn Azman, Steef, Steve Messina, Sunil Kim, Thomas Krcmar, Torsten Reil, Zach Boldyga, Zach Doty.

Károly Zsolnai-Fehér's links:

#OpenAI #GPT3 #GPT2

[Dixiklo-yltg]

"This of course means that we are late to the party and missed out on a whole lot of views and revenue, but that's okay, in fact, that's what we'll keep doing going forward to make sure you get the highest quality information that I can provide."

And that's why you're the best. Love you, KZF.

[hernanescu]

Long gone are the days when we were afraid of GPT-2... how innocent we were :P

[MichaelSHartman]

Maybe we have something that can read Terms and Conditions legalese then summarize.

[markosz]

"Hey, why are you not writing your thesis work? It's not going to write itself!"


Hold my beer.

[morthim]

why wasn't the algorithm called "most journalists"?

[olfmombach]

2:04
"While their origins are still unclear, some believe that perhaps the creatures were created when a human and a unicorn met each other in a time before human civilization."
Lol

[oddarneroll]

This is me from the future, we now have general access to gpt-4. And it really blows my mind. Its like having a personal expert college that never tires and is happy to even do my work for me.

[Maouww]

THE LotR FANFIC!! hahahahahahaha
Dear Frodo. I hope you're alive and not a bad guy, but if you are a bad guy, ok.
GOLD!

[KaaBockMehr]

imagine what beautiful roleplay games can be created with a computer giving you useful answers for each question you have..

[JustTryAnother]

few more month and it will be able to pass TOEFEL )))

[kuretaxyz]

I think the ethical thing to do is to release the source code. In the past people had the same argument over encryption and now everything pretty much depends on open implementations of RSA and similar algorithms. If as the authors are telling the truth when they are saying "a top tier research group would be able to replicate the results" then keeping the source secret means only a select group will have access to this technology, at least for a while. If the public has access to the code early on, people might develop some sort of a fake text detector, similar to image manipulation detectors that already exists.

[NicosLeben]

This thing should read the whole Wikipedia. And then we can ask him questions.

[BlackXxScopez]

it's funny how several people can come to the same decisions independently. When I created a markov chain analyzer for text synthesis, I used the works of shakespeare as my dataset. And now that I'm getting into and starting to understand neural networks, you know exactly what I'm going to be using to train my network

[vvilliam]

checking in on this after GPT-3's release to see if it's "good enough" for me to play around with.

[braelyn.b__]

I think you mispronounced "Seven and a Half Minute Papers"

[egrinant]

So this is a fake news generator. I see a fair application: a helper for writers (but definitely not to be abused). But yeah, basically everyone would use it for profit.

[mattiasfagerlund]

Love your stuff, but you used to describe how a method works, not just the results. Any chance you could get back to a more technical description of the methods in the papers?

[Geddy]

a lot of people commenting on how this can be used to be lazy with homework or used for nefarious deeds
and I'm sitting here thinking of all the entertainment value. Imagine having an idea for a story and all you have to do is give it a summery then bam, infinite things to read specifically tailored to you.

[MobyMotion]

I have mixed feelings about OpenAI turning into a for profit company, but if it means more papers like this that leverage a huge amount of (very expensive) compute, I think it's a good thing on balance

[Prisal]

damn... how far we've come in a reemarkable short time