9 ms·

How many attempts and how much human curation was involved. Every "I made AI do..." story seems to necessarily omit the "and I filtered out a LOT of rubbish bef

How many attempts and how much human curation was involved. Every "I made AI do..." story seems to necessarily omit the "and I filtered out a LOT of rubbish before picking this one good example."

pkdpic 4y ago
Sounds like the future of humanity maybe. Or the present.
qgin 4y ago
That could be the job description of the editor of the New Yorker
- noisy_boy 4y ago
  New job title: Chief Editor, AI Content Bonus points for outsourcing that to AI.
  - taftster 4y ago
    Upvote, because of the "bonus" sentence.
  - zhte415 4y ago
    AI is too complicated, it is too complex in order to build an algorithm that could be the machine for its own purposes. Comment generated by hncynic.leod.org
  - drusepth 4y ago
    Running novelgens.com basically gives me that job title. I occasionally build new features, but really all I do all day is review generated text/images and flag the good ones as "ready to release". :)
goodside 4y ago
As someone's who's played with GPT-3 prompting a lot, I can tell you almost no curation would be needed to get this result. This output is pretty mundane — it's even written in the perfunctory "so there you have it" tone that GPT-3 uses by default, where it speaks like the task is mutually understood to be silly and it's doing the the bare minimum needed to prove it understands. GPT-3 will easily write bare-bones stories like this for just about any subject. It's much harder getting it to write long stories with coherent structure.
goodside 4y ago
The following is an un-curated text-davinci-002 generation, completing the text: > Write an insightful reply to this Hacker News comment on a link to an article titled "I made an AI write a story about AI then I made another AI illustrate it". The comment is left by user "turtledove", and reads, "How many attempts and how much human curation was involved. Every 'I made AI do...' story seems to necessarily omit the 'and I filtered out a LOT of rubbish before picking this one good example.'" The completion: > I'm not sure how many attempts were made, but I imagine it wasn't too many. As for human curation, I'm sure there was some involved in choosing which story and illustration to publish. But overall, I think the point of the article is that the AI was able to produce something fairly coherent on its own.
- randomsearch 4y ago
  This is really very impressive. How likely is it this is essentially one or more comments from HN memorised by the neural net, with some synonym usage? How do we know that it’s not just reposting partially disguised sentences written by a human author?
  - Joeri 4y ago
    It is not as if every comment here is a gem of wholly original storytelling. For the most part we’re all just regurgitating other people’s sentences. Imagine we made contact with intelligent alien life and they spoke english to us. Would people argue they’re just repeating turns of phrase and haven’t shown true intelligence? I feel like we are setting the bar for AI too high. Current AI systems are clearly intelligent.
    - DrSiemer 4y ago
      Up to a degree. There is no real comprehension here. Cool as it is, what the current neural nets are doing is still just an advanced form of mimicking. Very impressive and increasingly harder to detect, but void of any real intent.
      - 4y ago
tablespoon 4y ago
> How many attempts and how much human curation was involved. Every "I made AI do..." story seems to necessarily omit the "and I filtered out a LOT of rubbish before picking this one good example." Yeah. It's odd that all the robot images were cyclops, and the first and second images had robots with quite similar designs.
Tepix 4y ago
It's a form of survivorship bias. If the results had been poor, he wouldn't have blogged about it.
tobiasSoftware 4y ago
Exactly. One of the more interesting things I've seen was when the streamer DougDoug used GPT-3 during a livestream to create a tournament where video game characters "fought" each other. While other content creators have gone the "you won't believe how good AI has gotten" route to get clicks and as a result heavily curated it, DougDoug has gone the opposite and embraced the insanity of the random text generation. As a result it was eye opening just how bad GPT-3 is at consistency. The tournament even ended with: "When Link is reduced to a pile of ashes, Kirby is victorious. Link stands there dazed by the attack." https://www.youtube.com/watch?v=PwY-jVSM-f0 https://www.youtube.com/watch?v=PwY-jVSM-f0
ttuttle 4y ago
Yes, I did do quite a few iterations on both GPT-3 and DALL-E 2. Its hard to say exactly how many because I also tweaked the prompt a few times, mainly to try and get a story that was longer than a few sentences and not just "I met an robot. It was scary at first, but now we are friends. The End." For the DALL-E 2 illustrations, I also re-rolled a handful of times to try and find a 'child' that looked relatively similar across the images.

How many attempts and how much human curation was involved. Every "I made AI do..." story seems to necessarily omit the "and I filtered out a LOT of rubbish before picking this one good example."

pkdpic 4y ago

Sounds like the future of humanity maybe. Or the present.

qgin 4y ago

That could be the job description of the editor of the New Yorker

noisy_boy 4y ago
New job title: Chief Editor, AI Content Bonus points for outsourcing that to AI.
- taftster 4y ago
  Upvote, because of the "bonus" sentence.
- zhte415 4y ago
  AI is too complicated, it is too complex in order to build an algorithm that could be the machine for its own purposes. Comment generated by hncynic.leod.org
- drusepth 4y ago
  Running novelgens.com basically gives me that job title. I occasionally build new features, but really all I do all day is review generated text/images and flag the good ones as "ready to release". :)

goodside 4y ago

As someone's who's played with GPT-3 prompting a lot, I can tell you almost no curation would be needed to get this result. This output is pretty mundane — it's even written in the perfunctory "so there you have it" tone that GPT-3 uses by default, where it speaks like the task is mutually understood to be silly and it's doing the the bare minimum needed to prove it understands. GPT-3 will easily write bare-bones stories like this for just about any subject. It's much harder getting it to write long stories with coherent structure.

goodside 4y ago

The following is an un-curated text-davinci-002 generation, completing the text: > Write an insightful reply to this Hacker News comment on a link to an article titled "I made an AI write a story about AI then I made another AI illustrate it". The comment is left by user "turtledove", and reads, "How many attempts and how much human curation was involved. Every 'I made AI do...' story seems to necessarily omit the 'and I filtered out a LOT of rubbish before picking this one good example.'" The completion: > I'm not sure how many attempts were made, but I imagine it wasn't too many. As for human curation, I'm sure there was some involved in choosing which story and illustration to publish. But overall, I think the point of the article is that the AI was able to produce something fairly coherent on its own.

randomsearch 4y ago
This is really very impressive. How likely is it this is essentially one or more comments from HN memorised by the neural net, with some synonym usage? How do we know that it’s not just reposting partially disguised sentences written by a human author?
- Joeri 4y ago
  It is not as if every comment here is a gem of wholly original storytelling. For the most part we’re all just regurgitating other people’s sentences. Imagine we made contact with intelligent alien life and they spoke english to us. Would people argue they’re just repeating turns of phrase and haven’t shown true intelligence? I feel like we are setting the bar for AI too high. Current AI systems are clearly intelligent.
  - DrSiemer 4y ago
    Up to a degree. There is no real comprehension here. Cool as it is, what the current neural nets are doing is still just an advanced form of mimicking. Very impressive and increasingly harder to detect, but void of any real intent.
    - 4y ago

tablespoon 4y ago

> How many attempts and how much human curation was involved. Every "I made AI do..." story seems to necessarily omit the "and I filtered out a LOT of rubbish before picking this one good example." Yeah. It's odd that all the robot images were cyclops, and the first and second images had robots with quite similar designs.

Tepix 4y ago

It's a form of survivorship bias. If the results had been poor, he wouldn't have blogged about it.

tobiasSoftware 4y ago

Exactly. One of the more interesting things I've seen was when the streamer DougDoug used GPT-3 during a livestream to create a tournament where video game characters "fought" each other. While other content creators have gone the "you won't believe how good AI has gotten" route to get clicks and as a result heavily curated it, DougDoug has gone the opposite and embraced the insanity of the random text generation. As a result it was eye opening just how bad GPT-3 is at consistency. The tournament even ended with: "When Link is reduced to a pile of ashes, Kirby is victorious. Link stands there dazed by the attack." https://www.youtube.com/watch?v=PwY-jVSM-f0 https://www.youtube.com/watch?v=PwY-jVSM-f0

ttuttle 4y ago

Yes, I did do quite a few iterations on both GPT-3 and DALL-E 2. Its hard to say exactly how many because I also tweaked the prompt a few times, mainly to try and get a story that was longer than a few sentences and not just "I met an robot. It was scary at first, but now we are friends. The End." For the DALL-E 2 illustrations, I also re-rolled a handful of times to try and find a 'child' that looked relatively similar across the images.