6 ms·

It's very good that OpenAI is relenting and opening up the API; however the Content Guidelines are still too onerous such that even if you can think of a good u

It's very good that OpenAI is relenting and opening up the API; however the Content Guidelines are still too onerous such that even if you can think of a good use case it will be a liability at best even if your app gets approval. At this point (1.5 years later), if you're looking to make a sustainable business on AI text generation, you may want to experiment working with large-but-not-as-large models like GPT-J-6B; it'll be much cheaper too in the long run.

deadalus 5y ago
Another alternative AI21 studio (creators of wordtune[0]) also recently released their GPT3-like model called Jurassic-1 with 178B parameters and comparable results ( they also have a smaller 7B parameters model). Here is the whitepaper[1] with comparative benchmarks on some tasks . [0] : https://www.wordtune.com/ https://www.wordtune.com/ [1] : https://uploads-ssl.webflow.com/60fd4503684b466578c0d307/61138924626a6981ee09caf6_jurassic_tech_paper.pdf https://uploads-ssl.webflow.com/60fd4503684b466578c0d307/611...
- andybak 5y ago
  > AI21 Studio They fell into the common trap of "signed up, quite liked it but could never remember the name of it to find it again." Does anyone else suffer from this? (and bookmarks don't help - I've got thousands of them)
- seeekr 5y ago
  Blurb from the bottom of the wordtune landing page: "Wordtune was built by AI21 Labs, founded in 2018 by AI luminaries." Even if this was Tesla's marketing department saying something like "founded by engineering luminaries", clearly referring to the engineering genius that is Elon, I'd be hugely turned off, and would seriously reconsider my view of the company. But this is a company & product in the field of "AI", where there's so much bullshit floating around, unfortunately, so much hype and buzzword bingo, that writing in such tone about yourself seems like it should clearly be an absolute no-go -- unless you're just riding the snake-oil wave, so to speak, whether in good faith or not. Not implying anything about the company or product, of course, as I know nothing about them otherwise. EDIT: Maybe to clarify the thought behind the above further: It seems that the "AI" industry has an integrity problem. Language like this extends the problem, rather than working towards fixing it.
- gwern 5y ago
  Since Jurassic-1 is behind their API just like GPT-3, why do you think AI21 will not clamp down just as much as OA?
dqpb 5y ago
The content guidelines are so onerous I don’t even waste time imagining what I might do with the API. OpenAI somehow managed to leech all the joy out of GPT-3 with their own overbearing self righteousness. For an organization with so many RL engineers, they have a surprisingly poor understanding of the exploration/exploitation tradeoff.
mushufasa 5y ago
Yes. And building off of a close-source API from an organization that has flip-flopped already on being a nonprofit versus being a company to being part owned by microsoft seems like a bad idea. At least that's why I haven't used it in our business.
humanistbot 5y ago
Why did they even choose the name "OpenAI" if they didn't want to make openness part of their mission?
- sillycube 5y ago
  I think they better change the name. It looks like lying
- revolvingocelot 5y ago
  To sucker people into thinking that they were, or were going to. Isn't it obvious?
  - dnautics 5y ago
    it's like "light yogurt" where "light" can refer to the colour
    - coolspot 5y ago
      Or Full Self Driving(tm) where “full” can be read as “fool”
    - Turing_Machine 5y ago
      Or "wheat bread" (as opposed to "whole wheat bread"). Essentially all bread is made of wheat.
- reducesuffering 5y ago
  I remember someone involved saying they regret it. It's been six years. They evolved their understanding of the safety vs. openness tradeoff.
  - coolspot 5y ago
    They evolved their understanding of the profit vs openness tradeoff.
- 5y ago
seeekr 5y ago
As I suspect many of us frequently do, I read the comments (including yours) before the actual submission. I thought I would find myself agreeing with what you're saying, but it turns out that I must say that I really like what OpenAI is doing here with the Content Guidelines! They seem to be doing the right thing, in trying to steer this powerful and highly likely to turn out very influential piece of technology into a positive and constructive direction of use. Yes, you might just build something that will be found in violation of their (good!) intentions, and will have to engage in a (at least partially public) discussion of what we, as a society, deem acceptable in terms of automated use of written content generation -- and that would be a good thing! Definitely not the easiest path to make some $$$ based on new and exciting technology, as lots of challenges like these and beyond are almost guaranteed to come up, but it seems not unreasonable to treat GPT-3 as something you can actually already start building businesses and products on, as long as you bring general awareness, sensitivity to relevant topics, willingness to engage in and maybe partially drive some of the conversations that we need to have in this new field, along with a general interest in R&D style work and the somewhat longer-term vision and resources it necessitates...
- sillysaurusx 5y ago
  > They seem to be doing the right thing, in trying to steer this powerful and highly likely to turn out very influential piece of technology into a positive and constructive direction of use. It's not going to affect society. It's little more than a markov chain. OpenAI doesn't need to do anything to steer it. > Yes, you might just build something that will be found in violation of their (good!) intentions, You're giving them way too much credit. I've seen them destroy someone's business after repeatedly saying that their business model was fine. It was for an AI assisted writing app. Then they decided one day "Nope, you're not allowed to generate arbitrary amounts of text." After that, I was no longer a fan.
  - kludgebidness 5y ago
    Not on the side of OpenAI in the slightest, but you’re upset someone founded a business ostensibly reliant on an external business and their overall strategy — and then suffered the predictable outcome of that arrangement? This industry brought the faceless, API-driven, tickets-sans-phone-number business relationship into the world as the dominant model and then claims one side owes the other something in the same breath? I’ve been growing more and more concerned about obvious entitlement slipping into folks’ ideas about business, particularly in this forum. One of the big ones is that an offeror of services is generally compelled to continue offering services because stopping hurts another business. There’s a word for that: business. You can’t have a detached customer relationship model and “don’t hurt your customers” in the same ideology. It’s incoherent. Pick one. Here’s your algorithm to understand threats to an existential partnership of your business: 1) Are you lacking a contract? You suck. 2) Does your contract fail to compel the continued offering of services in a definitive way? You suck. 3) Does the same contract actually compel continuing service with no mitigating circumstances (like serving up hacked child porn, which is still your bad because they can justifiably say “secure your shit”)? They suck. Start with a demand letter and go from there. That’s literally it. There is no 4. You’re favored to suck two-thirds of the times this happens to you. How you lost respect for OpenAI because a writing app single homed their entire future and paid for it mystifies me, and I say that as someone who respects OpenAI very little. Just a remarkably stupid business model your friend executed given the available competitors to their things. It’s literally common sense risk analysis. What did your friend tell their investors? Or is this a bedroom app where you’re no longer a fan because someone lost out on a couple bills ARR from a trivially resurrectable idea? I’m thinking the latter, and #1 above. And don’t misunderstand me, I’m not advocating for the above: I’m explaining it. Key difference. You might find I agree with your overall point in terms of progressive business, but consider it naïve to not look at it the same way today.
HWR_14 5y ago
I'm confused. The Content Guidelines (in my skimming) reveal only 9 prohibited categories: Hate, Harassment, Violence, Self-Harm, Adult, Political, Spam, Deception and Malware. Am I missing something?
- minimaxir 5y ago
  Yes, but those are open to very broad and potentially inconsistent interpretations.
  - HWR_14 5y ago
    Can you give an example of how that works. I get there could be gray areas at the border (esp. with the "adult" content), but they seem tightly constrained to me.
kordlessagain 5y ago
Then again, there's using it to explore something from an independent research standpoint that's already irrational by finding the edges of unexplored rational thought within the conversation. That could be used, for example, to figure out an approach to something which is typically rejected by convention, but not by proof.

deadalus 5y ago

Another alternative AI21 studio (creators of wordtune[0]) also recently released their GPT3-like model called Jurassic-1 with 178B parameters and comparable results ( they also have a smaller 7B parameters model). Here is the whitepaper[1] with comparative benchmarks on some tasks . [0] : https://www.wordtune.com/ https://www.wordtune.com/ [1] : https://uploads-ssl.webflow.com/60fd4503684b466578c0d307/61138924626a6981ee09caf6_jurassic_tech_paper.pdf https://uploads-ssl.webflow.com/60fd4503684b466578c0d307/611...

andybak 5y ago
> AI21 Studio They fell into the common trap of "signed up, quite liked it but could never remember the name of it to find it again." Does anyone else suffer from this? (and bookmarks don't help - I've got thousands of them)
seeekr 5y ago
Blurb from the bottom of the wordtune landing page: "Wordtune was built by AI21 Labs, founded in 2018 by AI luminaries." Even if this was Tesla's marketing department saying something like "founded by engineering luminaries", clearly referring to the engineering genius that is Elon, I'd be hugely turned off, and would seriously reconsider my view of the company. But this is a company & product in the field of "AI", where there's so much bullshit floating around, unfortunately, so much hype and buzzword bingo, that writing in such tone about yourself seems like it should clearly be an absolute no-go -- unless you're just riding the snake-oil wave, so to speak, whether in good faith or not. Not implying anything about the company or product, of course, as I know nothing about them otherwise. EDIT: Maybe to clarify the thought behind the above further: It seems that the "AI" industry has an integrity problem. Language like this extends the problem, rather than working towards fixing it.
gwern 5y ago
Since Jurassic-1 is behind their API just like GPT-3, why do you think AI21 will not clamp down just as much as OA?

dqpb 5y ago

The content guidelines are so onerous I don’t even waste time imagining what I might do with the API. OpenAI somehow managed to leech all the joy out of GPT-3 with their own overbearing self righteousness. For an organization with so many RL engineers, they have a surprisingly poor understanding of the exploration/exploitation tradeoff.

mushufasa 5y ago

Yes. And building off of a close-source API from an organization that has flip-flopped already on being a nonprofit versus being a company to being part owned by microsoft seems like a bad idea. At least that's why I haven't used it in our business.

humanistbot 5y ago

Why did they even choose the name "OpenAI" if they didn't want to make openness part of their mission?

sillycube 5y ago
I think they better change the name. It looks like lying
revolvingocelot 5y ago
To sucker people into thinking that they were, or were going to. Isn't it obvious?
- dnautics 5y ago
  it's like "light yogurt" where "light" can refer to the colour
  - coolspot 5y ago
    Or Full Self Driving(tm) where “full” can be read as “fool”
  - Turing_Machine 5y ago
    Or "wheat bread" (as opposed to "whole wheat bread"). Essentially all bread is made of wheat.
reducesuffering 5y ago
I remember someone involved saying they regret it. It's been six years. They evolved their understanding of the safety vs. openness tradeoff.
- coolspot 5y ago
  They evolved their understanding of the profit vs openness tradeoff.
5y ago

seeekr 5y ago

As I suspect many of us frequently do, I read the comments (including yours) before the actual submission. I thought I would find myself agreeing with what you're saying, but it turns out that I must say that I really like what OpenAI is doing here with the Content Guidelines! They seem to be doing the right thing, in trying to steer this powerful and highly likely to turn out very influential piece of technology into a positive and constructive direction of use. Yes, you might just build something that will be found in violation of their (good!) intentions, and will have to engage in a (at least partially public) discussion of what we, as a society, deem acceptable in terms of automated use of written content generation -- and that would be a good thing! Definitely not the easiest path to make some $$$ based on new and exciting technology, as lots of challenges like these and beyond are almost guaranteed to come up, but it seems not unreasonable to treat GPT-3 as something you can actually already start building businesses and products on, as long as you bring general awareness, sensitivity to relevant topics, willingness to engage in and maybe partially drive some of the conversations that we need to have in this new field, along with a general interest in R&D style work and the somewhat longer-term vision and resources it necessitates...

sillysaurusx 5y ago
> They seem to be doing the right thing, in trying to steer this powerful and highly likely to turn out very influential piece of technology into a positive and constructive direction of use. It's not going to affect society. It's little more than a markov chain. OpenAI doesn't need to do anything to steer it. > Yes, you might just build something that will be found in violation of their (good!) intentions, You're giving them way too much credit. I've seen them destroy someone's business after repeatedly saying that their business model was fine. It was for an AI assisted writing app. Then they decided one day "Nope, you're not allowed to generate arbitrary amounts of text." After that, I was no longer a fan.
- kludgebidness 5y ago
  Not on the side of OpenAI in the slightest, but you’re upset someone founded a business ostensibly reliant on an external business and their overall strategy — and then suffered the predictable outcome of that arrangement? This industry brought the faceless, API-driven, tickets-sans-phone-number business relationship into the world as the dominant model and then claims one side owes the other something in the same breath? I’ve been growing more and more concerned about obvious entitlement slipping into folks’ ideas about business, particularly in this forum. One of the big ones is that an offeror of services is generally compelled to continue offering services because stopping hurts another business. There’s a word for that: business. You can’t have a detached customer relationship model and “don’t hurt your customers” in the same ideology. It’s incoherent. Pick one. Here’s your algorithm to understand threats to an existential partnership of your business: 1) Are you lacking a contract? You suck. 2) Does your contract fail to compel the continued offering of services in a definitive way? You suck. 3) Does the same contract actually compel continuing service with no mitigating circumstances (like serving up hacked child porn, which is still your bad because they can justifiably say “secure your shit”)? They suck. Start with a demand letter and go from there. That’s literally it. There is no 4. You’re favored to suck two-thirds of the times this happens to you. How you lost respect for OpenAI because a writing app single homed their entire future and paid for it mystifies me, and I say that as someone who respects OpenAI very little. Just a remarkably stupid business model your friend executed given the available competitors to their things. It’s literally common sense risk analysis. What did your friend tell their investors? Or is this a bedroom app where you’re no longer a fan because someone lost out on a couple bills ARR from a trivially resurrectable idea? I’m thinking the latter, and #1 above. And don’t misunderstand me, I’m not advocating for the above: I’m explaining it. Key difference. You might find I agree with your overall point in terms of progressive business, but consider it naïve to not look at it the same way today.

HWR_14 5y ago

I'm confused. The Content Guidelines (in my skimming) reveal only 9 prohibited categories: Hate, Harassment, Violence, Self-Harm, Adult, Political, Spam, Deception and Malware. Am I missing something?

minimaxir 5y ago
Yes, but those are open to very broad and potentially inconsistent interpretations.
- HWR_14 5y ago
  Can you give an example of how that works. I get there could be gray areas at the border (esp. with the "adult" content), but they seem tightly constrained to me.

kordlessagain 5y ago

Then again, there's using it to explore something from an independent research standpoint that's already irrational by finding the edges of unexplored rational thought within the conversation. That could be used, for example, to figure out an approach to something which is typically rejected by convention, but not by proof.