OpenAI API pricing update FAQ • Hacker News Board

Re: OpenAI API pricing update FAQ

#11

Post by guipsp » Mon, Aug 22, 2022, 5:42 PM UTC

powerhour wrote: post #9

intrasight wrote: post #7

Honest question: What does GPT-3 add to my tookbox as a developer?

I wonder if anyone has successfully used it to create library documentation. Obviously you'd have to tweak whatever output you get but can GPT-3 provide a substantial starting point?

One thing to note (without commenting if this applies in this case or not) is that sometimes a bad starting point is worse than no starting point

Re: OpenAI API pricing update FAQ

#12

Post by tehsauce » Mon, Aug 22, 2022, 5:42 PM UTC

minimaxir wrote: post #4

hda2 wrote: post #3

Any GPT-3-level fully-opensource models yet?

BLOOM and OPT are both GPT-3 sized models that are open source and similarily sized (175B parameters) but a) text generation is not as good as GPT-3 for whatever reason and b) getting them to run in a cost-effective manner is not easy and with this new price decrease, it may be more economical to use GPT-3 for deployment despite that.

The reason the similarly sized open source models aren’t as good is because gpt-3 has had a 2 year head start of additional development and improvement. The gpt-3 that is available today is Much better than what was initially produced in 2020.

Re: OpenAI API pricing update FAQ

#13

Post by minimaxir » Mon, Aug 22, 2022, 5:43 PM UTC

edgyquant wrote: post #8

minimaxir wrote: post #4

Earlier quoted context omitted.

BLOOM and OPT are both GPT-3 sized models that are open source and similarily sized (175B parameters) but a) text generation is not as good as GPT-3 for whatever reason and b) getting them to run in a cost-effective manner is not easy and with this new price decrease, it may be more economical to use GPT-3 for deployment despite that.

Is this due to entirely to economies of scale?

Not entirely. There are many ways to optimize models for deployment both at the software and hardware levels: some are public such as quantization/fusing, others are not.

Re: OpenAI API pricing update FAQ

#14

Post by dvfjsdhgfv » Mon, Aug 22, 2022, 5:43 PM UTC

The so-called "open" in their name is an oxymoron. The whole thing is a trap. We need to work on our own, possibly publicly founded but really open solutions rather than depend on proprietary ones.

Re: OpenAI API pricing update FAQ

#15

Post by tehsauce » Mon, Aug 22, 2022, 5:43 PM UTC

edgyquant wrote: post #8

minimaxir wrote: post #4

Earlier quoted context omitted.

BLOOM and OPT are both GPT-3 sized models that are open source and similarily sized (175B parameters) but a) text generation is not as good as GPT-3 for whatever reason and b) getting them to run in a cost-effective manner is not easy and with this new price decrease, it may be more economical to use GPT-3 for deployment despite that.

Is this due to entirely to economies of scale?

Economies of scale and years of optimization, some of which likely have not been publicly shared.

Re: OpenAI API pricing update FAQ

#16

Post by vanadium1st » Mon, Aug 22, 2022, 5:48 PM UTC

Why is it still closed for users in Ukraine? Are there many other countries excluded, or are we the only unlucky ones?

Re: OpenAI API pricing update FAQ

#17

Post by ianbutler » Mon, Aug 22, 2022, 5:49 PM UTC

powerhour wrote: post #9

intrasight wrote: post #7

Honest question: What does GPT-3 add to my tookbox as a developer?

I wonder if anyone has successfully used it to create library documentation. Obviously you'd have to tweak whatever output you get but can GPT-3 provide a substantial starting point?

I've used it in a previous tool to generate documentation from code snippets, it works out pretty well.

Re: OpenAI API pricing update FAQ

#18

Post by karamanolev » Mon, Aug 22, 2022, 5:50 PM UTC

arbol wrote: post #5

Any chance of this making GitHub copilot cheaper?

Does Copilot use GPT-3 directly? I haven't found resources on this, but sounds doubtful?

Re: OpenAI API pricing update FAQ

#19

Post by JimmyAustin » Mon, Aug 22, 2022, 5:51 PM UTC

AJRF wrote: post #6

> We have been looking forward to reducing pricing for a long time. Our teams have made incredible progress in making our models more efficient to run, which has reduced the cost it takes to serve them, and we are now passing these savings along to our customers. Nothing to do with Stable Diffusion then, no?

Given that Dall-E 2's pricing hasn't changed, probably not.

Re: OpenAI API pricing update FAQ

#20

Post by naillo » Mon, Aug 22, 2022, 5:52 PM UTC

dvfjsdhgfv wrote: post #14

The so-called "open" in their name is an oxymoron. The whole thing is a trap. We need to work on our own, possibly publicly founded but really open solutions rather than depend on proprietary ones.

We really gotta come up with another thing to repeat as the go-to diss than this. I'm all for putting pressure on them to release weights etc but they are a lot more open than e.g. google or meta, and the whole "ClosedAI" is so over-repeated at this point.