Introducing ChatGPT and Whisper APIs • Hacker News Board

Re: Introducing ChatGPT and Whisper APIs

#551

Post by Tenoke » Thu, Mar 02, 2023, 6:33 AM UTC

alexb_ wrote: post #9

>Through a series of system-wide optimizations, we’ve achieved 90% cost reduction for ChatGPT since December This is seriously impressive. A MILLION tokens for 2 dollars is absolutely fucking insane. I hope that the gains reached here can also be found by open source and non-controlled AI projects. If so, that could be huge for the advancement of AI.

To be fair, it also seems like it performs worse. question: "Who is Scott Alexander?"

chatgpt-turbo:

>As an AI language model, I cannot determine the specific Scott Alexander in question without further contextual information. There may be several people with this name, including writers, politicians, entrepreneurs, athletes, and more.

chatgpt website:

>Scott Alexander is a pseudonym used by a prominent American psychiatrist and blogger who is best known for his writing on the blog "Slate Star Codex." He has also written articles for other publications, including The New York Times, The Atlantic, and The Wall Street Journal.

Scott Alexander's writing covers a wide range of topics, including medicine, psychiatry, economics, politics, and culture. He is known for his thoughtful, analytical approach to complex issues and for his ability to explain difficult concepts in a clear and accessible way.

In 2020, Scott Alexander announced that he was deleting his blog after he was doxxed by a reporter for The New York Times who planned to write an article about him. However, he later started a new blog called "Astral Codex Ten" where he continues to write.

Re: Introducing ChatGPT and Whisper APIs

#552

Post by gerash » Thu, Mar 02, 2023, 6:34 AM UTC

Is this sustainable or will the free content disappear from the web over time? because chatgpt does not generate revenue for the sources of content it uses.

OpenAI represents the opposite of open, responsible and non profit: it might as well be MS-AI

Re: Introducing ChatGPT and Whisper APIs

#553

Post by jgalt212 » Thu, Mar 02, 2023, 6:35 AM UTC

Garlef wrote: post #487

DoesntMatter22 wrote: post #420

Earlier quoted context omitted.

How is MJ making any money? Do they charge? When I went on their discord it was free.

you get 200 images for free. afterwards its $10, $30, $60 per month

Yeah, but what does one do with an infinite supply of ok-ish images?

Re: Introducing ChatGPT and Whisper APIs

#554

Post by cubefox » Thu, Mar 02, 2023, 6:37 AM UTC

vishal0123 wrote: post #470

cubefox wrote: post #452

Earlier quoted context omitted.

"We know the number of floating point operations per token for inference is approximately twice the number of parameters" Does someone have a source for this? (By the way, it is unknown how many parameters GPT-3.5 has, the foundation model which powers finetuned models like ChatGPT and text-davinci-003. GPT-3 had 175 billion parameters, but per the Hoffmann et al Chinchilla paper it wasn't trained compute efficiently…

https://arxiv.org/pdf/2001.08361.pdf . See the C_forward formula approxiamtion.

Thank you. Though it isn't quite clear to me whether the additive part is negligible?

Re: Introducing ChatGPT and Whisper APIs

#555

Post by 15155 » Thu, Mar 02, 2023, 6:38 AM UTC

Tepix wrote: post #500

rictic wrote: post #437

Earlier quoted context omitted.

Cost of power usage is marginal compared to that too: 300W per A100 * 8766 hours per year * $0.12 per kWh = $316 to power an A100 for a year

$0.12 per kWh is a very low price these days

Is this a low price for a datacenter negotiating their load with a utility provider (as most do?)

Re: Introducing ChatGPT and Whisper APIs

#556

Post by flangola7 » Thu, Mar 02, 2023, 6:51 AM UTC

theRealMe wrote: post #203

flangola7 wrote: post #149

Earlier quoted context omitted.

Agriculture reduced the global human economy/resource production doubling time from 100,000s of years to 1000s of years. Industrial revolution dropped it from 1000s to 10s or even 1s. If AI follows the same path it becomes 0.1 - 0.01 years. Your 401k wouldn't need 40 years to build a comfortable retirement, only 4 weeks.

I understand you’re joking, but surely it’s asymptotic to some multiple of human gestational periods.

This was not a joke. It's a serious concept that's been around for several years.

Re: Introducing ChatGPT and Whisper APIs

#557

Post by hiAndrewQuinn » Thu, Mar 02, 2023, 6:51 AM UTC

Hey, this is cool. Now that an official ChatGPT API is out maybe it's time for me to build that "Flavor text" Anki add-on I've been thinking about.

Re: Introducing ChatGPT and Whisper APIs

#558

Post by aidaman » Thu, Mar 02, 2023, 6:52 AM UTC

habitue wrote: post #166

Speculation: GPT-turbo is a new chinchilla optimal model with the equivalent capabilities as GPT-3.5. So it's literally just smaller, faster and cheaper to run. The reason I don't think it's just loss-leading is that they made it faster too. That heavily implies a smaller model.

it is a smaller model. it reveals this information if you ask it.

Re: Introducing ChatGPT and Whisper APIs

#559

Post by nodemaker » Thu, Mar 02, 2023, 6:53 AM UTC

wolverine876 wrote: post #542

nodemaker wrote: post #409

Earlier quoted context omitted.

In general trust is increased when all parties are speaking openly about whatever they are thinking.

It's a sort of philosophical idea - openness and free expression - taken to a logical and inhuman extreme. I cannot think of a situation where it is appropriate to say whatever I'm thinking. I think it would destroy trust, not least by demonstrating the unreliability of my judgment.

Here is a thought experiment for you. First think of the people you trust the most in this world, then imagine if they stopped speaking about whats on their mind with you. Would your trust in them increase or decrease?

Re: Introducing ChatGPT and Whisper APIs

#560

Post by aidaman » Thu, Mar 02, 2023, 6:54 AM UTC

dmw_ng wrote: post #68

rd wrote: post #8

This is incredibly cheap, it makes you wonder how in the world they managed to make it 10x cheaper than davinci-003 and still a better model? The world of robo-consulting is about to go insane.

It may just be that OpenAI corrected its 2020-era understanding of model size using 2022 insights from DeepMind.. https://towardsdatascience.com/a-new-ai-trend-chinchilla-70b... Seems model sizing, compression and quantization are still an art form, see also https://www.unum.cloud/blog/2023-02-20-efficient-multimodali...

yes turbo model admitted this when asked.