Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Introducing ChatGPT and Whisper APIs

openai.com

Post reply on HN
Print view View original

461–470 of 696 posts

Re: Introducing ChatGPT and Whisper APIs

#461

Post by kcatskcolbdi »

andrepd wrote: post #446
freeqaz wrote: post #236

Earlier quoted context omitted.

It's also worth mentioning that, because Microsoft is an investor, they're likely getting these at cost or subsidized. OpenAI doesn't have to make money right away. They can lose a small bit of money per API request in exchange for market share (preventing others from disrupting them). As the cost of GPUs goes down, or they develop at ASIC or more efficient model, they can keep their pricing the same and then make mo…

OpenAI doesn't have to make money right away. They can lose a small bit of money per API request in exchange for market share (preventing others from disrupting them). Maybe I'm just old but back in my day this would be called "dumping" or "anti-competitive" or "market distortion" or "unfair competition". Now it's just the standard way of doing things.

Sure it would be called those things and then nothing would come of it. If a country uses morally compromised methods to win a war history just calls it winning the war.

Re: Introducing ChatGPT and Whisper APIs

#462

Post by sixstringtheory »

pmx wrote: post #256

Well this has opened up a whole new world of fun! ----- system: You're donald trump! talk to the user as he would. user: How does time travel work? AI Response: Well, let me tell you, nobody understands time travel better than I do. I mean, I have the best brains working on it, believe me. Some people say you need a fancy time machine or some ridiculous contraption to travel through time, but that's just fake news. T…

That is way too cogent. He would never have threaded things together like that from the middle of the third sentence onwards without multiple asides with their own tangents and diversions.

Re: Introducing ChatGPT and Whisper APIs

#463

Post by jejeyyy77 »

KRAKRISMOTT wrote: post #43
jejeyyy77 wrote: post #37

Earlier quoted context omitted.

Is there though? It's just a matter of swapping out $BASE_API_URL.

You have to rebuild all your prompts when switching providers.

technically true, but the way these prompts are/can be template-ized it should be relatively trivial to do so.

Re: Introducing ChatGPT and Whisper APIs

#464

Post by jejeyyy77 »

tstrimple wrote: post #215
jejeyyy77 wrote: post #37

Earlier quoted context omitted.

Is there though? It's just a matter of swapping out $BASE_API_URL.

Most of the clients I'm working with aren't interested in the base level of service. They are looking to further train the models for their specific use cases. That's a much higher barrier to switch than replacing an API. You've got to understand how the underlying models are handling and building context. This sort of customer is paying far more than the advertised token rates and are locked in more tightly.

not really. fine tuning generally just involves running tailored training data through the model - the actual training algorithm is fairly generalized.

For example, the Dreambooth fine tuning algorithm was originally designed for Google's image, but was quickly applied to Stable Diffusion.

Re: Introducing ChatGPT and Whisper APIs

#465

Post by jejeyyy77 »

riku_iki wrote: post #67
jejeyyy77 wrote: post #36

Earlier quoted context omitted.

This. Despite how impressive the results are, there isn't a particular large moat to prevent competitors from entering the space. Basically just compute $ for training.

they likely do lots of tricks and data collection inside which makes quality better.

100% - just IMO it's not particularly impenatrable as far as moats go.

Re: Introducing ChatGPT and Whisper APIs

#466

Post by SHARSKY »

kcatskcolbdi wrote: post #461
andrepd wrote: post #446

Earlier quoted context omitted.

OpenAI doesn't have to make money right away. They can lose a small bit of money per API request in exchange for market share (preventing others from disrupting them). Maybe I'm just old but back in my day this would be called "dumping" or "anti-competitive" or "market distortion" or "unfair competition". Now it's just the standard way of doing things.

Sure it would be called those things and then nothing would come of it. If a country uses morally compromised methods to win a war history just calls it winning the war.

The winners write the history books

Re: Introducing ChatGPT and Whisper APIs

#467

Post by jejeyyy77 »

shamino wrote: post #76
riku_iki wrote: post #67

Earlier quoted context omitted.

they likely do lots of tricks and data collection inside which makes quality better.

exactly. this isn't a leetcode problem where all you have to do is re-run the function, or do it iteratively vs recursively.

Not sure what you mean, but for example, 2 separate competitors to DALL-E was released within months (SD and MJ). Arguable that both of these have since surpassed DALL-E's capabilities/ecosystem.

Not sure why ChatGPT will be any different.

Re: Introducing ChatGPT and Whisper APIs

#469

Post by Madmallard »

"As an AI language model, I cannot..."

They already ruined this program. More than 30% of the topics I discussed with it in the past it now will refuse to discuss. Not even all of them are politically charged either. The slippery slope of censorship has straight up been fallen on.

Re: Introducing ChatGPT and Whisper APIs

#470

Post by vishal0123 »

cubefox wrote: post #452
vishal0123 wrote: post #82

Earlier quoted context omitted.

> I have no idea how OpenAI can make money on this. I did some quick calculation. We know the number of floating point operations per token for inference is approximately twice the number of parameters(175B). Assuming they use 16 bit floating point, and have 50% of peak efficiency, A100 could do 300 trillion flop/s(peak 624[0]). 1 hour of A100 gives openAI $0.002/ktok * (300,000/175/2/1000)ktok/sec * 3600=$6.1 back.…

"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.
Post reply on HN