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

621–630 of 696 posts

Re: Introducing ChatGPT and Whisper APIs

#621

Post by Tenoke »

matteocontrini wrote: post #613
Tenoke wrote: post #551

Earlier quoted context omitted.

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 psyc…

Did you add the default ChatGPT system prompt at the beginning, when using the API?

I'm doing it via the openai library in the way they have in its docs.

>completion = openai.ChatCompletion.create(model="gpt-3.5-turbo", messages=[{"role": "user", "content": "Who is Scott Alexander?"}])

Re: Introducing ChatGPT and Whisper APIs

#622

Post by rmak »

manmal wrote: post #208
travisjungroth wrote: post #192

Earlier quoted context omitted.

If you think you're overpaying just hit the API yourself.

Any idea how to encode the previous messages when sending a followup question? E.g.: 1. I ask Q1 2. API responds with A1 3. I ask Q2, but want it to preserve Q1 and A1 as context Does Q2 just prefix the conversation like this? „I previously asked {Q1}, to which you answered {A1}. {Q2}“

ask chatgpt

Re: Introducing ChatGPT and Whisper APIs

#623

Post by matteocontrini »

Tenoke wrote: post #621
matteocontrini wrote: post #613

Earlier quoted context omitted.

Did you add the default ChatGPT system prompt at the beginning, when using the API?

I'm doing it via the openai library in the way they have in its docs. >completion = openai.ChatCompletion.create(model="gpt-3.5-turbo", messages=[{"role": "user", "content": "Who is Scott Alexander?"}])

Adding ChatGPT's initial prompt as a message with `system` role may make a difference (didn't try): https://platform.openai.com/docs/guides/chat/instructing-cha...

Also, we don't know ChatGPT's parameters (temperature, etc.).

Re: Introducing ChatGPT and Whisper APIs

#624

Post by qwertox »

graderjs wrote: post #597
qwertox wrote: post #576

Earlier quoted context omitted.

Last time I used it was last week ago, when I was comparing it to a docker install of Whisper/torch. Then again, I'm using it on very noisy audio recorded with a lavalier microphone while riding a bike. For transcription tests of recordings from an answering machine medium was more than enough, not sure about small. Also, for the recordings during a bike ride, sometimes medium is better than large. All this was used…

Transcript of 20230302 170602.m4a (at 2023-03-02 1711.02).txt Hey friend, your use case sounds really interesting. Actually that's why I created this app initially. I really love riding around on my bike in the city and doing voice memo debriefs about whatever. I also like to do it walking around outside. And as you say, the trouble with that is wind distortion. Full stop. On a day where it's not too windy, it's not…

I'm not preprocessing the recording. The thing I mentioned with considering VAD, if that's what you're referring to, is that I should somehow trim the files.

In order to minimize my interaction with my phone during the bike ride, I press a button which records 1 minute of audio. If I know that I need more time, I press it again before the minute ends, this then starts a second recording in parallel which also lasts one minute. So I just have to press a button and can forget about it. This is because I noticed that I usually don't require more than one minute to record a thought, and if I have multiple, I can put them in multiple files.

But since my recordings then usually consist of 20 seconds of audio, the 30 seconds at the end are only silence (with wind and tire noise). Whisper splits the files into 30 second segments, and apparently tries to find voice in each segment, so the remaining one which has no voice causes Whisper issues, where it starts hallucinating. This is why I would like to trim the files.

I now noticed that the service doesn't add punctuation and capitalization, so the funny thing is that I took that output and posted it into ChatGPT like this: "Correct the following: '[text from whisper]'", and it does an incredible job of fixing even words which Whisper erred on.

-

Whisper:

ich habe gestern erste tests mit open ai whisper gemacht um nozizen [sic!] zu transkribieren

[ Yesterday I did my first tests with open ai whisker to transcribe nozizen [sic!]. ]

es waren teilweise recht gute ergebnisse vor allem mit medium

[ there were some really good results, especially with medium ]

latsch [sic!] natürlich besser aber da sind die anforderungen zu hoch

[ latsch [sic!] better of course, but the demands are too high ]

wenn ich da einen server draus mache könnte ich mal eine zeit lang ausprobieren ob sich das lohnt

[ If I make a server out of it I could try it out for a while to see if it's worth it ]

auch für anrufe der anruf der antworten

[ also for calls the call of the answers ]

-

then ChatGPT:

Ich habe gestern erste Tests mit OpenAI's "Whisper" gemacht, um Notizen zu transkribieren. Die Ergebnisse waren teilweise recht gut, vor allem mit "Medium". "Large" funktioniert natürlich besser, aber die Anforderungen sind zu hoch. Wenn ich einen Server dafür bereitstelle, könnte ich mal für eine Zeit lang ausprobieren, ob sich das lohnt, auch für Anrufe und Antworten.

[ Yesterday I made first tests with OpenAI's "Whisper" to transcribe notes. The results were sometimes quite good, especially with "Medium". "Large" works better, of course, but the requirements are too high. If I provide a server for it, I could try it out for a while to see if it's worth it, also for calls and answers. ]

I'm sorry that this is in German, but I don't have anything in English I've been testing on.

Re: Introducing ChatGPT and Whisper APIs

#625

Post by gorbypark »

kierangill wrote: post #534
braingenious wrote: post #156

I hope this pricing impacts ChatGPT+ $20 is equivalent to what, 10,000,000 tokens? At ~750 words/1k tokens, that’s 7.5 million words per month, or roughly 250,000 words per day, 10,416 words per hour, 173 words per minute, every minute, 24/7. I uh, do not have that big of a utilization need. It’s kind of weird to vastly overpay

I use ChatGPT infrequently so the $20/month isn't worth it to me for pro. I stood up a open source & login-free UI here: https://www.chatwithme.chat/ It accepts your API token and stores it in your browser. It does not have feature parity with ChatGPT but gives you the basics.

Is this using the new GPT 3.5 turbo model?

Re: Introducing ChatGPT and Whisper APIs

#627

Post by counttheforks »

Doctor_Fegg wrote: post #577
komali2 wrote: post #520

Earlier quoted context omitted.

That seems to be changing. I've seen an uptick in criticism against the usa for unnecessarily (according to top military advisors, experts, generals etc at the time) dropping the atom bomb on Japan for example.

Absolutely. The bombing of Dresden has been viewed as a mistake - verging on a war crime - in Britain for the last 20 or so years.

Verging on? It was a mass murder of civilians. The US holocausted Japan and got away with it.

Re: Introducing ChatGPT and Whisper APIs

#628

Post by mach1ne »

npunt wrote: post #338
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…

Yeah we're in an AI landgrab right now where at- or below-cost pricing is buying marketshare, lock-in, and underdevelopment of competitors. Smart move for them to pour money into it.

Not very effective considering that it will be remade in open source 1-2 years from now.

Re: Introducing ChatGPT and Whisper APIs

#629

Post by karmasimida »

krelian wrote: post #33
polygamous_bat wrote: post #4

Earlier quoted context omitted.

> I have no idea how OpenAI can make money on this. This has to be a loss-leader to lock out competitors before they even get off the ground. The worst thing that can happen to OpenAI+ChatGPT right now is what happened to DallE 2, a competitor comes up with an alternative (even worse if it's free/open like Stable Diffusion) and completely undercuts them. Especially with Meta's new Llama models outperforming GPT-3, it…

I thought it was Midjourney who stole their thunder. Stable Diffusion is free but it's much harder to get good results with it. Midjourney on the other hand spits out art with a very satisfying style.

SD has won their plugin system crashed the game

Re: Introducing ChatGPT and Whisper APIs

#630

Post by thewataccount »

pharmakom wrote: post #591
freeqaz wrote: post #232

Earlier quoted context omitted.

Per instance (worker serving an API request) it requires 8x GPUs. I believe they have thousands of these instances and they scale them up with load. Because the model isn't dynamic (it doesn't learn) it is stateless and can be scaled elastically.

I expect some level of caching and even request bucketing by similarity is possible. How many users come with the same prompt?

In my experience running the same prompt always get's different results. Maybe they cache between different people but I'm not sure that'd be worth the cache space at that point? although 8x A100s is a lot to not have caching...
Post reply on HN