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

611–620 of 696 posts

Re: Introducing ChatGPT and Whisper APIs

#611

Post by CapsAdmin »

Tenoke wrote: post #551
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 psyc…

Can we really draw any conclusions on LLMs based on 1 sample? Maybe you've tried multiple times and with different semi famous people, but in general I see people comparing ML models in this fashion.

Re: Introducing ChatGPT and Whisper APIs

#612

Post by Tenoke »

CapsAdmin wrote: post #611
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…

Can we really draw any conclusions on LLMs based on 1 sample? Maybe you've tried multiple times and with different semi famous people, but in general I see people comparing ML models in this fashion.

Not really, I did try it with multiple attempts with multiple people and chatgpt had more issues. I just shared only one of them. If someone tests in a more systematic fashion that'd be great.

Re: Introducing ChatGPT and Whisper APIs

#613

Post by matteocontrini »

Tenoke wrote: post #551
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 psyc…

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

Re: Introducing ChatGPT and Whisper APIs

#614

Post by Havoc »

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.

90% for a presumably already semi efficient setup is insane

Re: Introducing ChatGPT and Whisper APIs

#615

Post by graderjs »

wriggler wrote: post #608
graderjs wrote: post #522

Wow not a single mention of Whisper this entire comment first page! I think Whisper is really cool: the large model can pull speech out of even heavily distorted (wind noise, clipping, etc) audio. I have a story to illustrate why running Whisper on your own locally is not so easy! Much easier to sign up to the OpenAI API. In my research I found that actually pre-processing the audio to reduce noise (using the IMO bes…

Hey, nice app. Can you add support for non-English transcription/translation, though?

Thanks! So, OK, if you write a review and provide your honest feedback about this on the App store, I will definitely consider doing that! Sound like a bad idea? :)

Re: Introducing ChatGPT and Whisper APIs

#616

Post by _andrei_ »

comment_ran wrote: post #336
b5n wrote: post #283

Earlier quoted context omitted.

The future is now: gptask() { data=$(jq -n \ --arg message "$1" \ '{model: "gpt-3.5-turbo", max_tokens: 4000, messages: [{role: "user", content: $message}]}') response=$(curl -s https://api.openai.com/v1/chat/completions \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $OAIKEY" \ -d "$data") message=$(echo "$response" \ | jq '.choices[].message.content' \ | sed 's/^\"\\n\\n//;s/\"$//') echo -e "$mess…

Not to sure how to make it one section, aka, one session mode, so chatgpi can understand the previous talk.

quick script

    #!/usr/bin/env bash
    set -uf -o pipefail
    IFS=$'\n\t'

    BOT='\033[33m'
    NC='\033[0m'

    messages=()

    trim() {
      local var="$*"
      var="${var#"${var%%[![:space:]]*}"}"
      var="${var%"${var##*[![:space:]]}"}"
      printf '%s' "$var"
    }

    function complete {
      local message="$1"
      local data
      messages+=("{\"role\": \"user\", \"content\": $(echo "$message" | jq -R -s '.')}")
      processed_messages=$(printf '%s,' "${messages[@]}")
      processed_messages="[${processed_messages::-1}]"
      data=$(jq -n \
        --arg model "gpt-3.5-turbo" \
        --argjson messages "$processed_messages" \
        '{ model: $model, messages: $messages }' \
        | sed 's/\]\[/,/g')

      response=$(curl -s https://api.openai.com/v1/chat/completions \
        -H "Content-Type: application/json" \
        -H "Authorization: Bearer $OPENAI_API_KEY" \
        -d "$data")

      message=$(echo "$response" | jq '.choices[].message.content')
      printable_message=$(echo "$response" | jq -r '.choices[].message.content')
      printable_message=$(trim "$printable_message")
      echo -e "${BOT}Bot:${NC} $printable_message"

      messages+=("{\"role\": \"assistant\", \"content\": $message}")
    }

    while true; do
      read -r -p $'\e[35mYou:\e[0m ' message
      complete "$message"
    done

Re: Introducing ChatGPT and Whisper APIs

#617

Post by RugnirViking »

CamperBob2 wrote: post #272
mritchie712 wrote: post #139

Earlier quoted context omitted.

https://news.ycombinator.com/item?id=34972791

That's just a list of exploits that will be fixed as soon as they come to OpenAI's attention, if they haven't already been fixed. Is anyone actually committed to providing uncensored models as either paid services or open distributions?

Maybe try GPT-J, its an open source model you can run locally. It's pretty massive though iirc

Re: Introducing ChatGPT and Whisper APIs

#618

Post by hifikuno »

awestroke wrote: post #585
jfan001 wrote: post #294

We've been struggling with costs because our application chains together multiple calls to GPT to generate the output we want, and it was starting to be ~$0.08 per call which obviously isn't feasible for high volume applications. This just made our business way more viable overnight lmao

lmao?

laughing my ass off

Re: Introducing ChatGPT and Whisper APIs

#620

Post by ChildOfChaos »

pksebben wrote: post #572
ChildOfChaos wrote: post #239

Earlier quoted context omitted.

While this is cool this doesn't change my opinion at all. Each still image is still not that impressive. Good for them using the tech in a clever way but i don't find this that relevant.

those trees are part of a forest, you know. the benefits of such fine grained control aren't a trick. it's why they were able to scrap together frames that don't jump all over the place (mostly). the other benefit of such a broadly hacked upon model is that it grows in leaps and bounds. All due respect to mid journey, but the stable diffusion hype is not just hype.

I agree, don't believe it's just hype, that level of control is useful, but for outright image quality and for most use cases, midjourney is better.

I still don't like the look of most of the Stable diffusion images, they just look slightly off/amateurish to me, where as midjourney produces images that make you go 'wow'

If you wanted to use these tools, midjourney would be my go too, with stable diffusion a backup for when some of the additional features were needed, perhaps inpanting on a midjourney image and using controlnet if needed but if you just want a pure image, midjourney is what you want.

Post reply on HN