I see what you are looking to do now. To be honest, I don’t know if there is any way to make sure the model will not include wording you don’t like. It will be great if you can try the fine-tuning method and see if that is possible. I’ll like to know your results as well.
Another option I think will work as well is that regardless of using prompt engineering or fine-tuning, you can use either semantic search or regex expression to validate the results you get back from the model.
If it you see words you don’t like, you can deny the result and do a retry request with slightly different temperature.
Another word, since we can’t guarantee the response from GPT, we add a validation step as quality control.
Hope that helps, looking forward to your experiment results.
Olá Nelson, eu realmente tive os olhos brilhantes ao ver sua mensagem. De onde sou há um provérbio que diz: " Quem diz a verdade não merece castigo". Aqui vai minha verdade. Sou novato em AI e em python. Estou tentando resolver um problema que é o seguinte: Eu tenho um conjunto de dados de temperatura no formato .tif são imagens de satélite que datam de 2003 a 2020, estas imagens contem pixels de 1km quadrado. Estou desenvolvendo um projeto de ilhas de calor urbana com base na temperatura de superfície e estou precisando usar este conjunto de dados para fazer uma regressão linear para prever até 2023. Isto é realmente muito importante para mim. Meu objetivo é usar isto para prever as área mais quentes de uma cidade e como isto pode impactar a saúde das pessoas idosas. A dengue também é um fator de risco nestas áreas. Gentil você já tem demonstrado que é, não sendo um projeto que considere interessante, necessitaria de sua imensa generosidade. O que você acha disto? Muito obrigado, Davi (Brasil)
Olá @daviluca
Sorry I don’t speak Spanish but I did translate your question to understand your comment. It’s a very interesting problem but I think this is out of my knowledge scope. Maybe you can post this question in other AI community?
OK. I’ll let you know how it worked with finetuning.
If I understand you correctly, I can use regex to prevent certain words and phrases.
But my problem is that davinci partially twists the meaning.
For example
Client: I have no need for your offer at the moment.
davinci: I am glad to read that you have no interest in our offer at the moment.
What can I achieve with semantic search? What does it do? How do I apply it?
Muito obrigado Nelson por sua resposta tão solícita.
Eu tenho este outro problema que talvez possa ser interessante e possa gostar.
Gostaria de construir um aplicativo para mobile que pudesse, enquanto o automóvel estiver em movimento, filmar a estrada e capturar a geolocalização dos buracos para que este banco de dados fosse levado às autoridades competentes.
O que acha?
Gostaria de ouví-lo, desculpe por mandar em português.
@JohnMichaels Sure, perhaps I can send you an example on how to do that.
How will you plan to integrate with OpenAI’s API.
Will you be using the google spreadsheet I provided or will you use custom code, if so which language?
Thanks.
Hi @daviluca
That is a great idea and I think it is possible using vision detection.
Thank you for sharing your idea. For now, I will like to focus solely on OpenAI and NLP related areas. Perhaps you can post this on other community such as reddit and get others involved.
Best of luck to you, and happy new year.
Hi @nelson thanks for your generosity. Our project is a little bigger than the time you have until end of January. Besides, I don’t like taking people’s time for free. So are you open to work with us on a paid freelance basis?
Hi @Icebergtec
For UI/UX, I think you can look into no-code programming such as bubble.io and FlutterFlow.
Are you looking to use the Open AI for content generation?
Can you me more on how that can apply to matches, criteria, prompts, reviews for a dating app?
Thanks.
It’d be cool to see a GPT-3 based coding thing that could automatically run/test/regenerate limited code it generates from text descriptions as well, for example no writing files, accessing the internet, drawing to the screen outside of its window, etc… just safe logic based code. The language would not have to be human readable.
Thanks for sharing your project with me.
What kind of problems are you running into right now while using GPT?
Feel free to message me here or add me to your linkedin network so we can go over some of the details.
Thanks. https://www.linkedin.com/in/nelsonchu247/