My name is José Ícaro, I’m a Dentist, I develop solutions in WordPress for 13 years and since 2020 I experience a trajectory of “Founder of Startup” in /solo/ mode.
I have a 3-year-old son named João Gabriel and loves to create images with his dad.
I’ve always loved participating in forums and virtual communities, since the 90’s I’ve been part of 3 large geek communities here in Brazil and I’m loving the format of the OpenAI forum, maybe now I can get more time and definitely try to participate more.
5 months ago when I started automations using simple settings with the OpenAI API, I received an audio from a person who was close to our main group who made me stop, breathe, save the audio and send it to my mom (yes I love my mom and she’s with me at all times possible).
The audio spoke next target : “What fantastic, this is going to be wonderful, how did you do it? Put my e-mail and the %user2% and %user3% for us to take a look and help you expand it!”.
This is all because of Long Prompts and Short Prompts. Perhaps.
At that time I was starting to crawl with a very simple setup in power automate. Two days later I was already using Copilot with Codex and GPT-3 to help me understand the codes being generated. I put everything aside and went back to programming and i felt happy, it’s been a long time since I experienced this feeling of relief. I attended 5 other universities, one of them was Computer Science, I program in C and C++ and use computer since I was 4 years old, today I am 36. The fact is that in 2010 I had chosen not to continue in computing the way it would have to be, I did not want to go to mathematics, calculus 3 and computer graphics at the time gave me chills and of course, were other times.
A month ago, I started implementing a front-end scenario to request application approval for OpenAI. I submitted an application idea, which is our repository of bug solutions and tips from WP, AI and ML, but that was in July and a scenario to offer People + AI connections via GPT-3 requires much more than I imagined.
I ran out of income, i chose not to sell or advertise the courses, and the time was ticking, the obligations were coming, I missed two important meetings with Microsoft, my son got sick for almost 30 days, my main computer (workstation) suffered a failure in the TPM and I ended up losing a motherboard, and I kept thinking about the codes and how I would do to turn the Long Prompts into something to help me and at the same time help people.
I started thinking about how people with vision disabilities would love to have an aid to expand everyday scenarios, or important things.
I started looking around, and I remember my dad when I was in the ICU in 2019 who tried to talk to me but couldn’t and to this day I don’t know what he wanted to talk about, but I’m sure they were good things.
We are Humans
It’s amazing how people prefer to look directly at something that wasn’t made for it, and I’m not referring to GPT-3 or an AI, but rather our ephemeral existence.
During late August and September I had a kind of “bug-blackout” and spent a few days trying to imagine something that could be marketed and do people good.
The Fear passes.
In October I was able to find a tool to calm my concerns about creating a secure application, but in a short time and protect the functions and prompts, it was then that I started researching on frameworks, react, nextjs and tested almost 30 tools. We received an invite from Retool and in 1 day I managed to virtualize the sketches and organize some of what was on the boards of Miro, Notion and Airtable.
At the same time, I discovered that several indigenous languages are not part of GPT-3, currently there are more than 50 million Indians in Latin America, with hundreds of indigenous languages and dialects, in Brazil there are about 5 million Indians and more than 170 dialects and languages, the main one being Tupi Guarani.
But what about the Long Prompts?
According to GPT-3, up to the subtitle above are 5,000 characters, so we already have a long prompt in history, even with a possible margin of error in the character count and format of the prompt I used.
Since January I am part of another context of life, were dozens of mind maps, we have a project focused on part of dentistry involving image detection to help the federal dental board reduce the amount of profiles and false posts, as I am hyperactive, sometimes do not understand simple things like “buy bread in the bakery and milk” and I return with milk, candy + other things – bread, so I had to go back in time, to understand the current processors and associate to content with Deep Learning, ML, etc.
In my case, a “long” prompt, with at least 1,000 words, without using NLP or NKLT or continuous summation or string improvements of an H2, could even be leveraged if there were such tips at the time I started using GPT-3, but after some time my long prompts would lose meaning, I would start to realize that they would need more guidance, that a dataset would be a mandatory step for improvement, but that no one says anything just because you asked “Say something with 10,000 lines”.
“Say something as best you can with 9,000 lines.”
With prompt [long or short] and response [long or short] that’s my trajectory with gpt-3.
Thank you =D