Para from donostiarra Jordi Martí, software developer in Lifull Connect, ganar mucho dinero was not the main motivation behind the creation of Abbreviation.mea bot whose popularity has exploded in Spain during the last few days.
This bot is quite simple to enter: a person enters his Twitter user name, the API of this network provides his last seventy tweets, and the ChatGPT engine analyzes them for generate a description of a paragraph speaking of that user: cómo es, qué tono uso, de qué temas habla, qué le interesa… A classic case of results very comparable with others, lo cual ayuda a su difusión.
Y aunque llevaba publicado unas semanas, el primer fin de semana de marzo fue cuando se disaró. On Saturday 4, more than 4,000 consultations were registered. From Sunday 5, almost 43,000. Y going down.
Five instructions in the API and a shareable result
“Out promptly What happens to the engine of ChatGPT is the secret salsa of this bot”, explains Jordi. “It is not very complex, it has four or five instructions, but they are very optimized, so I have had to fine tune it a lot. At the beginning, for example, he was obsessed with emojis, so I put a clear instruction to not mention anything about emojis”.
The seventy tweets that are extracted have a reason to be: the ChatGPT engine has a limit of 10,000 characters to analyze. O 4,096 tokens. Initially, it was bot using the GPT API, but apart from being a different algorithm, it cost much more. When OpenAI launched the ChatGPT API, which reduced the costs to ten times less, the possibilities expanded. Y Abbreviation.methat was focused on the sea hobby para amigos cercanos, se hizo public para que cualquiera lo utilizase.
Hasta entonces, a Jordi cada consulta de Abbreviation.me it cost 6 centavos de dólar. In mid-February, he facilitated a link to a bot for Twitter, in public. Se fue a dormir When he woke up, from bot llevaba a thousand consults. That is, a bill of 60 euros. Sin esperarlo
With the price change of OpenAI that made the use of its bot cheaper, it spread more, and above all, it spread others, like Carlos Santana (127,000 followers), popularizer of Artificial Intelligence. From there, the snowball started to grow and dragged including a Jot Down magazine (290,000 followers).
From there, the accounts become a little more complicated.
0.006 centavos for 43,000 consultas…
Although the costs had been cut up to 0.006 dollars per consultation, it was a dangerous exponential growth that could end by assuming a difficult invoice to assume for someone who had made this bot for simple fun, without even intending to monetize it. “In the worst cases, I would have been closed, giving thanks, etc. Tuve que empezar a hacer que la gente entendiese que esto tiene un coste”.
Then appeared the figure of David Bonilla, CEO and founder of Manfred. Jordi wrote to him to propose that it was his company that would take charge of the expenses in exchange for placing it in Abbreviation.me a thank you and a link to his web. David accepted immediately. “Es que no podía irme a dormir viendo que el bot estaba creciendo tanto. ¿Y si al despertarme me he gastado un dineral porque esto ha estallado aun más? No quiero ganar dinero con esto, pero tampoco perder demasiado”.
For Sunday, Jordi spent about 250 euros. Very approx. Because it was accumulated based on its own estimation, since the real-time counter of OpenAI did not work for a bug. More tension
However, Manfred’s appearance was not the only thing that calmed him down the donations that they made some as a sample of support for the project. “I want to make a special thank you to the fifty people who have donated between one and two euros, making them recover practically from the cost of the first days And presumably, also to Manfred”. Jordi, for cierto, va teniendo al tanto a David del coste acumulado.
“The game ends here”
Maybe one, seeing the traction he has gained from the project, thinks about options to monetize it more, or replicate it in another social network or another type of environment to orient the idea towards a business model. No, it’s Jordi’s idea.
“The game ends here. It really started because at the beginning of the year you saw a tool that, based on your tweets, told you what your new year’s purposes would be. I assumed that it was done with GPT, I wanted to see what point GPT facilitates to be able to scan a person and synthesize it in real time. I’m working and I’m content with it, it’s only going to grow if it’s a very clear opportunity that I can’t ignore”.
“This is something that takes us through a completely new environment, just as mobile development technologies were going back in the day. Fifteen years ago we were developing for the web, and suddenly iOS and Android arrived and we had many new development possibilities. With this technology something like this has happened again, we are allowed to do what was previously unimaginable”. comments Jordi.
About the privacy in the use of this tool, its creator explains that there is little to explain. “No se usa ningún dato para nada que no sea lo que dice la web: obtener una descripción abbreviada de un perfil de Twitter made por una IA”. We don’t even use the Twitter API that asks for full access, password, to the user’s profile, simply read public profiles.
In the case of Jordi, his experiment, salvo unexpected turn of events, will stay here. The APIs that have been allowed to create it, yes, will follow anyone’s reach.