@maximohistorico@GuillermoVersus Ya se las estamos dando, de primera mano y en mayor cantidad, cada que usamos un chatgpt, claude code, codex, cursor. Pero ¿Cuál es la opción?¿Es posible dejar de usarlos simplemente?
1/4. Hoy ha llegado a mis manos un libro jurídico publicado en una "editorial de prestigio" española en el que se citan reiteradamente varias publicaciones que realmente no existen.
One of the reasons I'm dubious about the 'nontechie builds a product with AI' posts is the sheer impracticality. Even if we put aside the fact that our nontechies will have to test, deploy, create a pipeline, handle cloud setup, and do other technical things, you can't specify a non-toy product in a single prompt. Even if you could, LLM-generated code is both overcomplex and larger than code written by a skilled developer. You'll have a mountain of output—more than you'd have if you'd hand-coded the app. Think about how big your current product is. Multiply that by at least two. In the worst case, put it all into one file. Our hapless nontechies who can't make hide nor hair of the code will have to load that entire humongous blob of goo into a single context to make an incremental change. The token cost will soon bankrupt them (and the LLM will be so slow you may as well write the code by hand 😄).
The solution is Architecture. A component architecture, for example, gives you the ability to work on one component at a time in a much smaller context where changes are cheaper. That moves us firmly out of the realm of the nontechie, however. Software Architecture is a hard-core technical skill, one that many programmers don't have. Even Spotify, which has been crowing about devs 'not writing a line of code' (which isn't strictly true—they are still reviewing and refactoring it), does not trust the LLM with the system architecture. That takes skilled architects.
The need for appropriate architecture is particularly thorny with legacy apps, most of which implement the blob-of-goo pattern. (Bad patterns are still patterns.) That structure will kill you when you try to integrate an LLM into your process. You can't just throw an LLM at a ball of mud. Changes become all-or-nothing across the entire app, side effects abound, and token costs skyrocket. You need to restructure the app to be LLM-friendly. That restructuring requires architecture.
What, you fired everybody who understood software architecture? Good luck with that.
@powerhdeleon@FaztTech ¿Y para cuando se podrá automatizar la publicación de contenido confiable y no tendencioso? ¿Eso es más difícil o también lo tendremos en 12 meses?
@_pegajoso@5eniorDeveloper Supongo que es porque no estás familiarizado, me pasa al revés, creo que con Spring se puede resolver cualquier cosa que necesites, puedes hasta integrar IA sin salir del ecosistema y de forma muy simplificada.
@Telmex Hola ¿Hay alguna forma de revisar un problema con mi servicio que no sea una llamada con un bot que me diga que me tengo que esperar 3 días para ver si se resuelve?
📢¡Atención aspirante!
🗓️Del 6 al 13 de junio de 2025 podrás realizar los pasos 6 y 7 de la Convocatoria de Admisión BUAP 2025.
🔗Ingresa a https://t.co/b4namBHlu2
Resuelve tus dudas en: [email protected] y WA: 8115431742.