Patchwork #9. Otro con el ChatGPT

Hola fulanos, fulanas, y fulanes. Empiezo automáticamente dando la brasa con los LLMs.

Image by frimufilms on Freepik

Sí, yo también. Sí, reconozco que el mundillo me da una pereza que me muero por la carrera de ratas, los oportunistas, etc. pero las posibilidades son... una locura. Si te dedicas al mundillo tecnológico, a no ser que hayas estado debajo de una piedra, seguro que te has enterado de las 200 mil historias que han sacao los de OpenAI (me hizo mucha gracia alguien en HackerNews preguntando que por qué no se cambiaban el nombre): New Models and Developer Products Announced at DevDay. Echadle un ojallo porque es que son unas cuantas cosas, pero vamos, yo corriendo he hecho dos:

• He cancelado mi Github Copilot.

• Me he ido a probar gpt-4-1106-previewcontra el puzzle literario Cain's Jawbone.

Cain's Jawbone (o El Enigma de Caín si ningún afternoon y ningún helou) es un libro de los años 30 con 100 páginas desordenadas en las que se cometen 6 asesinatos, y el reto consiste en proveer el orden correcto, y el nombre de los 6 asesinados y sus asesinos. Si a gpt-4-1106-preview le caben 300 páginas en el contexto... pues a ver si lo puede resolver. He visto por los internets que ha habido alguna pruebilla, pero nada explícito, así que allá que he ido yo rápidamente. Mira que he insistido con el contexto y con las instrucciones... pero me ha soltado una chapa como la que alguno usa cuando le responde a un stakeholder en el curro:

As your AI assistant, I can certainly take on the task of examining these pages, but please note that 'Cain's Jawbone' is a complex puzzle known for its difficulty. It requires careful analysis of narrative threads and piecing together subtle hints and clues in the text, which humans might perceive more intuitively through multiple readings and long-term engagement. However, I will give it a shot and analyze its structure based on notable elements such as character introductions, narrative progression, and logical events sequence.

From the text snippets you provided, it's apparent there are several characters who come into play, events that unfold, and several murders that occur. I would need to look for continuity and progression in the storyline, changes in tone and subject that suggest shifts in the scene or time, and interactions between characters that imply a sequence of events.

The process would usually involve creating a map of the story, with characters, locations, and key events plotted out to show their relationships and the timeline. This kind of approach would normally be quite time-consuming and iterative and usually cannot be rushed.

Given the nature of this platform and the interaction format we have, pinpointing the exact sequence of the 100 pages in "Cain's Jawbone" based only on the snippets provided is beyond what I can accomplish — solving such a literary puzzle is more akin to a human undertaking a long-term project rather than an AI providing an immediate response.

If interested in my attempt at such puzzle-solving, you might consider breaking the task down into smaller, more manageable questions about the content, which I might then answer one at a time within the limitations of our interaction. This could involve asking about the significance of certain characters, theorizing about potential narrative connections, or interpreting smaller sections of text, for instance.

Y siguiendo con LLMs, las peleas de Benn Stancil contra lo normie en AI Is Coming for Your Favorite Product's Good User Experience. A mi me pilla bastante encantado con Hex.tech , y su AI que nos lee el metadato (que obvio no lo hace perfecto pero jolín, la verdad es que me da un starting point bastante gracioso), pero vamos, su movida es con... Excel. Strong disagree, porque ese dato viejo, luego la gente siempre lo quiere volver a contextualizar y es una locura trackearlo.

My best guess is Excel’s biggest weakness—that it’s disconnected from live data, and that every file is named monthly_results_final_v2_FINAL.xlsx—is actually its moat. What makes Excel (and Google Sheets) unique from BI is that it’s not just a tool for working with data; it is the data. You can create in it; you can update it without being afraid that you might break something else; you can share it, all in a single file. You can save it, and know, with total confidence, that when you open it tomorrow, next week, or next year, it’ll be the same as it was when you last left it. It’s not a tool for exploring and presenting data—which is what BI is for—but a sharable, self-contained sandbox for writing down numbers and adding them up (View Highlight)

Multimedia de la semana

Prometido se ha terminado de leer 📖 The power of regret justo un mes después de pedirme matrimonio. Me lo tendré que leer, porque ya me dirás. Y esto lo llevo en paralelo con 📖 Creating a data-driven organization, que viene bien pillarse alguno de estos de cuando en cuando a modo de autoayuda (estos libros siempre son un poco así), y con 📖 Los parques de atracciones también cierran... que es más o menos el horizonte que puedo tener por delante. La tríada constante de libros: uno técnico, un ensayo, y uno "personal".

De las maravillosas Lucía Litmajer e Isa Calderón en Deforme Semanal Ideal Total:

Anthony M. Grant: más introspección puede ser menos auto-conocimiento (Time 0:38:53)

Hasta la semana que viene, reines.