Cachitos de junio de 2024

• Chollet y Dwarkesh presentando ARC. Creo que esta parte resume bien las diferencias (y similitudes) de Chollet con AGI-optimists:

D: When I talk to my friends who are extremely optimistic about LLMs and expect AGI within the next couple of years they also in some sense agree that scaling is not all you need. That the rest of the progress in undergirded and enabled by scaling, but still you need to add the System-2 and the test-time compute atop these models. Their perspective is that it’s relatively straightforward to do that because you have a library of representations that you built up from pre-training.
C: My intuition is, in fact, this System-2 architecture is the hard part. Scaling up the interpolative memory is the easy part. It’s literally just a big curve. All you need is more data. It’s a representation of a data set. That’s the easy part. The hard part is the architecture of intelligence. Memory and intelligence are separate components. We have the memory. We don’t have the intelligence yet. And I agree with you that having the memory is actually very useful.
D: The alternative hypothesis here is that a former guest Trenton Bricken advanced is that intelligence is just hierarchically associated memory.

• Five Nine Problems, it’s very difficult todo esto.

  The worst that happens when you are wrong about a problem being a 100% problem is that you are slow. You spend a lot of time thinking, a lot of time doing, and a lot of time testing. You burn political capital with your colleagues spending so much time on one thing. Your competitors spend almost no time thinking or testing, and outrace you to market. If you are wrong about a problem not being a 5-9 problem, the worst that happens is that you introduce errors. Often, you lose customers for being wrong. You can lose customer trust or you can get sued. Sometimes you can kill people, but usually you won’t.
  The core trade of “move fast and break things” is that asking for forgiveness is not that bad, a reputation for being untrustworthy fades, and what you are solving usually doesn’t kill people when it goes wrong. It is very much worthwhile to focus on building rather than proving when the stakes are low. (…)
  Thinking about it honestly is important, though: a vertical SaaS startup can likely eschew most reliability concerns beyond 2 nines. A database company or critical infrastructure service likely needs to think about 5 from the beginning. Anyone storing data needs to get 10 or more nines of durability. Understanding your nine requirements can help you figure out where you can skimp and where you need to spend effort.

• Gente que salió escaldada de construir Blue Origin, se pasan a la fusión nuclear intentando aplicar las ventajas con las que está triunfando SpaceX (construir más pequeño, tener una cadena de suministro más eficiente y flexible). “Infinite nine problems” y además ponte a mover átomos, tremenda movida.

• The weird nerd comes with trade-offs.

• Me está sorprendiendo la delicia de leer Poor Charlie’s Almanack. En dos sentidos: lo bien que está escrito y lo buena que es la edición. Lo segundo no lo había notado nunca (ni en otros libros de Stripe Press, y tengo unos cuantos como buena fangirl). Me refiero a que el libro se abre y mantiene abierto incluso en las primeras páginas, pero sin forzar el lomo. La textura de las páginas es agradable al tacto, y las letras sobre la página son definidas, se te van los ojos de palabra en palabra sin esfuerzo.

• También estoy leyendo Understanding Software Dynamics. Un poco al otro lado del espectro, más doloroso pero es como debe ser.

• Terminé de leer How the World Became Rich, básicamente una compilación de los últimos papers explicando esto, en el primer estado que lo consiguió (aka Gran Bretaña) y como el resto han ido/están siguiendo. (Ya hay bastantes disclaimers en el libre sobre que no todo el mundo es rico, y se explican bastante bien, no voy a volver a repetirlos aquí).

• Nuevos discos de Dellafuente y Kaytranada

Y el mes sin acabar.