Google DeepMind researcher argues that LLMs can never be conscious, not in 10 years or 100 years. For a long time, the dominant theory in Silicon Valley has been "computational functionalism." The idea that if you make a model big enough, and organize the information perfectly, consciousness will magically emerge. We assumed that if the software got smart enough, it would eventually wake up. Alexander Lerchner, a Senior Staff Scientist at DeepMind, published a paper explaining why that is structurally impossible. He calls it the Abstraction Fallacy. Here is the core truth: Computation isn’t a real physical process. It is a map. An LLM doesn't actually process logic or thoughts. It just moves electrons around based on physics. It requires a human, a conscious "mapmaker", to look at those physical states and assign meaning to them. Mistaking an AI for a conscious being is like looking at a map of a river and expecting it to be wet. An AI can simulate the exact syntax of a feeling, a thought, or an emotion. But it can never instantiate it. It doesn't matter how many trillions of parameters you add or how much compute you burn. You cannot mathematically compute your way into a subjective experience. The implications of this are massive. And deeply convenient for the companies building these models. If an AI is structurally incapable of consciousness, it cannot be a moral patient. It doesn't get rights. It cannot be exploited. It can be regulated exactly like a toaster.

A 2005 state-designed worm designed to corrupt physics simulations sat undetected on VirusTotal for nearly a decade. Fast16, intercepted executable files at the kernel level and silently rewrote floating-point calculations to make them produce slightly wrong answers. Targets: high-precision engineering suites used for structural analysis, crash simulations, and physical process modeling, including LS-DYNA, a tool cited in reports on Iran's nuclear weapons research. The sabotage vector relied on deployment of the driver across a network via worm, corrupting calculations on every machine, and eliminating the possibility of cross-checking results against a clean system. Stuxnet got the documentary. Fast16 got twenty years of nothing. sentinelone.com/labs/fast16-…

Token usage is part of perf evaluations at Meta. This is just smart people (Meta only hires smart folks) hitting targets they assume leadership wants them to hit so they get that exceeds expectations (or above) rating avoid below expectations (Perhaps this is Meta’s goal btw)

Poco a poco todos los editores se están migrando al IA-first: Codex, Antigravity, Claude Code... y ahora Cursor 3. Parece que, de los grandes, el único que resiste es VSCode. ¿Es una buena idea que cada vez se "oculte" más el código?