Francois Chollet

@k_dot_node Trust and capability aren't different axes

AI agents will soon graduate to fully-fledged economic actors that buy services, compute, and even data in the course of accomplishing high-level goals. 1-2 years before we start seeing this at scale.

When thinking about the societal implications of technology-induced abundance, you can use the present as a reference. The present is already a state of extreme abundance compared to 200 years ago and

It takes zero energy to stay certain of your current thesis. Meanwhile curiosity takes a lot of energy and discomfort. It requires constantly disassembling and rebuilding your world model. That's wha

pi**e sounds like some kind of swearword, but it's just ~22.46

The reason congenitally blind people are just as intelligent as we are despite ingesting a lot less "data" throughout their lives is that they navigate the same world as we do (i.e. same complexity re

We underestimate how much "abstract" thought is just repurposed sensorimotor control circuitry. A lot of reasoning is essentially about moving through idea-space the way we move through physical space

I keep reading this take (below) every few months, presented as if extremely profound, and it is just offensively dumb. It confuses data and information, it ignores the fact that not all information i

Unsurprisingly, discretizing continuous signals yields... discrete representations (hence why computers are digital). Even better: you can use these discrete representations to encode continuous repre

At its core, fluid intelligence is a survival strategy in novel, adversarial environments.

In the era of algorithmic distraction, the ability to maintain a single thread of thought for four hours is a superpower. It is the only way to solve hard problems.

It's basically impossible to predict what emergent properties you might get from scaling up a given algorithm. That's why AGI is much more an engineering endeavor than a theoretical one. It's a proces

If you ever feel like you're late to the game, consider that in the 1890s many scientists thought physics as a field was completely solved (quote below is from Albert Michelson in 1894). On the front

@mpvprb Examples being available on the Internet was a big deal in 2023, but today training datasets are vastly larger and richer than Internet data, due to programmatic generation and expert manual d

By explicitly training on specific tasks, we ended up covering a very large area (in absolute terms) of the space of all possible tasks humans can do, but this large area only amounts to 0.00...01% of

For benchmarks that target novel tasks, a common form of benchmark hacking that arbitrages this gap is to generate a dense sampling of potential tasks by manually parameterizing the space and then bru

Even after the steep progress of the past 3 months, it remains that AI performance is tied to task familiarity. In domains that can be densely sampled (via programmatic generation + verification), per

I gained a lot of respect for Dario for being principled on the issues of mass surveillance and autonomous killbots. Principled leaders are rare these days

@liminaltakes I've had colleagues who wrote Python without understanding memory management and let me tell you, it wasn't working out

@0x_Lotion It has held completely true for these layers so far (I count 4 of them in programming), so there's no reason to believe it will suddenly stop being true with the latest layer because that o