Thinking about artificial intelligence, consciousness, and what technology does to us. Published as it comes.
Consciousness may well be a fuzzy word. Qualia is not. An argument that the felt quality of experience resists both deflationary and materialist accounts.
Read
AI assistants adapt to everyone and speak from nowhere. An argument that plurality is not a matter of taste but a condition of thinking freely.
Read
Reading a model’s thoughts has become possible. But we can only recognise, inside a network, the concepts we already had a name for.
Read
AGI is not a threshold anyone measures. It is a word the labs are teaching the public, so they can decide for themselves when it has been crossed.
Read
Letting a model think longer is the missing ingredient. But a generator trained against a judge learned from human annotation inherits that judge’s ceiling.
Read
Faces and voices are being used to sell things their owners have never heard of. I watched this begin at Samsung in 2019, and nobody in the room asked.
Read
Every Stack Overflow metric is going down, and pretty much everyone draws the same conclusion from it. I think it is wrong, and that the subject is elsewhere.
Read
A model with frozen weights is procedural memory without plasticity. What gets replaced first is whatever had already been proceduralised.
Read
ChatGPT is not an innovation, and it is still a revolution. What changed sits in the third stage of its training, and in the tiny world that stage hands it.
Read
Creativity is not a human privilege. A machine found in Go what centuries of players had not: that capacity, not the artwork, is the thing to look at.
Read
Two years writing human-robot interaction software. What I learned: the hardware is beautiful, the software is not, and creating is easier than developing.
Read