3 Comments
User's avatar
Homo Viator's avatar

The ancients knew point 7 without a hippocampus: they made Mnemosyne, memory, the mother of all the Muses. Imagination as memory's child — it took neuroscience two and a half thousand years to confirm the genealogy.

Cyril Simonnet's avatar

The fixation on biological and nuclear risks is telling. Hassabis frames danger as something the model might *do* once it gets smart enough. That skips the threat that's already here: the infrastructure around the model is full of holes, and attackers don't need AGI to exploit them. Every org racing to plug AI into production is handing out credentials, API keys, and identity tokens faster than their security teams can track. I wrote about exactly this blind spot. The real attack surface isn't the model's capability level, it's the identity layer we wrap around it. Hassabis is solving for a future problem while the present one is wide open.

https://cyrilsimonnet.substack.com/p/the-model-was-never-the-attack-surface

The Catalyst Shift's avatar

Saying AGI is 2-5 years away in the fast developing AI ecosystem is conceding that they are nowhere close to it.