This is a special post for quick takes by simulus. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
The allocation of resources away from general computing hardware and towards specialized chips (Talaas, Cerebras, Etched, maybe Google TPUs) probably makes a fast takeoff less likely.
Optimized inference hardware probably would speed up AI-assisted R&D i.e. trying to elicit lots of interesting strategies out of the models, or help make prosaic continual-learning strategies more viable[1]. Also Cerebras' wafer chips seem performant for any compute including training (even if not as FLOPS-dense as datacenter Blackwell) and Anthropic has been expanding TPU capacity for training and inference
Talaas' and Etched's products do seem less for general compute though
I’m not sure. All these specialized chips have their own advantages which could be helpful in fast takeoff. For example, Cerebras has a bunch of densely packed SRAM on the chip and fast clock speeds, which might be useful if you need to run a model/algorithm that requires a lot of serial steps of computation at low arithmetic intensity. Also, a sufficiently advanced AI could design its own ASICs for various tasks in parallel if it needed to, perhaps within a few weeks/months, so the initial hardware might not matter that much.
3simulus
This is a good point. More options could increase p(there exists a chip that does what the AI wants).
[...]
Depending on who you talk to, weeks/months might not be a "fast" takeoff.
But this brings up another point: the diversification of chip manufacturing chains could make it faster/easier for an advanced AI to produce custom chips of its own design, which would increase takeoff speed.