Playing Tetris can also reveal another possibility for AI computing power
Acurast recently showcased Laya, which runs AI models directly on Android phones. For each move, it chooses the next step from five candidate options based on the position of the blocks.
In the official demo, each decision took about 0.2–1 second.
What makes this example interesting isn’t that AI can play a game, but that it points to a new division of labor: software calculates the options, AI makes the choice, and the phone carries it out.
Applied to Web3, this approach could let wallets compare transaction routes, agents choose their next step from a set of permitted actions, and apps delegate some lightweight decisions to distributed devices.
This doesn’t mean phones can replace GPUs. It shows that not every AI task warrants calling on expensive large models and centralized servers.
That’s the kind of computing demand Acurast is exploring.
#Acurast
Acurast recently showcased Laya, which runs AI models directly on Android phones. For each move, it chooses the next step from five candidate options based on the position of the blocks.
In the official demo, each decision took about 0.2–1 second.
What makes this example interesting isn’t that AI can play a game, but that it points to a new division of labor: software calculates the options, AI makes the choice, and the phone carries it out.
Applied to Web3, this approach could let wallets compare transaction routes, agents choose their next step from a set of permitted actions, and apps delegate some lightweight decisions to distributed devices.
This doesn’t mean phones can replace GPUs. It shows that not every AI task warrants calling on expensive large models and centralized servers.
That’s the kind of computing demand Acurast is exploring.
#Acurast