Sabitlenmiş Tweet
One thing we're trying to avoid with the design of Ambient is what I call "The ASIC Trap." Our proof of work is designed to operate at an algorithmic level, where the algorithm is generic across implementations and hardware, and the "usefulness" of the work is enforced by the choice of the network LLM and the supply and demand matching that occurs on chain.
Imagine if, instead, we made the proof of work something more fundamental, like matrix multiplication. Matrix multiplication is useful, right? But, it turns out that if your work is something fundamental like that, it's much cheaper to do a lot of nonsense matrix multiplications on a piece of hardware like an ASIC than it is to do meaningful operations on something like a GPU running an LLM. And at that point, you've stopped doing useful work and just become a clone of Bitcoin, because all your GPU providers are getting priced out. And Bitcoin ASIC fab capacity is decreasing year-on-year because ASICs are not inherently useful.
Thus, "The ASIC Trap": Any primitive mathematical operation that is claimed as inherently a proof of useful work reduces to a least-common-denominator of non-useful work on a highly specialized piece of hardware. An unfortunate economic inevitability
English
