Yonatan Sompolinsky@hashdag
A new paper worth being excited about: proofs of useful work (pouw) by Ilan Komargodski @komargodski, Itamar Schen, and Omri Weinstein @WeinsteinOmri.
arxiv.org/pdf/2504.09971
tldr; a pow function involving matrix multiplication such that the miner can choose the matrices freely (eg matrices used for own AI inference)---counterintuitive that main pow properties still hold, but they do. This is the first useful pow that makes sense since the idea was proposed about a decade ago.
Why definitely important: solves / significantly mitigates security budget draining threat, by adding a funding source (the usefulness) for mining.
Why could be important: mitigates wastefulness of pow by plugging into a high demand computation market, namely AI inference (/training). I believe addressing this issue is vital for political feasibility of pow, it's too easy for politicians to rally public opinion against wasteful pow; admittedly feels like the pow bon ton gets a kick out of this misalignment.
Possible criticism: some argue wastefulness is essential for security, and that usefulness effectively funds attacks or removes attack costs (related: arxiv.org/abs/1605.09193). One remedy is the constant wasteful overhead that still exists in pouw, but mainly I believe this is not a real threat if you successfully penetrate a huge computation market with so many diverse players; the vast scale of this market can arguably cover for model imperfections (which are inherent to the objective).
Notes: (i) afaiu from the team, current pouw requires adaptation to be fit for high bps; (ii) pouw unrelated to opow/kheavyhash though shares some common goals.
Congrats on this achievement pouw team, tracking your progress!
x.com/Lhree/status/1…
