this week's japanese tech discourse spans three layers: the legal framing of algorithmic trading, the state's push to fuse research with industrial muscle, and the quiet engineering of running massive models on minimal hardware.
@chuokeizai, algo manipulation
as legal object a new book from chuokeizai-sha, "market manipulation through algorithmic trading," lands today. it analyzes emergency injunctions against manipulation through the lens of algorithmic behavior. the text began as a paper in yokohama law review, now updated with fresh citations. the core question: how should japan regulate these trades while respecting their technical nature. a sign that the legal system is trying to catch up with code-driven markets. source: x.com/chuokeizai/status/2075421523008221523
@takaichi_sanae, the cabinet's innovation strategy
the 85th science and technology council finalized the integrated innovation strategy 2026. the takaichi cabinet's line: win in tech, win in business. they plan a real doubling of research funding in the fy2027 budget, using the "strong and prosperous japan" investment framework to boost grants and university operating subsidies. a revised R&D tax system, strengthened by law, will be pushed on companies and universities for fy2027. the growth strategy also expands the SBIR system for startup procurement and explores new university clusters around 17 strategic fields. the framing is clear: economic strength rests on science and technology. source: x.com/takaichi_sanae/status/2075518524139716900
@rS_alonewolf, running glm-5
on a laptop a 744b parameter model, glm-5, now runs on modest hardware via "colibrì." the trick: keep the 9.9gb shared base in ram, leave the 370gb of expert weights on nvme, and read only what's needed. no gpu required. on 25gb ram, it crawls at 0.05 to 0.1 tok/s. on an m5 max with 128gb, it hits 1.06 tok/s. this is not about speed. it's about making giant mixture-of-experts models run at all without loading everything into memory. the post ends with a quiet hope: maybe soon these models will just zip along on a personal machine. source: x.com/rS_alonewolf/status/2075723437725934006
these three pieces sketch a japan where the state is betting on science as economic engine, the law is belatedly grappling with algorithmic markets, and individual engineers are finding ways to run frontier models on consumer hardware. the gap between policy ambition and technical reality is still wide, but the direction is consistent: more compute, more regulation, more access.
Originally published on FalsifyLab Substack.
— research and educational content. not investment, legal, or tax advice. do your own research. positions and views may change without notice.