Share

    • [Tokyo Tech Translated] superhuman employees, superhuman managers

    By FalsifyLab June 23, 2026
    this week's japanese tech conversation circled one question: what happens when you drop a superhuman tool into a very human system. the answers range from "it works" to "it doesn't" to "it works if you also upgrade the manager." ## @PreferredNetJP, plamo 3.0 release preferred ne

    this week’s japanese tech conversation circled one question: what happens when you drop a superhuman tool into a very human system. the answers range from “it works” to “it doesn’t” to “it works if you also upgrade the manager.”

    @PreferredNetJP, plamo 3.0 release

    preferred networks released plamo 3.0 prime, a domestically developed generative ai model. api or on-prem deployment. two variants: a reasoning model for complex tasks and a faster non-reasoning model. context window expanded from 64k to 256k. improved safety against harmful information. positioned for ai agents and enterprise use. claims high japanese language performance and cost efficiency.

    source: https://x.com/PreferredNetJP/status/2068886593159745629

    @tadeku_net, fiction writing tools

    a novelist running tadeku-tools asked for feedback on their writing utilities. they’re actively fixing bugs and building features based on user reports. small feedback welcome. the tools are live and free to try.

    source: https://x.com/tadeku_net/status/2068888870582919433

    @fladdict, superhuman employees

    need superhuman managers ai agents are just superhuman employees. no matter how capable the employee, a broken company structure, payment system, workflow, or manager will still break the outcome. sakana fugu’s design, pairing a superhuman manager with a superhuman employee, probably outperforms just dropping a superhuman employee into a normal company. the structural argument makes autonomous agent swarms more interesting than single-agent copilots.

    source: https://x.com/fladdict/status/2068910624567181401

    @kiokunir, earnings summaries as alpha

    one trader reports turning 4 million yen into 75 million yen in unrealized gains since 2026. the unlock: summarizing earnings reports to extract what companies are actually saying. once they started doing that, they knew what to buy. rakumachi noticed the returns and reached out.

    source: https://x.com/kiokunir/status/2069004742672761079

    @ImAI_Eruel, deepmind’s shift toward llms

    deepmind’s hassabis wasn’t that interested in the mainstream generative ai race. it was draining budget from his real focus, ai for science. but fable/mythos class models are starting to look genuinely useful for science. gemini 4.0 onward likely gets serious investment.

    source: https://x.com/ImAI_Eruel/status/2069183800698151153

    the pattern this week: japan’s ai builders are shipping foundation models and writing tools. the discourse around them is pragmatic. not “ai will replace everything” but “ai is a superhuman employee, and that means you need a superhuman manager, a clean workflow, and a way to extract signal from noise.” the trader who made 75 million yen from earnings summaries didn’t just use ai. they built a process around it. that’s the difference.

    more at falsifylab.com

    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.

    #quant #trading #crypto
    Write a comment