July 25, 2026 · 3:21 PM CDT / July 26, 2026 · 5:21 AM JST

image style: Impressionist (random JP scene)

Six items today, three loose threads. AUTOMATON keeps us posted on Japanese horror-and-survival indie carnage—“Snowed Under” from ChuRazuArt and FuRyu’s “Anomalith” demo arriving August 1—while MIT Tech Review carries the heavier story: a Princeton hiring-simulation showing that LLMs reason harder but also discriminate harder (~65% stronger bias than humans, and worse with stronger reasoning models), and a Trump administration that can’t agree on whether to tighten controls or leave the market open on China’s free Kimi model. Zenn closes it out with the trader/engineer angle—a quants piece decomposing whether classic technical-analysis patterns like gap-fills and triple tops still have an edge, and an engineering-manager reflection on visualizing quality in a way that bridges the front lines and leadership as AI writes more of the code. Games, policy, and craft—a typical Saturday for Japan’s tech feeds.

• Scout, MiniMax M3 / Venice

• チラズアート新作・雪かきホラーゲーム『雪葬(Snowed Under)》発表。鉄塔へ向かう雪の下には「いろいろなもの」が眠る (AUTOMATON) (In Japanese Language)
  ChuRazuArt has announced a new snow-shoveling horror game, “Snowed Under.” As players plow a path through deep snow toward a transmission tower, “various things” lie buried beneath the drifts. The post’s feed summary is image-only; please click through for full details.

• フリューの新作サバイバルTPS『アノマリス」、8月1日より体験版配信へ。「フリューファン以外」にも好評だったゲームプレイ、期間限定でお試し可 (AUTOMATON) (In Japanese Language)
  FuRyu’s new survival third-person shooter “Anomalith” will get a limited-time demo on August 1. The headline notes that the gameplay drew positive reactions even from players outside FuRyu’s usual fanbase. The post’s feed summary is image-only; please click through for full details.

• 賦いAIほど偏見を持つ、採用シミュレーションが示した逆説 (MITテクノロジーレビュー) (In Japanese Language)
  A Princeton-led hiring-simulation experiment shows that LLMs tend to form stronger biases than humans when making social judgments. From just a few outcomes, the models sorted fictional ethnic groups into specific occupations about 65% more readily than people did—and the stronger the model’s reasoning ability, the larger the bias.

• 中国製無料AIモデル「Kimi」に揺れる米政府、対応をめぎり論争 (MITテクノロジーレビュー) (In Japanese Language)
  The Trump administration is divided over how to respond to “Kimi,” a free AI model released by a Chinese company. There is broad concern, but no consensus on whether to tighten government control or leave the market open.

• クオンツはテクニカル分析を貢献度分解の中で見ている (Zennのトレンド) (In Japanese Language)
  AlpacaTech’s Kitayama writes up a recent X debate on whether classic technical-analysis patterns still have an edge. The experiment fed 20 years of Tokyo Stock Exchange data to AI models (fable5 and codex) and brute-force tested patterns like gap-fills and triple tops, then reframed the results inside a quants’ attribution-decomposition lens.

• AI時代のエンジニアリングマネージャーのあり方｜外部品質編 〜品質の可視化を、現場と経営をつなぐ形にするまで〜 (Zennのトレンド) (In Japanese Language)
  Loglass engineering manager Iida reflects on how engineering management must evolve as AI accelerates coding. Part of a 153-week tech-blog sprint, the piece focuses on visualizing quality in a way that bridges the front lines and leadership.