ReadingResearchRadarInvestment framework
Sign in / Sign up中文
Sign in / Sign up中文
ReadingResearchRadarInvestment framework

Reading

2026-09-274 posts

Hong Kong delays AI copyright amendment, will issue guidelines first

Material
2026-09-27 08:00 UTC

Hong Kong has put off its AI copyright amendment and will issue operational guidelines first.

According to the South China Morning Post, the Copyright (Amendment) Bill 2025, which would have introduced an "opt-out" mechanism for text and data mining, was removed from the Legislative Council's agenda last August and does not appear in the 2026 blueprint or the city's first five-year plan, both delivered earlier this month.

The opt-out mechanism drew opposition from copyright holders after it was proposed in 2024, who said checking whether AI companies had used their works would take too much effort and counter-proposed an opt-in model. The shift to guidelines first means the legal boundary for AI training data in Hong Kong remains unsettled for the near term.

Sources:https://www.scmp.com/news/hong-kong/society/article/3368913/hong-kong-delays-ai-copyright-law-focus-guidelines-first

Research

UK's largest board game expo bans AI-generated games, with narrow assistive exceptions

Quick take
2026-09-27 05:18 UTC

UK Games Expo, the largest tabletop gaming expo in the UK, announced new rules this week banning AI-generated games from the show floor, with only a narrow set of exceptions.

Per the policy text as relayed by IT之家, exhibitors displaying or selling AI-generated works can be ordered to remove them; refusal means the organizer can shut down the booth and bar the exhibitor from future shows. The exception covers only "minor editing, assistive production, or accessibility work."

According to Rascal, UKGE follows Gen Con and other major expos in clearing AI-generated works from their floors. The expo drew criticism in 2025 for a sponsored social media post promoting the AI-generated board game Gates of Krystalia. The full rule text is currently known only through media relay; the original Rascal report is paywalled, so enforcement details should be checked against the organizer's official text.

Sources:https://www.ithome.com/1/007/506.htm

Research

Insurers say hospital AI coding added $942M in two years

Quick take
2026-09-26 21:02 UTC

An analysis by the Blue Cross Blue Shield Association says hospitals' use of AI tools when submitting insurance claims added $942 million in healthcare spending over a two-year period.

The association found a sharp increase in patients documented as having complex conditions, with no corresponding change in care delivered, arguing that medical coding and actual treatment have disconnected.

The association is the paying side, in direct conflict with hospitals over payments; the New York Times notes that both hospitals and insurers now use AI, and the friction is getting worse.

Sources:https://techcrunch.com/2026/09/26/insurers-claim-ai-is-already-increasing-healthcare-costs

Research

Nvidia says tuning the harness, not the model, can nearly halve coding-agent token use

Quick take
2026-09-26 10:30 UTC

Nvidia researchers report in a preprint that SoL-Pi, a system that uses a research agent to rewrite a coding agent's harness, cuts token use by 49 percent in its leanest variant while retaining 93.7 percent of the Pi harness's score.

The system explored 152 directions across 535 executable environments and more than 3,000 runs, producing four mechanisms including action fusion and context compaction. At current API prices, the authors estimate savings of $8.75 to $13.50 per hour versus native Codex and Claude Code harnesses.

The gains do not hold everywhere: on 63 Terminal-Bench 4 tasks, SoL-Pi solved 15 while Codex and Pi each solved 18. The authors themselves note that automatically optimized harnesses tend to overfit their training tasks.

Sources:https://the-decoder.com/nvidias-sol-pi-system-cuts-coding-agent-token-usage-nearly-in-half-by-optimizing-the-harness

Research

You’re all caught up in this view