AI Advice Slashes Accuracy to a Third, Jensen Huang's Japan Visit Spreads Deals, AI Progress Demonstrated on 4 Metrics — AI News Digest

Study: AI Advice Cut Human Accuracy to a Third While Doubling Confidence

In an experiment reported by The Next Web, respondents who received AI advice saw their accuracy drop to roughly a third of those who answered on their own, while their confidence in their answers doubled. The study drew notable attention on Hacker News, picking up 93 points and 35 comments, and its finding that "AI advice suppresses critical thinking and increases wrong answers" resonates as a practical warning at a time when agent usage is going mainstream. It once again throws into sharp relief the kind of cognitive bias that comes from trusting AI as a black box.

Jensen Huang's Japan Visit Leaves Deals Spread Across Japan's Entire Tech Ecosystem

As Kate Park reported for TechCrunch, even after NVIDIA CEO Jensen Huang left Tokyo, deals spanning Japan's entire tech ecosystem remained in place. The specifics of the partners and the scale are covered in the article itself, but the agreements appear to be comprehensive arrangements bundling together semiconductor procurement, research institutions, and industrial uses — and are likely to influence Japan's domestic AI infrastructure policy as well. The focus now shifts to the coming weeks, as observers decipher the visit's "afterglow" effects.

WSJ Reports on the Companies Pouring Money Into Top AI Models

The Wall Street Journal's CIO Journal section rounded up the realities of companies investing heavily in top-tier AI models. The article, which gained traction via Hacker News, reflects a reality in which model selection is no longer purely a technical decision but one of budget and risk. With licensing and API costs for frontier models climbing year after year, CIOs are being forced to rethink how they measure cost-effectiveness and how to steer clear of vendor lock-in.

A Skeptical CTO Verdict: "Is AI Progress Real?" Tested Across 4 Metrics

An analysis published by SkepticCTO on Substack examined whether AI progress is real using four independent metrics, concluding that all of them showed a consistent upward trend. At a moment when benchmark inflation and data contamination are being called out, the approach of combining multiple metrics from different angles carries real persuasive power. Even as an analysis from a skeptical standpoint, it managed to confirm the "direction" of progress.

Forbes Cheers the "Birth of New Professions"; a Historian Sees "a Familiar Pattern"

While Forbes reported that AI has given birth to new professions, Future Hangover (Substack) pushed back, arguing that strikingly similar patterns of new occupations emerging have been observed repeatedly throughout history. Society's responses to technological shocks follow a recurring structure, the argument goes, and this time too, caution is warranted before declaring anything "entirely new." It makes for a useful lens for viewing media hype with a cooler eye.

When Senior Lawyers "Sharpen Their Hunting Spears" with AI

A Financial Times article made the rounds on Hacker News. Titled "AI should have senior lawyers sharpening their hunting spears," the piece appears to argue that rather than being displaced by the spread of AI, senior lawyers may instead find themselves concentrating more on strategic judgment and litigation design. The push by those at the very top of the professions to harness AI as a weapon looks set to accelerate.

Claude Code v2.1.215 Stops Auto-Running /verify and /code-review

As moha0918_ explains on Qiita, Claude Code v2.1.215 no longer auto-runs `/verify` and `/code-review` from the agent side. They now work only when you explicitly invoke the commands, reining in the behavior of silently burning tokens in the background. It's a change in the direction of handing control of the developer workflow back to developers.