Google Brings "Gemini Intelligence" to Android as AI Agents Automate Everyday Tasks

On May 12, Google announced "Gemini Intelligence," a new set of AI capabilities for Android. The feature has AI agents take over multi-step operations that users previously had to perform manually.

Specifically, it includes booking and arranging travel, summarizing web content, auto-filling forms, and converting spoken thoughts into well-organized text messages. For example, just telling it "Arrange a business trip from Tokyo to Osaka" would let the AI search for and book flights and hotels across multiple apps.

Integrating AI agents as a standard Android feature goes a long way toward lowering the barrier to AI for everyday users. AI agents have so far remained confined to experimental tools and developer-oriented platforms, so reaching billions of Android users directly marks a major turning point.

OpenAI Publishes Results from the "Parameter Golf" Competition

OpenAI has published the results of "Parameter Golf," a competition on AI-assisted machine learning research. More than 1,000 participants made over 2,000 submissions, exploring topics such as AI-assisted coding, quantization, and novel model design under strict constraints.

The competition set out to validate the potential of "AI-assisted research" — using AI itself to accelerate AI research. The constraint of achieving the best performance within a limited parameter budget reflects the resource constraints of real-world research. The results appear to show that while AI is useful for making the research process more efficient, human researchers' judgment remains indispensable.

FBI Announces Major AI Overhaul for Criminal Investigations

FBI Director Kash Patel has revealed a plan for large-scale deployment of AI technology in the bureau's criminal investigation work. The main elements reportedly include accelerating data analysis with AI, strengthening pattern recognition, and optimizing investigative resources.

AI adoption is progressing across government agencies, but a full-scale rollout to law enforcement sparks particularly sensitive debate. Privacy concerns, accountability for AI decisions, and the risk of false positives make this a field that demands careful operation. How the FBI's embrace of AI will influence other law enforcement agencies is also drawing close attention.

Using AI in Japanese Costs About 1.5x More, ITmedia Finds in Tests of Major Models

An ITmedia independent investigation suggests that using AI in Japanese may cost roughly 1.5 times more than in English. In real-world measurements of major models such as GPT-5.5 and Claude Opus 4.7, Japanese was confirmed to have lower token efficiency than English.

As AI pricing shifts toward usage-based billing, this "cross-language cost gap" is a problem that Japanese users and businesses cannot afford to ignore. Even for the same content, Japanese consumes more tokens, so the cost difference accumulates with sustained AI use. The findings underscore that model selection should take token efficiency into account, not just raw performance.

Paper on AI Agents Under EU Law Published on arxiv

A paper titled "AI Agents Under EU Law" has been published on arxiv. It analyzes how autonomously acting AI agents are positioned under the current legal system following the enforcement of the EU AI Act.

Unlike conventional chatbots, AI agents are capable of making their own decisions and acting on them. As a result, issues arise that existing legal systems never anticipated, from liability attribution and data protection to consumer rights. The paper considers how to fit AI agents into the EU AI Act's classification framework. Japan has also begun debating the legal regulation of AI agents, making the EU's developments an important reference case.

Debate Heats Up Over AI Making Things "Worse Instead of Better"

A question on Hacker News is drawing attention: "If AI has made engineers more efficient, why does everything feel worse?" It drew numerous firsthand accounts from people on the ground about frequently occurring service outages, UIs that never improve, and security issues that never end.

Behind this debate is the possibility that while AI code generation delivers short-term productivity gains, it also produces side effects such as the accumulation of technical debt and struggling quality control. AI tools have not been widespread for long, and verifying their long-term impact is still at an early stage.

---

Exclusive Investigation: Does Using AI in Japanese Cost "About 1.5x More"? - ITmedia AI+

[AI Agents Under EU Law](https://arxiv.org/abs/2604.04604) - arxiv

[Ask HN: If AI has made engineers more efficient, why does everything feel worse?](https://news.ycombinator.com/item?id=48113873) - Hacker News