OpenAI Engineer Leading ChatGPT's Biggest Transformation

According to WIRED, OpenAI's Thibault Sottiaux is leading a complete overhaul of ChatGPT. Sottiaux originally drove OpenAI's fast-growing business in the AI coding space, and he now oversees a large-scale overhaul of ChatGPT as a whole.

This move is seen as a critical turning point in ChatGPT's evolution from a mere chatbot into a more comprehensive AI platform. The details of the redesign are discussed in WIRED's interview article, but what deserves attention is how his AI coding experience is being applied to the redesign of the entire ChatGPT product.

OpenAI Releases Threat Report on Malicious Uses of AI (June 2026 Edition)

On June 11, OpenAI published the latest edition of its threat report on malicious uses of AI. The PDF report, which drew attention on Hacker News, summarizes how AI is being abused and how OpenAI is responding to it.

The regular publication of threat reports is positioned as part of AI companies' efforts to increase transparency around safety. For the specifics, refer to the PDF report itself.

Microsoft President Warns of "Gen Z's AI Backlash"

Microsoft President Brad Smith said that the AI backlash spreading among Gen Z should be taken as "a warning to Big Tech." The remarks, reported by Business Insider, were made during a graduation speech.

Younger generations' skeptical stance toward AI is becoming a factor that can no longer be ignored when thinking about the societal deployment of AI technology. In fact, Stanford's Digital Economy project has released a dashboard visualizing the relationship between AI exposure and changes in employment, signaling growing interest in AI's impact on the labor market.

AI's Power Problem — The Bottleneck Isn't Generation Capacity

According to ITmedia, Gartner Japan pointed out that "delays in power supply are affecting data center construction." However, the real bottleneck is said to lie not in generation capacity itself, but in deeper structural issues.

The surge in power consumption driven by AI demand is a global concern, but in Japan, factors other than power generation—such as grid infrastructure and regional imbalances in electricity supply—have become the binding constraints. Solving this problem will likely require more than simply expanding generation capacity.

Physical AI Paves the Way for Japan's Unique Bargaining Power

Also reported by ITmedia, Laboro.AI held an AI study session for media outlets, covering industry trends for 2026. The company presented its view on how software development is changing amid "the death of SaaS," as well as Japan's path to winning in the global ecosystem through "Physical AI."

Physical AI is a field that fuses the physical world with AI, and it is drawing attention as a domain where Japan holds unique bargaining power—one where "things don't work without Japan." For Japan, with its strengths in manufacturing and robotics, this could be an opportunity to establish a presence on an axis separate from the software-centric AI race.

Frontlines of Local LLMs — Gemma4 26B QAT Hits 110 tok/s

According to a benchmark report on Qiita, Google's Gemma4 26B QAT (Quantization-Aware Training) variant was fully loaded into the 16GB of VRAM on an RTX 5070 Ti, achieving 110 tok/s. Speed and quality comparisons against standard Q5/Q6 quantized versions were also conducted.

The QAT variant is said to reduce memory usage while preserving quality better than standard quantization, which is a big deal for running large models on consumer GPUs. Meanwhile, Hacker News reports of DiffusionGemma running at 775 tok/s on an RTX 6000 Pro show that local inference keeps getting faster.

---