OpenAI: "Chat Is Dead" — But the Super App Vision Isn't Over Yet
According to TechCrunch, a senior OpenAI employee has said "Chat is dead." The company is reportedly still moving forward with its "super app" vision, aiming to build a platform that goes beyond a mere chatbot.
It remains unclear what form OpenAI's super app will actually take, but it's noteworthy that there are internal efforts to rethink the chat interface itself. It hints at a possible shift in AI interactions, from the text chats we've known toward more immersive, multi-modal experiences.
DeepSeek Takes the Top Spot in U.S. Enterprise AI Adoption
According to data from Ramp, a corporate spend management platform, DeepSeek has surged to the top of the trending software vendors for June 2026. Cases of U.S. companies sending data directly to DeepSeek in search of cheaper AI models are reportedly skyrocketing.
Ramp's chief economist Ara Kharazian points to heightened cost consciousness as the driver, while also warning about the security risks of using Chinese-made models. How to weigh data sovereignty against lower cost is becoming a critical decision axis for enterprises.
Qwen3.6 35B-A3B Reaches Practical Usability on a Laptop
A firsthand report of running Qwen3.6 35B-A3B on a laptop with an RTX 4060 (8GB VRAM) and 64GB RAM has been drawing attention in Reddit's LocalLLaMA community.
According to the report, it achieves practical generation speeds of roughly 27 TPS at 32K context and about 18 TPS even at 256K context. File reads and writes, CLI command execution, and Git operations all stay local, which the author describes as a groundbreaking experience in that "no private information is sent to cloud models."
The fact that local LLMs have finally reached a "practical" level even on laptops marks a major turning point for privacy-conscious users.
The Limits of OpenAI's Lockdown Mode — Prompt Injection Can't Be Fully Prevented
On June 6, OpenAI made ChatGPT's "Lockdown Mode" generally available. It's a feature designed to protect sensitive data from prompt injection attacks, but TechCrunch reports that "even with Lockdown Mode, ChatGPT may still be vulnerable to prompt injection."
Japan's tech community has also taken up this issue, with many pointing out the weight of the fact that "the AI company with the most abundant resources in the world has officially admitted that this attack cannot be fully prevented." A growing discussion argues that defending against prompt injection requires a fundamental architectural rethink rather than simply stacking filters.
New Hardware for Local LLMs: GMKtec EVO-X3
GMKtec has announced the EVO-X3, a mini PC equipped with OCuLink, Wi-Fi 7, and dual PCIe 4.0. The headline: a 192GB Ryzen AI MAX+ 495 model is slated to arrive later.
The Ryzen AI MAX+ 495 is built on the Strix architecture and may offer enough memory bandwidth to run local LLMs. Pricing has not yet been announced, but it is attracting strong interest in the local LLM community.
---