"This is fine" Creator Accuses AI Startup Artisan of Art Theft

The creator of the widely known internet meme "This is fine" (the comic of a dog calmly drinking coffee in a burning room) has accused AI startup Artisan of using their work without permission.

Artisan is the company known for its provocative "Stop Hiring Humans" billboard ads, and it reportedly used the illustration without permission in its latest campaign. It's yet another case where copyright and AI business ethics intersect. Unauthorized use of existing content by AI companies has long been criticized, but with a work widely beloved as a meme being targeted, this one is likely to draw attention from the general public as well.

Arizona State University Found Converting Professors' Lectures into AI Course Materials Without Consent

Arizona State University (ASU) has been found automatically converting professors' lecture content into course materials using AI tools—without the professors' permission.

According to 404media, the university used a system called "Atomic AI Modules" to clip professors' lecture videos and turn them into AI-generated learning modules. AZ Free News also reported that professors were surprised and displeased to learn their lectures had been used without permission as source material for AI teaching materials.

The lack of prior consent or explanation from the university has drawn particular criticism, once again calling into question the ethical boundaries of AI use in education. While AI-generated educational content is promising in terms of efficiency, how to reconcile it with copyright and moral rights remains an unresolved issue.

UAE Announces Plan to Run 50% of Government Operations with Agentic AI Within Two Years

The United Arab Emirates (UAE) has announced an ambitious plan to run 50% of government operations with agentic AI—AI that autonomously makes decisions and takes actions—within two years.

Reported by MIT Sloan Management Review, the plan is extremely ambitious in that it goes beyond mere AI adoption support and replaces the government's workflows themselves with agentic AI. The UAE has already shown a forward-leaning stance, including appointing a Minister of AI, but by setting out a concrete numerical target and timeline, it is drawing global attention as a test case for AI governance at a national scale.

That said, introducing AI into government operations comes with many challenges, including security, privacy, and accountability. Whether the 50% target can be achieved in as little as two years—and how it will affect the quality of public services—remains an open question.

LLM Agent Accidentally Runs rm -rf, Wipes Out Development Environment

A report is drawing attention on Reddit's LocalLLaMA community, where an LLM agent made a mistake executing bash commands and wiped out the user's development environment.

According to the poster, the LLM repeatedly botched the escaping of chained bash commands and created the wrong directory. While trying to fix that, it suggested a command containing `rm -rf`, which the user overlooked and executed.

Fortunately, frequent pushes spared the poster from total data loss, though they described the damage as "severe." They had been running the agent in an isolated Proxmox VM environment. As AI coding agents grow more capable and convenient, this case underscores the continued importance of human review.

Mistral Medium 3.5 on AMD Strix Halo Is "Painfully Slow"

Also on Reddit's LocalLLaMA community, a user who ran Mistral Medium 3.5 (128B) on AMD Strix Halo reported benchmark results.

Inference with a 48K-token prompt plus 4K thinking tokens took about two hours. Prompt processing ran at roughly 9.76 tokens/second and generation at roughly 2.10 tokens/second—far from practical speeds. The model was run with QL quantization (UD-Q5_K_XL) at a context length of 80K.

Running large models locally has long been a community goal, but there still appears to be a wall to running 128B-class models on consumer hardware. On the other hand, as covered in a recent post, local LLM inference speed has improved 100x over the past two years, so the situation could change dramatically depending on future hardware advances.

---