I’m not sure if we’re ready for agentic browser control. Yes, you can click each time to accept the risk, but how many of us read the T&Cs before we click accept?

Their 123 adversarial prompt injection test cases saw a 23.6% attack success rate when operating in “autonomous mode”. They added mitigations:

When we added safety mitigations to autonomous mode, we reduced the attack success rate of 23.6% to 11.2%

I would argue that 11.2% is still a catastrophic failure rate. In the absence of 100% reliable protection I have trouble imagining a world in which it’s a good idea to unleash this pattern.

While many educators in the West see AI as a threat they have to manage, more Chinese classrooms are treating it as a skill to be mastered. In fact, as the Chinese-developed model DeepSeek gains in popularity globally, people increasingly see it as a source of national pride. The conversation in Chinese universities has gradually shifted from worrying about the implications for academic integrity to encouraging literacy, productivity, and staying ahead.

The FDA’s head of AI, Jeremy Walsh, admitted that Elsa can hallucinate nonexistent studies.

“Elsa is no different from lots of [large language models] and generative AI,” he told CNN. “They could potentially hallucinate.”

Sounds like a need for some kind of tool to verify that studies are at least real, if not accurately represented, in the answers government scientists are using to make critical decisions.

This looks like a handy package for converting documents (PDF, .docx, .pptx, and more) to .md. There’s also a MCP server so you can use it with your LLM.

To install as a uv tool:

uv tool install markitdown --with 'markitdown[all]'  

A fascinating look into OpenAI the company:

[…] you probably shouldn’t view OpenAI as a single monolith. I think of OpenAI as an organization that started like Los Alamos. It was a group of scientists and tinkerers investigating the cutting edge of science. That group happened to accidentally spawn the most viral consumer app in history.