What I found interesting this time

A Short and Concise Article on How to Use Tor and Stay Anonymous OnlineThis guide is great for those who have never tried it, as it clearly shows step by step how to install and use the Tor browser.

Blog post that describes the struggle of designing a good API for Claude MCP connections.
It’s interesting to see how others experiment with LLMs.

The article covers prompt injection attacks from various vectors and compares prevention solutions to OODA.
“Observe, orient, decide, act” is a framework for understanding decision-making in adversarial situations.
If you’ve never heard about prompt injection, this might get your interest. The problem is still not solved and probably never will be, as it is inherited in the architecture of LLMs, the datasets, and the way it communicates with users.
#llms #ai

This small entry I’ll split into two separated parts which will contain quotes from the article
"What we’ve found confirms our initial concerns: indirect prompt injection is not an isolated issue, but a systemic challenge facing the entire category of AI-powered browsers."
HAHAHAHHAHAHAHHHAHAHAHAHAHHAHAHahahahhahahahahahahahahahahahhahahaha ehhh… "initial concerns" which researchers brought up right after GPT-3 was created and marked it as architecture flaw.
"This lets simple natural-language instructions on websites (or even just a Reddit comment) trigger cross-domain actions that reach banks, healthcare provider sites, corporate systems, email hosts, and cloud storage."
Jesus …
By the way it really shows that current AI does not have reasoning capabilities and every step of processing needs to be approved by human.

Someone decided to combine my two passions in one post: punk rock and shitposting about LLMs.
Yes, someone asked Google’s LLMs if Fat Mike was a skater.
It’s really funny how articles about him were not summarized correctly.

Just as in title. This is insane and I think governments of Europe should quickly react and put fines on OpenAI.
Thier theft has no limits and it’s disgusting.

A highly detailed blog post about image dithering. I was not aware of any of those techniques. As always, it’s amazing to see an applied algorithm modifying something in a graphical way.

LLMs can be really helpful when working with older technologies. I think this is not something people typically take into consideration when selecting technologies, and it’s a valid point.