The Dark Dark Show

The solution was actually pretty simple. Use my email provider to create an email address solely for use with Microsoft login; not to be shared with anything else. Then, set this as the ONLY email you can use to log in with (excluding even outlook email addresses). This has no effect on your access to your outlook addresses or inboxes, and you can still use them as normal. However, you just can't log in with those addresses.

ToolHive in theory is a great idea, but seems to break everything when you try to use it with LM Studio. It seems intuitive to add custom servers to the interface, but then they don't actually work with clients.

AnythingLLM seems ok, but it seems bloated for what it is... Which would be fine if it also was decent as a central MCP server-host... But it's missing a lot of plugins, and only STDOUT servers work with it.

I really wanted to avoid having a "central" json file, but it seems like that's what I need to do. Then not helped by the fact different software is designed for different versions of the MCP standard.

All I've been aiming for is a central repository, including a searchable database of my bookmarks (of which I have a lot of), maybe augmented with my last.fm listen history, and with previous AI conversations irrespective of platform also included in it. A place I can keep track of my knowledge.

Continue.dev was an extension for Visual Studio Code which allowed you to use AI models to, ideally, vibe code (though I used it to learn coding, as AI coding always throws up errors which I can then try to fix).

The big difference between it and other pieces of software (like Windsurf) was it was open source and let you use your own models. After some setup I got performance about as good as using Windsurf, except I was using my own models run on my graphics card meaning the cost was, at most, my electricity bill and some time (rather than paying an external server for a certain amount of tokens).

Maybe there's a better solution these days; I feel like Visual Studio Code isn't that far from having built-in support for your own AI models. It's just frustrating to have something open source get swallowed up like this, with no direct replacement.

It could be I'm just not read up on AI news, and actually there's already a solution out or around the corner. The big problem with AI news is it feels dominated by techbros and hype, rather than grounded articles that allow a user to experiment with AI at home. There are a few sources that have help like this, but they tend to spread out- probably because they're not as profitable to redirecting you to buy tokens from OpenAI, Google, Grok or something.

Just a reminder that this book exists, from a former Meta employee with a gag order against her not allowing her to be anywhere near her book, to the point of spying on her every minute of the day to ensure she keeps to it. Including, among other things, the dangers of Facebook's algorithms to society and our children. Literally billed as the book Meta doesn't want you to read. https://amzn.eu/d/05Lqtlyl

I was really concerned by this post, and I've been there in my younger years. Coming across messages in a relationship that changed my perspective and believing I was in the right for it (I wasn't). The only difference being the advent of AI in the last few years.

I particularly didn't like the framing, essentially, of "What would you do if you saw this?". I know what I'd want to do, of course I'd want to know what was said. What I would have done, these days, is invite the conversation: I saw a thread with this name in your ChatGPT logs, and to be honest it has really knocked my confidence, could we talk about it? Maybe he'd be dismissive or cross; maybe he'd be completely open and even show you what was written.

I would argue he did exactly the right thing: talked about these issues in complete confidence. I find it really hard not to make the comparison of whether he would be a better, or worse, person for posting these thoughts to a personal Substack instead (without identifying you). This isn't intended as an attack on you, as your piece was thought provoking and lead to me leaving a comment on a Substack for a first time on this account because it stuck with me. But the fact he wasn't talking about this in a public manner suggests he really did want to protect or save the relationship.

I also feel people use AI in the wrong way. When I use it, I often take an argument to the extreme conclusion, then ask the AI to "steelman" the alternative argument. LLMs that I use tend to have strict instructions to not be sycophantic, and to back up all points with evidence. This has been a huge help in identifying friendships that were actually becoming toxic, bringing me to the next step of asking actual friends whether this is actually what is happening or not. Being able to test thoughts to the extremes is a huge help for me; indeed sometimes shouting the worst about someone is also cathartic. I don't think this is the first, or the last, relationship to come to an end as a result of an LLM.

I really wanted to add, however, that where breaking this privacy would have been reasonable would have been where there's evidence of a crime being committed, or evidence of abuse. And where it would be much more understandable would be where there was evidence of cheating.