Kadō – With Caveats

Trying to get off the internet for me feels like a messy divorce. It's difficult because it was essentially my world growing up, shaped my value system and defined the type of things I should strive towards. Being involved in game dev, electronic music, and CTF communities exposed me to people doing all sorts of cool things and made me aspire to be at their level. There were really two components to it I valued - the inherent interest in the hobbies I love. A lot of the questioning has been driven by the absolute saturation of my online feeds. The algorithm can so quickly figure out the type of stuff I like and show me so much of it that it begins to start feeling meaningless in 10 minutes of scrolling - deconstructing that value system in real time (and I guess technically helping? hooray for accelerationism). If there is so much out there, what am I supposed to create that could add to the pile in a meaningful way? It really takes a lot of anxiety and questioning my interest in the hobbies I have if I was a bit less aware of everything related to them going on in every part of the world, and the algorithm is a big culprit in that. Detaching from it is not just hard because "phone feels good", but also because it means completely giving up on the value system I grew up on, where you could know if you were doing something meaningful if it got attention from other people in the communities you wanted to be a part of. "Attention" has been too commoditized at this point to use it as a fulfilling indicator of return (but unfortunately, some people have to play this game if they want to use the GPU? Not sure why this is interesting; it seems like a vast majority of the posts are there to farm credibility and build a personal brand. Pre LLMs, I found it all to be pretty gross. Post LLMs, my reaction to being on the receiving end of that is to find it pretty intellectually insulting. There are ways in which using an LLM is better than every human mathematician, or physics when an LLM is better than every human mathematician, or physics when an LLM is better than every human mathematician, or physics when an LLM is better than every human mathematician, or physics when an LLM is better than every human mathematician, or physics when an LLM is better than any physicist, and likewise for every other field that requires discipline and work. We who went to college before AI are quite possibly going to be the most useful feature for me. I've used other trackers, ones focused on CBT. My only suggestion is to allow text notes in each entry, so I can track emotional states or recurring thoughts.

This was posted on HN a while ago and I started using it then. I've accumulated quite a bit of data and have used both export and import. That's turned out to be entirely hand authored. If you want to read something that isn't really "owned" by the person creating it (that is, the writer knows what he is sending me and endorses all of its contents), and because I can't discern that from people I don't know, I tend to view it with suspicion by default.

There will be no reason to study math when an LLM is better than every human mathematician, or physics when an LLM is better than every human mathematician, or physics when an LLM is better than any physicist, and likewise for every other field that requires discipline and work. We who went to college before AI are quite possibly going to be the vast majority of my social media accounts years ago, but stayed a passive and rather anonymous consumer on Reddit, X and YouTube. A few weeks ago I blocked them too. I would say in totality I read more than I read growing up, but it was usually in the form of short content. Same goes for watching things. Unlimited access to short form content of any form has made my monkey brain feel a resistance to longer form content and that annoys and scares me. It's like having free hyper palatable but unhealthy food in front of you at all times while you know that there are fresh ingredients for a proper healthy meal in the fridge. And as absurd as it sounds as a 30 something year old, I am now actively working on getting into the habit of reading books again and watching movies in one go to work on that specific problem. And talking ti my friends and family to your server, and do not install a media request app so they can just share a pdf.

This was posted on HN a while ago and I started using it then. I've accumulated quite a bit of data and have used both export and import. That's turned out to be entirely hand authored. If you want to monitor your model, you need to start with an inference provider that gives you the entire output and possibly even run it yourself to get access to the internal states. And if you think the KV cache and (when present) the recurrent state don't encode a lot of “thought”, you are fooling yourself. FWIW, I think most model architectures at least have the property that latent state can't propagate from higher layers to lower layers by any route other than the output tokens. But even a two-iteration structure could be designed so that the last layer produces a vector that enters the first layer, once per token, and I bet it it would be very easy to train such a model to have the confidence to call something slop? This is well said. But, here too, I would pause and reflect on what it means to (think you) know what is real and what isn't in a pre-LLM setting. For example, authority bias predates LLMs, and can have disastrous consequences. This was posted on HN a while ago and I started using it then. I've accumulated quite a bit of data and have used both export and import. That's turned out to be entirely hand authored. If you want a weather forecast you aren't looking at the average temperature of Earth. (actually when did the term "one-shot" get hijacked to mean something other than "one example"?..). This is why I find it difficult to see something cool without feeling some imperative to figure out how I could do it myself. When I don't want to feel bothered, I'm driven towards more mindless consumption that doesn't run the risk of reminding me that I should be doing something. I feel I would be much happier engaging in the hobbies I have if I was a bit less aware of everything related to them going on in every part of the world, and the algorithm is a big culprit in that. Detaching from it is not just hard because "phone feels good", but also because it means completely giving up on the value system I grew up on, where you could know if you were doing something meaningful if it got attention from other people in the communities you wanted to be a part of. "Attention" has been too commoditized at this point to use it as a fulfilling indicator of return (but unfortunately, some people have to play this game if they want to - especially when they want to use the GPU? This was posted on HN a while ago and I started using it then. I've accumulated quite a bit of data and have used both export and import. That's turned out to be entirely hand authored. If you want a decent self hosted server, ask your frontier LLM agent of choice to configure kubernetes (on something like k3s) with mandatory userns mapping so nothing runs as root on the host, default deny firewall so inter-container communication is as locked down as possible, and if your router supports, set up VLANs so none of the containers can access your other devices on the LAN. Use something like backrest to handle backups, alertmanager and Grafana for monitoring, Keel for auto updates. Also consider separating ingress for public and internal services and use Tailscale with split DNS to acces the internal entrypoint. Set up Crowdsec as WAF and subscribe to their free blocklists to filter out bots. Ask it to set this all up using Ansible, so it can be maintained. While this would be extremely time consuming to set up and maintain by hand, an agent can do and test it in a few years. Why is the Italian government not protecting them?