jmacc93

I've heard it's related to the mathematician Bernoulli somehow

Actually, I just did a search along those lines and I found this thread https://news.ycombinator.com/item?id=49988852

They link to https://news.ycombinator.com/item?id=49723267

I'd still like to see a more in depth explanation why they chose that word. It's probably going to enter our lexicon to some degree

I've been reading stuff from the rationalist community for a long time. Here's some related comments that might be helpful:

* Writing to think -- writing about a subject to understand it better -- is a common technique in online rationalist circles

* I often just throw long (esp rationalist) essays at my LLM text summarizer script to see if they would be something I'd like to read. It works well for me

* Rationalism the subculture, rationalism the skill, rationalism the subject, rationalism the mindset, etc, all seem pretty distinct. This comment is more for the comments here (mftch)

* Inducing a trait from a few examples to a whole group (esp a long tailed (ie diverse) group) is usually a bad idea. Exposure is often a good remedy for that. (mftch)

* It might just be me, but it seems that both the online rationalist community and hacker news comments cultures have evolved a lot in the last few years (mftch)

Every time I see stuff about age verification (which I do admit is likely just a justification for control and surveillance over the internet) I can't help but wonder if there's some way to coordinate a large enough number of internet users to adopt a parallel internet that inherently cannot be regulated, or at least one in which no law can be reasonably enforced. For instance a fully encrypted p2p internet. I guess ultimately the problem is banks and other tie-ins to the real world like businesses with public reputations. And so I guess since endless content streams (including porn and other child-unfriendly media) follow the money primarily, any non-monetized / volunteer parallel internet would likely contain mostly unmonetized content like information (eg: wikipedia, Q&A sites, etc) and slow, un-gameable social media

In a perfect world, I'd prefer law and regulation minimization, but obviously that's just not reasonable in our actual world. As much as I have libertarian tendencies, I just can't argue that regulation is necessary. The really interesting part is which regulations are the ones you actually need, which do more harm than good, and which should be refactored -- all as measured (probably) by something like counterfactual real world preference (in the best case, at least). My sense (which isn't worth much; only should act on empirical evidence and falsification-resistant modelling) is such a parallel internet, plus an analogous parallel economy (like bitcoin), would be a net positive

Semi-related: I always wished there was something like the unholy combination of a spreadsheet and notebook rolled into one. I picture it notebook-like at the top level, then each cell is a widget that the host language can reference parts of in other cells (probably with a variable (eg: something like mathematicas `Out[_]`) or a built in construct (like the `$ABC123` forms in spreadsheet formulas)). A notebook interface would also be good (I think) as a straight up terminal as well, as you (I) typically want to run commands in an order like in a notebook

The thing that gets me is these large companies obsess over their UIs, so don't they consult with UI experts? Yet why do their UIs become predictably worse over time? Perhaps a certain subset of people actually really like featureless, pseudo-minimalist, vague, vacant, and contrary UIs??

"Please don't write lots of code unless I explicitly request it, or its the only way. Just show the minimal code necessary to respond to my request"

It ignores it literally every time lol

I use Alacritty and really like it :). I use it primarily for scripts and other cases where I want a terminal to open extremely fast. Some day I'll get around to finding a way to easily use tabs in it and I can replace Terminator (which I also really like and use, of course; but its a lot slower to open)

Yes! Finally something relevant I can post to hacker news

I use this little script https://github.com/jmacc93/paste-to-tmux-script daily. Its for pasting your current clipboard item to a target tmux session, along with very simple dsl for controlling which tmux target to send to, opening the tmux target, killing tmux sessions, etc. It makes a vastly, vastly better and more productive repl for me. The workflow for using it looks like typically is like: use `@sw name_of_target`, `@open`, `name_of_executable`, then move my cursor to something I want to evaluate, use ctrl-c to copy the line, and ctrl-. to send it to tmux

I think I've changed stuff around since I last updated that repo, though, but the general idea and skeleton is there, if anyone wants to use it / hack on it. And I'm gonna do a rewrite of it soon, I think, so that instead of using a dsl (the `@...` forms above) it uses `!...` or some similar form to execute arbitrary shell commands

I'm hoping to (soon hopefully) integrate it with my https://github.com/jmacc93/noca notebook canvas program as well

I've long thought a good challenge to aim development toward in machine learning is to make an agent that can finish Skyrim zero-shot, unassisted, in real-time. I'm just a hobbyist but I thought I'd try to see if I could figure out what this would look like

I'm very interested in this topic and if anyone is interested in talking about it, please let me know. You can also email me at joe_miller_93@protonmail.com

I'm thinking about trying to write and train a superagent (an agent composed of agents) for my next project. If anyone is interested in talking about that, again: let me know :)

This is a demonstration prototype I made of a transpiler to Javascript from a variant of Javascript extended with type conjunctions and tag types (I describe both on the page) for a form of dynamic dispatching via implicit guard clauses. It is very rough behind the scenes and is really likely to break if too much is done with it haha. But its really just to demonstrate and illustrate type conjunctions and tag types. The explanation on the page is hopefully worth more than the demonstration, anyway

I hope you all like it :)

P.S. the demonstration requires javascript but the other stuff on the page I wrote in markdown then used pandoc to turn that into html and manually spliced that into the page's index.html. It pulls CodeMirror from the web, and I think that's it iirc

When I want to use behavioral directives, I've been putting the directives for ChatGPT in a parenthetical reminder header at the top of all of my messages to it

For example, you would say "Don't apologize", and then after that you would start every message with "(Remember to not apologize)"

I've also learned from using local LLMs that if you force the LM to start its response with something (via the 'start response with' field in text generation webui), then it will just go with that thing. This can be used to prevent RLHF-trained models from not responding because they think the question is unethical or illegal, and from giving the typical "I am a language model bla bla bla" responses. To be clear, if you put in the start response with field "Sure, I can answer that!", the LM will just go with it and not respond with "that's unethical" or "you're a horrible person for even conceiving of such a thing. they should lock you up for life!!", etc etc. It seems there is a similar effect when you edit the LM's past responses in some way, the LM's new responses will mimic that way of responding

Carrying that over to ChatGPT, it seems that if you request ChatGPT always starts its replies (via parenthetical reminder as well) with "Sure, I can answer that!", or with "I apologize", that that does seem to affect how it starts its replies. There appears to be some cases where it will say, for example "Sure, I can answer that! As an AI language model, ...", but it seems that forcing it to start its response a particular way helps prevent it from apologizing

But generally, for the apologizing thing, I just downvote it when the apology doesn't make any sense, and otherwise ignore the apologies, as others are doing. This indicates a potentially hazardous 2nd order effect where people are trained to ignore ChatGPT's apologies. eg: Boy who cried wolf, etc

This webpage caused my laptop (4 GB) to OOM. Really cool though!

I've discovered that when role-playing (or really doing anything that extends over many messages) with ChatGPT, its a good idea to add a header to all of your messages that describes what you're doing, a synopsis of the conversation so far (you can use ChatGPT for this), and what you want ChatGPT to continue doing. ChatGPT (GPT 3.5) seems to lose context really, really fast, so this technique helps keep it focused on what you actually want

I'm like 70% confident that a LLM that is constantly condensing its own memory (its immediate memory / input, and condensed memory) using internal prompting, and conditioning its output using both that condensation and its immediate memory, would lead to effective long term memory. You could probably make a model that does this on multiple scales, with condensations on top of condensations, and so on