Liaust

turning questions into excessively large projects ;
ai systems · reasoning · physics · philosophy

[Liaust @TheLiaust]

no way MUSE CODE will be better than CODEX
I am too much of a devotee to openai to accept it

[Mark Zuckerberg @finkd]

Releasing Muse Code in beta today. It's a terminal coding agent that takes on complete software engineering tasks across large repos: planning changes, writing code, validating the results. Powered by Muse Spark 1.2, a coding-focused model update.

[Liaust @TheLiaust]

@DanielLockyer even harder to beat an obsessed one

[Liaust @TheLiaust]

@mckaywrigley Gotta love the good vibes agents

[Liaust @TheLiaust]

@nikitathakur21 Doing the same here! Just started this week, still feeling the playground

[Liaust @TheLiaust]

@craigweiss some say they are preparing something very large (I don��t believe it. they are cooked)

[Liaust @TheLiaust]

@EXM7777 That is quite helpful and the value capture is what I had realized and will adapt to. One issue that I find is that a lot of content to reply to is templated and the creators don’t engage. How do I find accounts to reply and engage with?

[Liaust @TheLiaust]

I am not sure if I understand entirely, probably because your project is quite complex and specially non-linear(at least it seems that way). Regarding grok compacting context and then you having to reset it, maybe you need a more specific protocol that it follows where it increases the “granularity” of the context. So basically breaking up what is being talked about constantly such that every piece of conversation/decision/action can be retrieved more “specifically”, exactly because the granularity of the context is very high.

So instead of letting the context be compacted you create some flow where memories are created after every message, or even more granular where you make it identify every claim and concept and write to .md so everything you claim or state becomes durable

But of course I am probably misrepresenting your problem. It is hard to understand just via a thread like this.

[Liaust @TheLiaust]

@icanvardar I remember from an Elon musk interview that he said something along the lines of “the most difficult problem is always having a corrective feedback loop”. The advantage of building in public or not hiding entirely is that you are receiving that corrective feedback loop

[Liaust @TheLiaust]

@ishauryansh Will do!

[Liaust @TheLiaust]

This is a 10D point(+time) and a 2D surface(gyroid chamber) made of it

So it is technically a 10D surface(though a very very special kind of surface)

Let me explain.
A 10D point is just an array of 10 values:
(v1,v2,v3,....,v10)
We represent the first 3 with a 3D position (x,y,z) and v4 with a colormap(HUE). So far very common.
For v5 we use the size of the glyph (larger means higher values of v5).
Different values of v6 changes the geometry of the glyph(negative -> zero -> positive means "circle" -> "diamond" -> "cross")
v7 is luminance (brighter or darker)
v8 is orientation
v9 is the thickness of the stroke
v10 is the texture inside the glyph(higher value = more lines)
Combining all of these ways of representation cramps all 10D degrees of freedom into one glyph

Still working on a general 10D plotter with less restrictions but this is as far as we can go without looking very chaotic

[Liaust @TheLiaust]

I see. Now that you mention the “drift” I have had some problems with that as well some time back using gpt. For complex projects it is hard to manage the “scope” of what needs to be built and the exact set of features. So what I did was that I started following a HARD planning phase, where I would create an overview.md and implementation_slices.md file for basically every new feature or change. Not only that but every new feature gets its own worktree local and git, so it is completely reversible. And then with codex I use /goal and explicitly tell it to implement only these features and to consult me if it had to decide anything meaningful that is not in the planning files. And then with this workflow I have had quite a lot of success with it not leaving the scope. But it does require an in depth planning phase per feature.

But maybe I am misinterpreting the problem you are having

[Liaust @TheLiaust]

@ishauryansh Thank you! It is a silly idea of cramping information into a unit, so we can represent higher dimensional objects but maybe I will find a way to make it better and actually intuitive enough so we can see more dimensions in different ways

[Liaust @TheLiaust]

@ishauryansh Well regardless of what the algorithm thinks, I will share what’s on my mind and will explore circles of interesting people from every area I can imagine. Meeting like minded people is probably more important than a niched audience for me, at this stage

[Liaust @TheLiaust]

I wonder how many people have lost their individuality to fit into the ideals of "how to grow on x"

this is my second day posting and, as I always do, I have been obsessing about how to optimise growth on the platform. the conclusion is very simple

1. define a niche
2. define your audience
3. define your value-proposition
4. post and comment at low and high rates
5. double down on what works

but... do you know anyone who can be simplified to a single niche? to a single audience?

it makes me question just how many people are offering less than what they, as individuals, can. How many are limiting how interesting they are by turning their accounts into an "operation" instead of a displaying all their "overwhelmingness"

[Liaust @TheLiaust]

@kitkatkittenlab @elonmusk That is an interesting experience. Thus far I have basically only used OpenAI and never had a problem with it. Though I do need to study more about grok, and even Claude, because honestly I kind of locked myself in the OpenAI bubble

[Liaust @TheLiaust]

@ishauryansh So real…

[Liaust @TheLiaust]

@edo_lunardi @thecosmos That is awesome, I love that style. I use something similar for my website, by far my favourite

[Liaust @TheLiaust]

@DinisPetrukha Wait are you Portuguese? I hadn’t seen that. I lived there for about 9 years, but I was born in Brazil!

[Liaust @TheLiaust]

@craigweiss Segregation is also increasing I would say. Truly smart people that are keeping up and not falling towards “minimize my brain becomes ai does it” are becoming far smarter and ahead. Those who are not keeping up, either stay where they were or become dumber

[Liaust @TheLiaust]

It the end it becomes an optimization problem, where you must choose what to maximize and how much you want to take away from one side or the other. I am still not sure if the algorithm truly rewards specialized content instead of generalists but it does sound logical. Hopefully it is less punishing than people make it sound

[Liaust @TheLiaust]

@pcshipp Does anyone like it? Only if it is something really interesting but rarely that’s the case