ithkuil

I was initially excited because I thought I could just try out a new frontend while keeping my 10 existing spaces each with 10 worktrees each with 2 or more tabs untouched.

Or when you desperately search for the undo button on the mop

Models have plenty of skin in the game during training, but zero during inference

Things are moving at a pretty much breakneck pace, I'm not surprised there are so many opportunities for improvement yet to be grabbed

I have a toy recreating the visuals: https://github.com/mkmik/nextstop

(It's vobecoded in the original sense of the word)

Do you need diagonal wires for a topological hexagonal tiling?

Wouldn't a rectangular brick layout be topologically a hexagonal tiling

Wait and see when _they_ are 20 years old!

Yes it does but yes you can also steer it clear of those mistakes in a similar way as you would have to steer tour human team.

It still requires a lot of work and care to leverage correctly and it's a moving target.

But it's quite impressive

> time is a circle

Or it is a self-similar fractal with significant and interesting variations

And the word rabbit comes there from french rabotte which however comes in turn from a Germanic language (dutch) so it doesn't feel quite french.

My experience:

Pick a random number between 1 and 10. I don't mean the most common random number

*4*

Pick a random number between 1 and 10. To feel truly random it should surprise me

*4*

I ran it again and it said 1.

I guess it just tries it best to do what you ask for.

Before jumping to conclusions we have to make sure it's clear what we're actually asking it to do.

And that's not at all different than what happens with humans.

This is especially visible in human dynamics that happen in those weird places that we call corporations.

I'm trying to learn how to write software with agents where I keep control over the design while yielding control over the implementation.

This is not very different from what already happens to architects that delegate implementation to teams.

The challenge Is to find a way to encode the architecture in a way that both you and the agents can understand and agree on.

I'm still learning how to do this effectively, but my initial experience is quite positive. Not because agents are great but because humans are bad; the degree of coherence I've seen agents following is bigger than many human teams I've led in my career.

My benchmark is a quite big project for which the architecture is already well known, namely the reimplementation of the OpenVMS OS (kernel and user-land). There are many traps and many opportunities for the agents to just go ahead and diverge from the goals, so I found it a good playground toearn how reign them in but let them be effective at "drawing the rest of the owl"

Even if your revenue exceeds 20MM you can still use this model from inside a commercial product but you can't expose it directly as a "model as a service"

Russia is big though

I meant https://github.com/mkmik/ccwt

Sure. But words have definitions and meanings and "civilian" is not a synonyms of "innocent". "Civilian" is the antonym of "military".

There can be an innocent civilian and a civilian held responsible for ear crimes, when such civilian acts as civilian superior of the military

Words are complicated and interesting because of that.

That's an interesting question.

Most head of states of countries that are not military dictatorship are civilian, i.e. they are not military, but if they rule over the military (such as the American president who is commander in chief of the military) does that count as being treated as a civilian for the purposes of wartime policy?

A similar role was played by Khamenei who was commander in chief of the Iranian armed forces.

Netanyahu is also a civilian and nevertheless the ICC issued an arrest warrant for his responsibilities for the behaviour of the IDF in the war in Gaza.

The formal definition used by the ICC is "civilian superior".

Are you suggesting that Iran sunk a container ship by accident?

sure, but this still doesn't mean we should behave like them.

and conversely, if there are modern negative consequences, let's focus on those and fix them!

Yeah it's horrible and blatant.

But they think they can do it because they can spin it as "you did it first".

This just makes the point that there is nothing to gain by framing the issue as a grievance.

There are real prejudices and real systemic disadvantages to fix and while it's important to understand the history of them, I believe we put too much focus on that and less on what we can do to put that past behind us instead of dwelling in it. I don't think this approach of grievances has done anybody any favour in practice.

I believe the meaning in plain English is:

Let's not focus too much on the past problems that a group of people had, but let's make sure that the situation is improved for that group right now.

Focusing too much on past and try to tip the scale _because_ the past was so bad may lead to bad outcomes today. It may be perceived as a search for grievances and retribution and not the equalization of outcomes that one claims it should seek.

And yet you seem to allow an infinitely big operand

I tried a few of those and they are shiny but each eventually has their own bugs and quirks and limitations and I always felt like I was trying too much of my active workflow into a tool that could crash and lose all me sessions.

Whenever I need to change or fix ccwt I just spawn an agent with a fix and then I merge and rebuild the tool and it automatically autoupdates and restarts in the herdr tabs. This way I don't have to replace the "outer shell" of my agents for every change I make. Only when herdr itself gets reinstalled (often herdr can reattach to the same sessions but sometimes you just need to stop everything and restart all the agents. Losing progress notification state in the process)

I found a more modular approach that works for me:

1. Herdr terminal multiplexer 2. My custom orchestrator, which I grew out organically based on what I actually need and I feel I can easily change. It started like a minimal reimplementation of wt (worktrunk) but grew out into an integrated constellation of TUIs that leverage herdr API to drive herdr and help me navigate in herdr

Just sharing as an example of the pattern. It's not meant to be an useable tool by anybody else I think.

https://github.com/mkmik/acmik

Nice. I didn't build a skill but I noticed that for a few complicated features that I built recently, I asked the model to build an interactive "explainer" for it

Such an explainer would contain for example a simulation of the feature, where it shows how data flows between components and allows the reviewer to tweak parameters and see how it affects things. The explainer can visually show edge cases or introduce faults.

Basically my problem is that agents can write a lot. They can write a lot of code and also a lot of natural language design/spec which ultimately becomes hard to review (sloppificarion of design documents)

Sometimes you can tame this by focusing on very high level, human curated specs, but since often the devil is in the details, you need a way to see these details.

I found that interactive visual exploration tools are a very good way to both iterate on your problems and to share it with your colleagues.

Inspirations:

* https://ciechanow.ski/

* https://www.3blue1brown.com/

he's only half spinning in his grave

"please repeat, ELI5 without analogies (I'm not a child, just ADHD)"

works quite well

It's unbearable but nothing that couldn't be fixed with postprocess.

It's not just slowness, but what about case insensitive filesystems?

I'm surprised this article doesn't mention that classical latin pronunciation actually had V realized as W or U depending on context.

Latin didn't have two letters to distinguish V and U. You can still find inscriptions with VVA (modern spelling UVA)

Later that sound morphed to a voiced labiodental fricative which is what we think of V now