Show newer

Chapter 9 of [[Quantum Computing Since Democritus]] started very strong, you could say we're vibing.

All social networks should run ranking algorithms client side in a user-debuggable, user-overridable way. Is this as obvious to everyone as it is to me? Because if so, I think we should just petition all governments.

I just had a galaxy brain moment, perhaps it was the [[epoisses]]

To encode a mapping of the range (0, 1) to the naturals, picture a scheme based on conventional prefixes. For example of nines that lead:

template like: 0.<nines>.<number>
regex like: 0.(9+)(\d+)

\1 is a natural number in unary coding, \2 is another natural or a program or data

Show thread

Because you can encode anything in a string of numbers (per Gödel), you can state a number [[r]] in (0, 1) and essentially communicate a program or a description of an arbitrary Turing machine and input.

Show thread

In essence you could use 0..1 to map reals into natural numbers, essentially using a subset of the reals to encode an arbitrary mapping. A relationship between two sets.

Show thread

In this scheme, the version could also be taken to mean an intent of meeting an SLO. But in that case you would lose bijection in the 0..1 range.

Show thread

Because I'm a hack only 0 and 1 are implemented.

I'm thinking of perhaps adding nines instead of going the natural 2, 3, 4...

That is, it could go 0.9.anagora.org, 0.99.anagora.org, 0.995.anagora.org and so on.

Show thread

I guess I never told you about the versioning scheme: 0.anagora.org is production running version v, 1.anagora.org is v-1, etc.

So if HEAD is broken, you can go back in time and run something stabler.

Show thread

The [[agora]] is a bit slow, load is high and I shipped a computationally intensive change. Please excuse any disruption while I work on it and reach out if you need anything.

As usual, you can visit 1.anagora.org to access v-1

---
RT @flancian
I use screenshots a lot and plan to record my screen and perhaps run on a snapshotted VM while I work on Flancia, so it seems to me like this could ship with an example model called [[flancian]].
twitter.com/flancian/status/14

Show thread

When you see a [[typo]] in my posts, you can assume there's a follow up tweet that corrects it. I'll try to keep them linked.

IMHO this beats deleting/reposting because it keeps full history.

This is relevant for project snapshot precisely.

---
RT @flancian
Idea for a project: train an ensemble of neural networks to move a cursor and type text. By default it seems to me GPT/Bert/Meena solve the latter.

Corpus comes from the [[agora]]…
twitter.com/flancian/status/14

Show thread

Idea for a project: train an ensemble of neural networks to move a cursor and type text. By default it seems to me GPT/Bert/Meena solve the latter.

Corpus comes from the [[agora]] and [[snapshot]].

@an_agora
@agora

I use screenshots a lot and plan to record my screen and perhaps run on a snapshotted VM while I work on Flancia, so it seems to me like this could ship with an example model called [[flancian]].

Show thread

Idea for a project: train an ensemble of neural networks that to move a cursor and type text. By default it seems to me GPT/Bert/Meena solve the latter.

Corpus comes from the [[agora]] and [[snapshot]].

@an_agora @agora

Show older
social.coop

A Fediverse instance for people interested in cooperative and collective projects.