scribu

I own a Kinesis Freestyle2. It's collecting dust because I've found the keyboard layout suboptimal. My thumbs would awkwardly reach inwards for the modifier keys.

That's so fun!

Having a background rhythm makes it a lot easier to create a decent groove.

Did you read the part where it says that the people writing the content don't receive any of the donations?

And that the foundation's software initiatives have been largely innefective at getting more/better content? [0]

(I've also donated to Wikimedia in the past and contributed to Wikipedia itself)

[0] https://forum.effectivealtruism.org/posts/9qqds7Z3Ykd9Kdeay/...

The paper proposes a ban on almost all advertising:

> It is hard to believe that as late as the 1970s the Federal Trade Commission (FTC) viewed non-false, non-misleading advertising as anticompetitive conduct capable of violating the antitrust laws.

> Advertising has the power, through repetition and brand image creation, to induce consumers to buy things that they do not really want, to the disadvantage of competitors selling the things that consumers would otherwise buy.

Ah, this is so handy!

It explains why some Economics degrees have a course on Cybernetics.

I'm curious how it differs from PyPortfolioOpt.

https://github.com/robertmartin8/PyPortfolioOpt

The lifecycle diagram is legit.

As for the vendors, Nvidia is just promoting their own services or their partners.

But there aren't any established players yet. It's still a nascent field.

Indeed:

> test_onefile.c - This file implements a demonstration VFS named "fs" that shows how SQLite can be used on an embedded device that lacks a filesystem.

Dropshipping has been around for a while.

What's new is using Influencer Marketing to lend more credibility to knockoffs.

IMHO, a better title would be "Probabilistic Theories Of Everything"

Contrast with the recent work from David Deutsch, which eliminates all probabilities from physics.

Intro to Constructor Theory: https://youtu.be/wfzSE4Hoxbc

The post makes a good case for why it's a bad idea to split a business problem into slices. Even though you have multiple small teams, they still need to closely coordinate with each other.

But I can't find the next post in the series which proposes a better alternative. Anyone else had better luck?

Model visualization can be useful even for experts, either as an additional sanity check or to debug surprising results.

I couldn't evaluate this Otto thing because I'm on a laptop right now and it looks like the developers assumed that everyone has a 40" screen.

Source: ML is (part of) my job.

Did anyone else notice that the biggest non-academic supporters of functional programming seem to be financial institutions?

Is it due to the nature of the problem space? Is it because finance people tend to be more analytical? Something else?

I feel like Black Mirror Bandersnatch gets closer to the promise of interactive video. You get choices which lead to meaningfully different endings. And you get a little reward for playing it through multiple times. :-)

Not the author; just a fellow ML practitioner.

In theory, yes, you can feed these new distorted images into a new model and defeat the cloaking.

In practice, building new training sets is much more expensive than slightly changing the cloaking algorithm.

Shapecatcher is a free webapp that allows you to draw any Unicode symbol:

https://shapecatcher.com/

> unless the product changes to suit my needs in some way

But isn't that the point of the article? To figure out how to change the product so that it meets more people's needs.

> No question about whether potential users are interested in the product in the first place?

They do show interest, tho. From the article:

> There are a set of users who show intent for your product but are not quite able to get over the hump. Those are your adjacent users.

Why is it pathological?

It's cool that you can immediately update the output whenever the input changes.

But what if the code inside the function changes? Do you have hot reloading functionality?

Streamlit handles this by re-executing the whole script when detecting any code changes. You can cache certain resources (such as the model), but it's fiddly to get right.

sklearn's BallTree has a query_radius method [1], but it's exhaustive, rather than approximate.

[1] https://scikit-learn.org/stable/modules/generated/sklearn.ne...

Are you point out the removal of the Apace License header or what?

Potentially!

Looks cumbersome to set up for iOS, tho: https://www.reddit.com/r/Syncthing/comments/ese82l/ios_users...

It's not that. Wikipedia, for example, has a separate "pages that link to this" view.

What makes Roam Research different is "block addressing", i.e. you can get a reference to a single paragraph and use it in other docs (either referencing it or embedding it).

(and of course a nice UX for editing)

The drawback to working directly on Markdown files is that it's hard to synchronize your notes from/to mobile.

If I didn't care about sync, I would use org-roam or some Vim plugin, personally.

This is one of the clearest explanations I've read on the difference between traditional Statistics and Machine Learning.

Where you using a different search solution previously? What was it?

What if you first trained a classifier that told you if the uttereance is a single word vs. multiple words? Then, based on that prediction, you would use one of two separate models.

The technique you're thinking of is called oversampling and there are many other general techniques for dealing with imbalanced datasets, as it's a very common situation.

It's easy to see from the uptime history [1] that there have been many more incidents in April-June than in January-April.

Don't know if that has anything to do with the Microsoft acquisition, but it is concerning.

[1] https://www.githubstatus.com/uptime?page=1