Pinned toot


Software "engineer" by trade and study. enthusiast, loves . Also looking to pivot away from pure software, and into , or . Or in any other way contribute something meaningful back to this world. is my OS. I may have made a for it once.

TeMPOraL boosted
TeMPOraL boosted

RT @MachinePix@twitter.com

A Liebherr LR13000 lifting a Liebherr LR11350 lifting a Liebherr LR1350/1 lifting a Liebherr LTR1100 lifting a toy crane.

🐦🔗: twitter.com/MachinePix/status/

TeMPOraL boosted

"In case you've never heard of Java, its an object oriented
programming language, similar to the latest .NET development languages like C#. However, it's been around far longer and is widely consider the original true developer language of choice." -- _planetminecraft

TeMPOraL boosted

Remember, peeps, units matter.

If someone asserts you have a room-temperature IQ, ask if that's Celsius, Fahrenheit, or Kelvin.

TeMPOraL boosted
TeMPOraL boosted
TeMPOraL boosted
TeMPOraL boosted

@publicvoit Charles Bridge (Karlův most), across the Vltava (Moldau) river in Prague, Czech Republic, construction animation, showing medieval engineering techniques.

This video is the longer-form (3m), higher-quality version of a gif circulating earlier (see parent).

youtube.com/watch?v=nJgD6gyi0W

Background on the bridge:
en.wikipedia.org/wiki/Charles_

Article (in Czech); praha-archeologicka.cz/p/212

HN discussion: news.ycombinator.com/item?id=2

#CharlesBridge #KarluvMost #MedievalEngineering

8) Bottom line: InsightMaker is OK for creation, bad at exploration. It's not enough for what I'd like to have.

What I want is a tool for creating, exploring and *discussing* models. Replacing plaintext backs-and-forths or argument trees (a-la kialo.com) with refining and commenting on dynamic models, forking and presenting alternatives.

I'm not sure if this would be a useful addition to social conversation, but it may as well be. I guess I've found a new side project.

8/end.

Show thread

7) If it was so easy for me to make such stupid mistakes modelling and almost miss them, how difficult would it be for a consumer of such model to spot them? If they wouldn't, they could've played with the parameters and leave with wrong conclusions in their mind. It's the opposite of gaining insight.

(Note that it would be easier to spot these mistakes of the presentation of dynamic state was better; see point 5.).

This also raises possibility that such models could be used to easily lie to people.

7/?

Show thread

6) On : not only is it hard to do right (per point 1.), but also it's somewhat easy to get it wrong and not realize it.

I've made two mistakes that exemplify this, that I caught last minute before publishing the model.

One, in use of EROI, which I noticed because I set EROI to < 1 and the model did not explode to infinity, as it should. Two, in CO₂ emissions calculations, because increasing plant efficiency increased emissions (it shouldn't!).

6/?

Show thread

5) What this means is that InsightMaker isn't the perfect tool I'm looking for. It's too hard to be a consumer of a model.

What am I looking for then? Something like Bret Victor did towards the end of worrydream.com/LadderOfAbstrac - interactive, multidimensional diagrams allowing you to abstract over some parameters. And easy inline graphs like he demoed here: vimeo.com/67076984#t=350s.

Basically, the ability to inspect the dynamic behavior of a system interactively, as you tweak it.

5/?

Show thread

4) InsightMaker is good at constructing simple , but sucks at exploring them.

There's the "Link results to model" feature, but it's hidden under a "map-pin" shaped button on the results graphs. Only activating that gives you ability to see the charts change in realtime.

There's bunch of other obscure features like mini-charts that show when you hover over a corner of a item in the model, but only when "Link results to model" is active (results window has to be open for that).

4/?

Show thread

2) On a positive note, I've learned some things about the problem domain. Like, what coal is made of (not just carbon) and how different types of coal and burning conditions affect emissions (not accounted for in the linked model).

3) The model is slightly more complex than it should be, because it was easier to find usable kWh->CO₂ numbers than tons of coal -> CO₂. So the part that goes through "Perfect burn energy equivalent" to CO₂ is just laziness on my part.

3/?

Show thread

1) Modelling a conceptually simple feedback diagram as a dynamic simulation turned out to be complex and time consuming. Some of that was because of unfamiliarity with the tool, some because I went the hard way and did the simulation accounting for units of measure.

But most of it was because turning a vague idea into executable model is *hard*. Making flows correct, avoiding circular equations (though that's probably a limitation of the simulation engine), sanity-checking dynamic behavior.

2/?

Show thread
TeMPOraL boosted

144 000 times per second, They ask a question.
144 000 times per second, It answers.

- Is this a dog?
- Yes.
- Incorrect. Is that a dog?
- Yes.
- Incorrect. Is that a dog?
- Yes.
- Incorrect. Is that a dog?
- No.
- Correct. Is that a dog?
...

Perhaps a stupid question, but: did anyone ever try to drug an AI?

I mean, take one of these DNN models - particularly generative ones - and apply a biased noise function to its weights. What exactly would happen?

We've already seen what DeepMind dreams of. But what does it dream of when high?

TeMPOraL boosted
Show more
Mastodon for Tech Folks

This Mastodon instance is for people interested in technology. Discussions aren't limited to technology, because tech folks shouldn't be limited to technology either!