lachlan_gray

It mentions a greedy decoder in the readme, I'm curious what would happen if you run it stochastically with T>0. Vibe NES? Or does it just crash.

In old school ML a pathological solution is often a collapse of some kind due to an underspecified objective. Maybe one parameter explodes or the model predicts "0" for everything

I read this as "the code is working as intended but the bug is an artifact of the original design"

For anyone like me who finds theory a little dense, you can prepend "quick" to an arxiv link to get a blogpost-style summary

e.g. quickarxiv.org/abs/2608.29530

Ha! Never looked into :r, but I certainly will now

Funny, I was really impressed with how concise and goal oriented the directions are!

I would have done better in uni if the lab manuals were written like this

I think they were doing something like this, the tradeoff is that it's hard to do without an irritating number of false positives and/or wasting loads of precious tokens on useless audits.

Yes, so many times I've been jumpscared by work-related audio in my playlists because of this

Almost too much so, it often feels like opus is pushing back for the sake of pushing back. The way old models used to add disclaimers to every message regardless of content

Tangential, but made me think of this YouTube channel I like.

I have no plans to own a tractor but for some reason many others and I enjoy videos like this one:

https://m.youtube.com/watch?v=pQO-pVxvKvA

Robotics/control systems is exactly what came to mind when I saw this release! What struck me is the possibility of look ahead search in real time, a bit like alphazero's mcts.

Mentioned elsewhere, but

:term claude

In a split goes a long way for me!

Ymmv, but I have been very happy using classic vim’s “native claude support”

:term claude

It will also expand special characters so you can do something like

:term claude “refactor %”

And Claude starts work on your current file right away. Also your buffers will update with Claude’s edits!

According to this, notifications are possible if you add the app to the home screen, which I didn't know.

A feature more devs should use- I've been surprised how much websites behave like native apps if you just "add to homescreen" instead of downloading an official app, e.g. twitter, instagram.

When you open the shortcut, it doesn't launch as a tab in safari, but appears independently in the app switcher. They are often indistinguishable from official apps!

Seems like a great way for devs to avoid app store pains

Not sure about the ssd in particular but the neo is apparently pretty modular

https://www.youtube.com/watch?v=5k7Lv7f-5CQ

Another crazy one is SectorLISP, 223 lines of asm

https://justine.lol/sectorlisp2/

fwiw there are more granular controls, where you can for example allow/deny specific bash commands, read or write access to specific files, using a glob syntax:

https://code.claude.com/docs/en/settings#permission-settings

You can configure it at the project level

Does this mean I get to take a cut for assisting their training data?

Earlier this year moving home to Canada, instead of flying I bought a van in California and drove it back with all my stuff.

I was taken aback to learn my dad did the exact same thing at my age!

What I meant is, assuming that they do find solutions. If they're not doing anything of course that's different.

In the article, "speed" is about reaching specific answers in a specific window of time, the bane of ADHD.

I think part of the message is that speed isn't a free lunch. If an intelligence can solve "legible" problems quickly, it's symptomatic of a specific adaption for identifying short paths.

So when you factor speed into tests, you're systematically filtering for intelligences that are biased to avoid novelty. Then if someone is slow to solve the same problems, it's actually a signal that they have the opposite bias, to consider more paths.

IMO the thing being measured by intelligence tests is something closer to "power" or "competitive advantage".

I think what you're describing is a form of conflict aversion, where the (tiny) conflict is what would clear up your read, or the group's attitude on something, for going forward. Short sighted kindness is a nice way to put it

They have helped me a lot with chunking tasks, and guiding me through tasks that I can't hold in focus.

There's a prompt I used while moving out, where I had claude ask me questions, what is in each room. And then once we had this item list, organizing it.

Is there a similar thing on iOS? I always wonder when a random app asks to “find devices on my network”

This could be a fun example to work with :p

https://en.m.wikipedia.org/wiki/Shakespeare_Programming_Lang...

Reminds me a lot of thermodynamics. Microstates and transition probabilities are a more “fundamental” description, but when you hit the metal, temperature and pressure are more useful in practice to human engineers

I was super obsessed with this for a while! When you have a string instrument tuned in 4ths, there are 2D patterns that emerge which you can use to "derive" or "extrapolate" what a scale shape/pattern will look like across the whole neck

Using a 6-string bass as an example: https://bradleyfish.com/the-notes-on-the-6-string-bass-guita...

You can find a 2D pattern in the white notes (green notes in the pic) that you can use to understand how the pattern will extend from a given point. For example notice EF+BC always appear in the same 2x2 box shape. Also how those boxes repeat in a diagonal line, and how boxes are connected vertically by a "strip" of 3 notes ADG

The only difference for guitar is that you have to correct for the G/B strings which are separated by a 3rd instead of a 4th, by scooting the pattern on the B+E strings up by one fret

Hey hn! I wanted to share my blog post. It's about the historical relationship between engineering and science, and an interpretation of the current AI movement. I would love perspectives.