Aaron Brooks

BTW, this is why I use a separate device for 2FA -- specifically a primary and backup YubiKey 5C. The primary is on my keychain, the backup is at home. I always register both.

Swapping phones is no longer a disruption to 2FA.

Isn't that stuff in the secure element that can't be copied (and intentionally so, for security reasons)?

New type of context engineering? When using chat-mode LLMs, I'm really wanting to "fork" conversations. I've created a useful context but don't want to pollute it with several different directions of exploration.

I want this as a feature.

Q: How many LLM agents does it take to change a lightbulb?

A: Either 1000 who will burn through through enough tokens to light the room directly or just one who is thoroughly confident that the lightbulb is already changed.

LLMs are the foie gras geese of data science

I am SO THOROUGHLY BORED of the AGI-is-next-week / AI-doesn't-think debate. The problem is, we don't have a good definition of what thought *is* -- because it is not just one thing. We'd get a lot farther if we start teasing out by what we mean by thought and all of its different facets

The lesson from the past 6 months is to do stuff in your communities with your neighbors and plug in where you can.

I want someone to create a Pokemon card game but for gut flora

"Oh, you have a Lactobacillus rhamnosus! That gives a +7 immuno regulation -- so cool!"

My watchgha lets you watch the progress of GitHub Actions in your terminal. Now the job names are sorted, since GitHub now returns them in a randomish order.

github.com/nedbat/watch...

😆. Particularly for FOSS projects, being able to do synthetic user usability testing across a range of user capabilities is pretty interesting. For many of these applications we're looking to maximize how smart the model is but this is a case where we might not want that

This, interestingly may be a use case for using "dumber" models in the future, even when more capable models exist, serving as a proxy for a lower range of user capability during this sort of synthetic user testing

Another Claude Code experience report. This is a followup to bsky.app/profile/kana.... Last time I tried to get Claude Code to create a new mal implementation in Fortran. It couldn't finish step1 due to memory management. This time I asked for a from scratch implementation using python3.

GitHub - kanaka/mal: mal - Make a Lisp mal - Make a Lisp. Contribute to kanaka/mal development by creating an account on GitHub.

Claude Code usage report.

Task: implement a fortran implementation of mal/make-a-lisp (github.com/kanaka/mal). Mal has implementations in 88 languages but none in fortran yet.

1/n

The Statue of Liberty from the grounds behind the monument, cast mostly in silhouette against an evening sky. Trees of the promenade bracket the photo.

Give me your tired, your poor,
your huddled masses yearning to breathe free,
... or give me $5,000,000 USD and we'll call it even

I love seeing Welsh in the chat

GitHub - Computational-Turbulence-Group/SURD Contribute to Computational-Turbulence-Group/SURD development by creating an account on GitHub.

Does anyone have any experience reports using SURD?
github.com/Computationa...
#SURD #causal_inference

bsky.app/profile/alxn...

I've found the CausalSky feed to be useful. That's how I saw your post :D

Want to use bluesky replies as your blog’s comment section?

`npm install bluesky-comments`

Built by @coryzue.com

One of the greatest values you can provide to an employer is saying No. To have experience to know what's important. If we don't log this thing we will have more retention of what matters MORE, longer. The real world has costs, and those have to be paid.

I learned a new term this year "exercise snacking." Not a replacement for regular workouts but a really good complement.

Welcome, folks! 😁

Since when are there more ollama supported models than cannabis strains?

Also, I'm not sure which are which... moondream, yi, zephyr, deepseek, nuextract, mathstral, notus, megadoplhin, codebooga, magicoder

I don't know who needs to know this but the 1992 Sneakers movie soundtrack is now on Spotify 🎉

A screenshot from a FedEx shipping status page showing "Shipment arriving On-Time" followed by "NAN" where the location should be.

That's confidence inspiring...

A screenshot of Twitter/X notifications, all of which are procedurally generated NFT images being sent as spam with lines like "Just minted!" or "Don't forget to mint!" Also, Twitter is begging me to turn on push for notifications in the midst of these.

This is all my Xitter notifications are now, NFT minting spam and Xitter asking to turn push notifications on. As if...

P.S. Xitter is pronounced "shitter"

To be clear, I'm not subtweeting anyone. This is preaching to myself.

Aaron's law (and maybe someone else's) of not being (so much of) an @hole:

The need for you to say That Thing is inversely proportional to how badly you feel like you need to say it.

Not saying it Right Now will allow you to find the right time & words and, maybe, that it doesn't need to be said.

"Please explain this gap in your resume"

Oh I'm sorry Mr. LaTeX formatting genius

Real-time collaboration made apps like Google Docs and Figma popular, and many other apps have also been adding similar features. However, the algorithms tha... "New algorithms for collaborative text editing" by Martin Kleppmann (Strange Loop 2023)

My talk at Strange Loop (which I had to give remotely) is now up! www.youtube.com/watch?v=Mr0a...

It covers our experimental text editor Upwelling www.inkandswitch.com/upwelling/ as well as the internals of Automerge and a bit about Peritext www.inkandswitch.com/peritext/