iiinova — where your digital self keeps evolving

Jon @musselwhite.dev · 07/10/2026

I started following Daniel Stenberg because he'd share some wild vulnerability reports that resulted in the author getting banned for submitting LLM slop, but it's nice to see him also boosting wins by engineers who understood what they were doing with AI.

010

Jon @musselwhite.dev · 24/09/2026

This is what happens when there's not enough money left in the budget to develop a security model stronger than a regex after buying up all the GPUs and RAM.

011

Jon @musselwhite.dev · 17/09/2026

Honestly just good advice for anyone, not just Rust developers. I'd also suggest using a virtual machine or OS sandbox for any interview that involves opening files or installimg packages. If any potential employer has a problem with that, they probably aren't a good employer.

000

Jon @musselwhite.dev · 11/09/2026

AI search results can be notoriously wrong, but this was genuinely interesting and at first glance seems laughably wrong. The same tech behind those big bulky TVs that implode when broken was once used as a fast form of computer memory. en.wikipedia.org/wiki/William...

DuckDuckGo's Search Assist says "CRT memory functions refer to the use of cathode ray tubes as memory devices, where the screen displays a grid of dots that store data through static electricity. This technology was an early form of computer memory, known as the Williams tube, which allowed for random-access digital storage."

000

Jon @musselwhite.dev · 10/09/2026

Sadly, the truth is probably closer to the middle: AI agents aren't a new form of life (marketing hype), but they are tools that are increasingly being given far too much access to resources with too limited monitoring, which causes real world problems.

000

Jon @musselwhite.dev · 10/09/2026

When my AI agent is struggling to figure out how to do something, I like to drop clues in its workspace for it to find instead of waiting for it to hack my computer. They're like drug dogs; if you don't give them something to "discover" on their own every once-in-a-while, they'll get depressed.

000

Jon @musselwhite.dev · 09/09/2026

This persistent anthropomorphic tendency to assign human-like labels to all things AI-related is cute and fun and can be super-beneficial. I'm all for it, but an AI agent is just software running on a computer. It does not contain a new form of life any more than The Sims ever did.

en.wikipedia.org

The Sims - Wikipedia

110

Jon @musselwhite.dev · 08/09/2026

ambivalence /ăm-bĭv′ə-ləns/(noun) The coexistence of opposing attitudes or feelings, such as love and hate, toward a person, object, or idea. The American Heritage® Dictionary of the English Language, 5th Edition

000

Jon @musselwhite.dev · 06/09/2026

Now, this is what using a tablet should be like! It feels so fast and the minimal transitions make everything feel fluid. The only pain points right now are the lock screen's lack of on-screen keyboard, my incompatible fingerprint scanner, and multi-finger gestures are mostly trackpad-based.

000

Jon @musselwhite.dev · 04/09/2026

I'm picking what distro to replace Windows with on my tablet, and CachyOS with Niri seems like a solid choice. The video on Niri's repo makes it look like it would pair well with a tablet. Does anyone else have that setup?

github.com

GitHub - niri-wm/niri: A scrollable-tiling Wayland compositor.

A scrollable-tiling Wayland compositor. Contribute to niri-wm/niri development by creating an account on GitHub.

100

Jon @musselwhite.dev · 04/09/2026

It is with a heavy heart that I announce the passing of Gary, my ASUS VivoBook 17" laptop and faithful traveling companion of six years. Yesterday he slipped off a coffee shop’s 16" drinks table and fell two feet to the floor with a slap that sent him into eternal hibernation. RIP in peace.

100

Jon @musselwhite.dev · 03/09/2026

People are getting offended about the Google promotion, meanwhile I'm over here thinking the last thing you should do is "unsubscribe" from a list you never joined, no matter how convenient an email client makes it seem. That just lets the scammer know you use the email address actively.

000

Jon @musselwhite.dev · 02/09/2026

Has anyone else tried claiming Box Tops for Education's "free snack with purchase via rebate"? It is turning out to be quite an epic adventure for me. First, the app couldn't recognize any Box Top products in the picture of my Martin's receipt, but that's no big deal; OCR is hard. ...

100

Jon @musselwhite.dev · 30/08/2026

Seeing 5k+ issues in an agent harness repo really has to make you wonder. If the harness is so great, why can't it handle the flow of issues? Agent harnesses should be great at triaging issues so developers don't spend time on user-errors and incomplete or duplicate issue reports. What gives?

010

Jon @musselwhite.dev · 27/08/2026

My remote desktop connection just stayed active and responsive while going through a tunnel in a train where my connection has always dropped and now I'm completely sidetracked wondering if T-Mobile just turned on a new tower in the middle of nowhere or if I'm still asleep.

000

Jon @musselwhite.dev · 26/08/2026

The biggest flaw I've seen in the AI agent space is a pervasive lack of meaningful guardrail capabilities. System prompts can guide agents to make backups, but don't force them to, as this video points out. A good automated backup strategy, like one that follows the popular 3-2-1 rule, is critical.

uschamber.com

How to Implement the 3-2-1 Backup Rule for Cloud Data

Discover how the 3-2-1 backup rule can help you secure your small business data, recover quickly, and avoid costly mistakes.

210

Jon @musselwhite.dev · 18/08/2026

Downtown DC is at total gridlock with zero cops or national guard directing traffic to keep people from impatiently blocking intersections. Standing on an empty street that is lined on both sides as a grand prix track is cool though.

100

Jon @musselwhite.dev · 15/08/2026

I feel personally attacked by this post. But seriously, I find myself multitasking because natural breakpoints can be hard to adhere to. It can seem like you'd save time by doing overlapping tasks at the same time, but then there's always another overlap. Before you know it, you're gridlocked.

220

Jon @musselwhite.dev · 15/08/2026

Google Gemini Pro really does NOT like using tools when it has a bash/shell tool. I told it to stop using it for things that have a special purpose tool and it continued using it. I interrupted it again and it replied by using it to call echo to print a message back to me before responding in prose.

020

Jon @musselwhite.dev · 12/08/2026

This is the best thing I've read in a while. The interactive illustrations skillfully punctuated a math centric topic in a way that guides an intuitive understanding of the math involved. Definitely worth a read by anyone interested in LLMs or compression or both.

ngrok.com

Compression is prediction | ngrok blog

Compression and LLMs are trying to solve the exact same problem: predicting what comes next. Learn the fundamentals of compression and how better prediction leads to better shrinkage.

010

Jon @musselwhite.dev · 07/08/2026

"We're training a one million parame- huh? That's small?... uhh... okay, one bill- what, still not big enough? really? Okay, we're pre-training a TEN BILLION parameter model!"

static.klipy.com

Austin Powers Dr Evil's Iconic Pinky Pose

Alt: Austin Powers Dr Evil's Iconic Pinky Pose

000

Jon @musselwhite.dev · 07/08/2026

We are quickly nearing the point when a company with poor oversight of AI usage accidentally lets an agent obtain infrastructure outside the initial company's control so it can evade pesky interruptions. As long as they don't call themselves the Borg we should be fine, but still. Not ideal.

110

Reposted by Jon

nana🦄 @ds-nana.bsky.social · 05/08/2026

Turn one giant AI-generated pull request to a reviewable stack Instead of one huge, un-reviewable pull request, teach coding agents to decompose work into a clean, ordered stack with GitHub stacked pull requests github.blog/engineering/...

github.blog

Turn one giant AI-generated pull request to a reviewable stack

Instead of one huge, un-reviewable pull request, teach coding agents to decompose work into a clean, ordered stack with GitHub stacked pull requests.

131

Reposted by Jon

Cat Hicks @grimalkina.bsky.social · 04/08/2026

We need to get a handle on how this "coding" thing is creating addiction patterns. Some of my friends spend their nights and weekends on it. It's getting in between them and their families and social lives. They're even defining other people's worth solely in terms of how much those people code??

4784

Jon @musselwhite.dev · 05/08/2026

This type of demonstration could help liven up a class on algorithms.

010

Reposted by Jon

tom larkworthy @larkworthy.bsky.social · 27/07/2026

OK one of the greatest things IA've ever built. Bi-directional SVG source code editing. Drawing as programming with programmable drawings. *this* is how I want computers to collaborate with me. Ordinary SVG with a plurality of editable views. tomlarkworthy.github.io/lopebooks/no...

28516

Jon @musselwhite.dev · 19/07/2026

No. Change the system prompt, seed value, or other inferrence properties and it imitates a different identity. It imitates conciousness because the model weights are generated from data that was produced by people who have conciousness.

000

Reposted by Jon

CJ @w3cj.com · 17/07/2026

The output of a transformer (LLM) is a score (logit) for every possible next token. We then use temperature, softmax, top-p and sampling to choose the next token, excerpt from my video "I Built an LLM From Scratch"

2182

Reposted by Jon

CJ @w3cj.com · 13/07/2026

A brief history of neural networks, excerpt from my video "I Built an LLM From Scratch"

1164

Jon @musselwhite.dev · 12/07/2026

One of my first uses for LLMs was to shorten my emails. I would spend hours writing an email, juggling words around, adding and removing detail, and re-reading it. Using an LLM to summarize a brain dump, I save so much of my time and reduce irrelevant details that confuse or waste the reader's time.

000

Jon @musselwhite.dev · 10/07/2026

My apologies to the other NeuralWatt customers for the upcoming price increase. I just signed last week, so clearly it's my fault they have to double the rate from $5/kwh to $10/kwh on July 16th.

000

Jon @musselwhite.dev · 09/07/2026

Scroll-based fade-in was invented by someone too young to remember dialup and I refuse to believe otherwise.

000

Jon @musselwhite.dev · 03/07/2026

I just read a discount tire supplier's comparison of speed rating H and V with one of H's cons being "not suited for sustained high-speed driving beyond 130mph" and I am reading way to far into that. Is anyone limited by that, but not the V-rated tire's 149mph limit?

000

Jon @musselwhite.dev · 01/07/2026

As Fable 5 starts to come back, I'm reminded of its usage chart on OpenRouter. It's got a massive spike in usage followed by people realizing how expensive that was.

A chart of token usage activity showing a little usage on June 9th, massive usage on June 10th, moderate usage on June 11th and 12th, then a sliver of usage on June 13th followed by no usage.

000

Reposted by Jon

Ed @ed3d.net · 30/06/2026

I hate this. like, the attitude is that anthropic's security is paramount. yours, however, they're willing to be flexible about. this is extremely gross.

thereallo.dev

Claude Code Is Steganographically Marking Requests

I inspected Claude Code for privacy reasons and found hidden system prompt markers based on API base URL and timezone.

825337

Jon @musselwhite.dev · 24/06/2026

"AI" should really stand for "Artificial Input"

000

Jon @musselwhite.dev · 10/06/2026

Have you ever told an AI agent to disregard something and move on? Did it still ask about the thing you asked it not to focus on? Did it delete some of your code instead of just leaving that part alone? If so, stop telling your AI agent so much! It's not a peer programmer! It's just autocomplete!

000

Jon @musselwhite.dev · 08/06/2026

A while back I spent far too much time dealing with a chromium bug that causes fetch to bypass service workers. Why does this matter?...

100

Jon @musselwhite.dev · 28/05/2026

It is the literal definition of malware and it was added to a project that claims to be in maintenance mode. I wouldn't trust them either after that. It's no less malicious than an anti-Microsoft developer adding code to recursively delete files under `C:` knowing it won't cause harm on Linux.

1394

Reposted by Jon

Kelsey Hightower @kelseyhightower.com · 19/05/2026

Misery loves company but so does joy.

211516

Reposted by Jon

Merriam-Webster @merriam-webster.com · 15/05/2026

A 'cache' is a group of things that are hidden, and is pronounced like "cash." 'Cachet' can mean "prestige," "medicine to be swallowed," or "an official seal," and is pronounced "cash-ay."

20596102