Lightnews — Scholar-powered news

Jeremy Morrell

@jeremymorrell.dev

2.4K followers 600 following 5K posts

Principal engineer @ Cloudflare. I talk about tech sometimes Location: Squamish, BC 🇨🇦 Website: https://jeremymorrell.dev/

jeremymorrell.dev

Posts Media Videos Starter Packs

Pinned

Jeremy Morrell @jeremymorrell.dev · Oct 22

There have been some good posts on what "wide events" instrumentation is but not as many on how to go about it, what attributes you should add, or how to work with OpenTelemetry

I put everything I've learned in the last few years into one guide jeremymorrell.dev/blog/a-pract...

A Practitioner's Guide to Wide Events | Jeremy Morrell

The existing articles on Wide Events define the concept well but leave the implementation details to the reader.

jeremymorrell.dev

10 27 78

Jeremy Morrell @jeremymorrell.dev · 10h

I love frog and want good things for frog

lena @lenazun.bsky.social · 11h

i hope the frog has good opsec

Jeremy Morrell @jeremymorrell.dev · 10h

The 9 year old is learning about the dark side of sugar free gummies this evening. I assume he will be more diligent about reading packaging in the future when he raids the local corner store with his allowance

Jeremy Morrell @jeremymorrell.dev · 18h

This is good, and I learned things!

𝙃𝙤𝙪𝙨𝙚 𝙤𝙛 𝙇𝙚𝙖𝙫𝙚𝙨 Audiobook Narrator @jefferyharrell.bsky.social · 22h

I posted this last night cause I kind of wanted to bury it. I got cold feet about putting it out there.

embedding-space.github.io/sparse-netwo...

The subject is WHY neural networks work, and I think the answer I offer is kind of interesting. Maybe even a little correct, possibly.

A line chart titled “Accuracy vs. Sparsity (Iterative Magnitude Pruning)” showing model accuracy as weights are pruned. The x-axis represents sparsity from 0% to 100%, and the y-axis represents accuracy from 0% to 100%. A blue line with circular markers shows that accuracy stays around 80% from 0% to roughly 90% sparsity, then drops sharply toward 55% near 100% sparsity. A dashed red horizontal line labeled “80% target” runs across the chart near 80% accuracy, indicating the desired baseline.

Jeremy Morrell @jeremymorrell.dev · 1d

It’s still the best current option though as far as I can tell?

2 6

Jeremy Morrell @jeremymorrell.dev · 1d

What I want out of Twitter-a-likes is interesting technical discussion, slice-of-life stuff, experts from domains I don’t follow closely talking with their peers, and yeah, some news and experiencing it together

Bluesky lately… is not that. I feel more and more disconnected from the Discourse here

2 6

Jeremy Morrell @jeremymorrell.dev · 1d

This is good, and I wish I had had someone earlier in my career explain things like this www.seangoedecke.com/seeing-like-...

Seeing like a software company

The big idea of James C. Scott’s Seeing Like A State can be expressed in three points: Modern organizations exert control by maximising “legibility”: by…

www.seangoedecke.com

Jeremy Morrell @jeremymorrell.dev · 1d

I haven’t given it a try yet. What issues are you hitting? cc @micahw.com

Jeremy Morrell @jeremymorrell.dev · 2d

Tecate but pronounced like Hecate

Jeremy Morrell @jeremymorrell.dev · 2d

Oh, I don't care who makes it or what it's called. I doubt Honda will. I just want Utility Box EV, and I'm not sure if anyone wants to make one 😢

1 1

Jeremy Morrell @jeremymorrell.dev · 2d

but also smol

Screencap from a youtube walkthrough of the car. The boxy car only comes up to just below the man's shoulders, and takes up maybe 60% of a parking space.

Jeremy Morrell @jeremymorrell.dev · 2d

PLEASE someone make an modern EV Honda Element and release it in North America. I will line up to buy it

I miss my Toaster 😢

☠️deadward niedermurder☠️ @niedermeyer.online · 2d

Dacia made a boxy, low-cost EV that weighs less than 2,000 lbs, and just in case that wasn't enough to win me over they called it the "Hipster"

www.autocar.co.uk/car-news/new...

Dacia Hipster, a boxy, minimalist two-door, four-seat EV concept

3 1 8

Jeremy Morrell @jeremymorrell.dev · 2d

I imagine that's the big one, but we load JS bundles at the edge, and are also sensitive to bundle size since it affects cold-start time. If we're going to encourage library authors and users to include it, ideally it should at least tree-shake well if possible

Jeremy Morrell @jeremymorrell.dev · 2d

IIRC Vercel had their proposal a while back: github.com/vercel/otel/... but I'm not sure where discussions are today if anywhere. Bundle size is going to be a pretty big concern for us, and may be a place where we can contribute?

otel/packages/otelzero at main · vercel/otel

OTEL tracing for Vercel. Contribute to vercel/otel development by creating an account on GitHub.

github.com

1 1

Jeremy Morrell @jeremymorrell.dev · 2d

@aparker.io Is there any existing work / discussion on shrinking the bundle size of @opentelemetry/api? bundlephobia.com/package/@ope...

@opentelemetry/api v1.9.0 ❘ Bundlephobia

Find the size of javascript package @opentelemetry/api v1.9.0. Bundlephobia helps you find the performance impact of npm packages.

bundlephobia.com

1 4

Reposted by Jeremy Morrell

Danilo, Juicero Aspect 🇵🇷 @daniloc.xyz · 3d

The ~ AI skepticism ~ camp, so overcome with zealotry and in-group/out-group signaling

risks becoming a fully anti-intellectual movement

hostile to fact, new information, and any nuance that challenges whatever telephone game version of the truth they heard from the last viral post

Cabel Sasser @cabel.panic.com · 3d

i don’t like or understand ai. but it’s wild to see this person, who clearly DOES understand it and obviously knows a LOT about it, be repeatedly told they are stupid?

i think a lot of bluesky thinks the simple solution to discomfort is bullying

which is so weird given we were all surely bullied??

then: you don't need to "understand how the human mind works" on any level in order to make a computerized map of the connectome including every protein and cell. this is similar to how we do not currently understand how LLMs work but we can run them using opaque frozen weights.

bluesky: We do know how LLMs work. What the fuck are you talking about? Lmfao.

them: this is a common misconception about modern ai systems. they are grown, and not crafted; we understand the process that builds them but lack a paradigmatic understanding that answers the question
"how do these 1 trillion floating point numbers do ‹x>". this has been the case since the DL revolution

them: the way it actually works is that we initialize 1 trillion random floating point numbers as activation weights and biases in an enormous multilayer artificial neural network, give it tasks, and use vector calculus to update those 1 trillion neurons based on empirical task performance

source: i am literally one of the engineers/ research scientists who builds and optimize

them: and when i say "we update the neurons based on empirical task performance",
, the “we" is not any human at all, it is an automatic process called gradient descent

the job of these engineers is to make the framework that grows the neurons and feed the pipeline that executes, measures, and backpropagates task pertormance

geoffrey hinton won the nobel prize in physics in 2024 for his contributions that helped make this possible

bluesky: i think you maybe unwell

4 32

Jeremy Morrell @jeremymorrell.dev · 3d

The other thing that I will say is that if you go full background agents and don't have a high bar for what you will merge, you can add enormous amounts of technical debt to your codebase in a hurry

Jeremy Morrell @jeremymorrell.dev · 3d

Ultimately, my own time and attention is still a huge bottleneck. I don't see that changing any time soon.

1 2

Jeremy Morrell @jeremymorrell.dev · 3d

Like Simon I've also found some benefit in research and proof-of-concepts. "I want to accomplish X, what all needs to change?"

or a recent example I posted about where I needed to read a zip file from a worker: jeremymorrell.dev/sketches/clo...

Reading a zip file from a Cloudflare Worker | Jeremy Morrell

Experiments in reducing dependencies via LLMs

jeremymorrell.dev

1 2

Jeremy Morrell @jeremymorrell.dev · 3d

I've been trying the same to make progress on a side project to mixed results.

For things that play well with the existing architecture (extend this, add new method there, emulate this other implementation in a new class) this works pretty well. Changes that you can review in a few minutes.

Simon Willison @simonwillison.net · 3d

I've become one of those people who run multiple coding agents at once - here are my notes on embracing the parallel coding agent lifestyle simonwillison.net/2025/Oct/5/p...

Embracing the parallel coding agent lifestyle

For a while now I’ve been hearing from engineers who run multiple coding agents at once—firing up several Claude Code or Codex CLI instances at the same time, sometimes in …

simonwillison.net

1 2

Jeremy Morrell @jeremymorrell.dev · 3d

Nah, self-hosted DOs with dedicated hardware is actually very close to what I think I want here

(but also I would imagine a very hard thing to market to devs. Explaining the virtual actor model usually just results in a glazed expression in my experience 😅)

Jeremy Morrell @jeremymorrell.dev · 3d

No, this is just “Jeremy is a noob at data modeling” stuff 😅

Many other things I want to write about (and am procrastinating on 😬)

1 1

Jeremy Morrell @jeremymorrell.dev · 4d

Today I have learned many things about how not to do SQL migrations

1 9

Jeremy Morrell @jeremymorrell.dev · 5d

A more concise way to describe it might be “let me give Cloudflare money to run workers / DOs on dedicated hardware in locations I can choose” so I can get minimal latency between bits of my code and no cold starts but I trade off cost and potentially end-user latency

Jeremy Morrell @jeremymorrell.dev · 5d

I think it might be, though there are a couple of limitations of DOs that I’d want to work around in an ideal world:
- dedicated hardware vs serverless
- 10gb size limits
- latency between DOs

Perhaps all of these can be solved with DOs themselves? Facets is an interesting direction

Jeremy Morrell @jeremymorrell.dev · 5d

Wait, is this just Orleans? 🤔

Jeremy Morrell @jeremymorrell.dev · 20d

If I could design a compute platform from scratch today I think I'd explore vertical-scaling only, but make it really easy to adopt cell-based architecture with geographically distributed cells. SQLite is your storage layer.

2 4