John Hodge

ML, modeling & optimization for hardware and physical systems. Applied Scientist. Background in EM, antennas, & deep learning. Opinions my own.

[John Hodge @johnhodgeml]

@photonjaeger @RYANHINGSHING @PhotonCap @AcquiredFM I got you. I’m just naturally biased to think about the modulation scheme. I’m a big fan of the Aquired pod and the quote, btw! https://t.co/VmEsb2MmA0

[John Hodge @johnhodgeml]

@photonjaeger @RYANHINGSHING @PhotonCap @AcquiredFM I looked it up and see that this IM is intensity modulation. They are different, but not mutually exclusive.

[John Hodge @johnhodgeml]

@photonjaeger @RYANHINGSHING @PhotonCap @AcquiredFM What is IM-DD? A form of index modulation?

[John Hodge @johnhodgeml]

Can AI improve financial planning without inventing numbers?

Planner-lab keeps LLMs out of math: tested code calculates, a ledger traces results, and a critic blocks unsupported claims.

For CFPs and ML builders:
https://t.co/HnC6eLYnUP

[John Hodge @johnhodgeml]

https://t.co/m67ikchO3y

#Baseball #Sabermetrics #Python #Physics

[John Hodge @johnhodgeml]

Ever wondered why a curveball breaks or why a fastball appears to “rise”? Baseball

I built pitchphys, an open-source baseball pitch simulator that models gravity, drag, and the Magnus effect in your browser. Adjust velocity, spin rate, and spin axis and watch the physics unfold.

[John Hodge @johnhodgeml]

More info here: https://t.co/8c9V2wrUwn

[John Hodge @johnhodgeml]

I made Gumline, a private dental habit tracker for people who want to actually remember what their hygienist told them.

Log brushing, flossing, water-pick, focus areas, visits, and notes. No account, no cloud, no ads, no analytics.

https://t.co/LiI5UQpFSy

[John Hodge @johnhodgeml]

It’s easy to forget the little things when I’m tired, busy, or rushing around with my toddler.

So I made the app to help me stay on top of my dental routine.

[John Hodge @johnhodgeml]

New post: how I built the Connect Four-style AI behind Neon Drop.

Minimax, alpha-beta pruning, center-column heuristics, blunder rates, and why a strong game AI does not have to be a perfect solver.

https://t.co/b4N6s0XeOH

[John Hodge @johnhodgeml]

@omarsar0 Honestly, I thought LLM-as-a-judge was a joke when I first heard the term.

[John Hodge @johnhodgeml]

SlopScore also shows why the score happened.

This sample was driven mostly by lexical markers, genericity, and significance inflation, exactly what you’d expect from deliberately vague enterprise-speak.

Try it here: https://t.co/tu6gvWq5AI

[John Hodge @johnhodgeml]

Weekend project. Give it a try and let me know what you think! 📝

https://t.co/tu6gvWq5AI

#AIWriting #WritingTools #BuildInPublic #OpenSource #Python #SlopScore

[John Hodge @johnhodgeml]

The goal is to flag writing patterns that make prose feel generic, inflated, overconfident, or marketing-sloppy.

Here, it caught things like “unlock synergy,” “ever-evolving ecosystem,” “pivotal role,” and “indelible mark.”

[John Hodge @johnhodgeml]

@natolambert Interested.

[John Hodge @johnhodgeml]

@TheStalwart As LLM writing becomes normal, taste and QA become the bottleneck.

I built SlopScore to catch the lazy tells, canned transitions, faux profundity, template-y prose, before people or agents publish them.

[John Hodge @johnhodgeml]

I built slopscore: an open-source prose linter for AI-slop writing patterns as a personal hobby project.

It scans Markdown, text, JSON, websites, docs, and code comments, then returns a 0–100 SlopScore with evidence spans.

Blog post: https://t.co/zIuOkCop3P

[John Hodge @johnhodgeml]

I built it to make low-quality prose easier to see, discuss, and fix.

Try it out! I would appreciate your feedback!

Install:

pip install slopscore-lint

Repo: https://t.co/kMuhfKhcDH
Docs: https://t.co/5V23NQTTRi
PyPI: https://t.co/UCrbZtYlNg

[John Hodge @johnhodgeml]

I built slopscore: an open-source prose linter for AI-slop writing patterns as a personal hobby project.

It scans Markdown, text, JSON, websites, docs, and code comments, then returns a 0–100 SlopScore with evidence spans.

Blog post: https://t.co/zIuOkCop3P

[John Hodge @johnhodgeml]

The goal is practical editing.

A sentence can be weak even if a human wrote it. A paragraph can be useful even if AI helped draft it.

slopscore flags patterns like vague claims, generic filler, repetitive scaffolding, prompt residue, and low specificity.