Wednesday, October 7, 2026ArchiveSearchAsk the paper

The Computomatix Times

All the posts fit to save — curated from @computomatix's bookmarks & likes on X

Edition of Saturday, April 4, 2026

7 stories

Andrej Karpathy Shares Idea File for Building LLM Knowledge Bases

Andrej Karpathy publishes a gist describing an 'idea file' approach, where an agent builds a personal LLM wiki from shared concepts rather than shared code. The post builds on his earlier thread on using LLMs to compile markdown knowledge bases from raw sources.

Original post · 1 min read
Wow, this tweet went very viral!

I wanted share a possibly slightly improved version of the tweet in an "idea file". The idea of the idea file is that in this era of LLM agents, there is less of a point/need of sharing the specific code/app, you just share the idea, then the other person's agent customizes & builds it for your specific needs.

So here's the idea in a gist format: gist.github.com/karpathy/442a6bf555914893e9891…

You can give this to your agent and it can build you your own LLM wiki and guide you on how to use it etc. It's intentionally kept a little bit abstract/vague because there are so many directions to take this in. And ofc, people can adjust the idea or contribute their own in the Discussion which is cool.
Andrej Karpathy @karpathy
LLM Knowledge Bases

Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:

Data ingest:
I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki…
♥ 26.8K · ⟲ 2.8K · 👁 7.3MView on X ↗

Addy Osmani Releases Open-Source Agent Skills for Coding Agents

Addy Osmani Releases Open-Source Agent Skills for Coding Agents

Addy Osmani of Google released Agent Skills, 19 engineering skills and 7 slash commands that enforce specs, tests, and reviews for coding agents including Claude Code and Cursor. The project is free, open source, and installable via npx.

Original post · 1 min read
🚨 You need to see this.

@addyosmani from Google just dropped his new Agent Skills and it's incredible.

It brings 19 engineering skills + 7 commands to AI coding agents, all inspired by Google best practices 🤯

AI coding agents are powerful, but left alone, they take shortcuts.

They skip specs, tests, and security reviews, optimizing for "done" over "correct." Addy built this to fix that.

Each skill encodes the workflows and quality gates that senior engineers actually use: spec before code, test before merge, measure before optimize.

The full lifecycle is covered:

→ Define - refine ideas, write specs before a single line of code
→ Plan - decompose into small, verifiable tasks
→ Build - incremental implementation, context engineering, clean API design
→ Verify - TDD, browser testing with DevTools, systematic debugging
→ Review - code quality, security hardening, performance optimization
→ Ship - git workflow, CI/CD, ADRs, pre-launch checklists

Features 7 slash commands: (/spec, /plan, /build, /test, /review, /code-simplify, /ship) that map to this lifecycle.

It works with:
✦ Claude Code
✦ Cursor
✦ Antigravity
✦ ... and any agent accepting Markdown. Baking in Google-tier engineering culture (Shift Left, Chesterton's Fence, Hyrum's Law) directly into your agent's step-by-step workflow!

`npx skills add addyosmani/agent-skills`

Free and open-source.

Repo link in 🧵↓
♥ 2.5K · ⟲ 339 · 👁 423.0KView on X ↗
AI7/10

Google Releases Open Source App to Run Gemma 4 Offline on Phones

Google Releases Open Source App to Run Gemma 4 Offline on Phones▶

Paul Couvert highlights an official Google app that runs Gemma 4 models fully offline on iOS and Android. The app supports text, audio and image input through the E4B and E2B variants.

Original post · 1 min read
Friendly reminder that Google has an official app to run Gemma 4 on your phone.

- 100% open source
- Fully offline and private
- Multimodal with text/audio/image
- Works with Gemma E4B and E2B

And the app is available on both iOS and Android.

Steps and download below
♥ 5.3K · ⟲ 572 · 👁 727.1KView on X ↗
AI7/10

Google Publishes Visual Guide to Gemma 4 Architectures

Google Publishes Visual Guide to Gemma 4 Architectures

Google Gemma shares a visual guide explaining the new Gemma 4 architectures and how they process text, images, and audio in the smaller models.

Original post · 1 min read
Who wants to know how Gemma 4 works?

This visual guide breaks down the new architectures and how they process text, images, and (for the smaller models) audio.

👇
♥ 4.2K · ⟲ 480 · 👁 246.9KView on X ↗

Andrew Farah Releases Fieldtheory CLI to Sync X Bookmarks Locally

Andrew Farah Releases Fieldtheory CLI to Sync X Bookmarks Locally▶

Andrew Farah shares his first open source project, a free CLI called fieldtheory that downloads and syncs X bookmarks locally so an agent can access them. The post shows install and sync commands plus viz and classify features.

Original post · 1 min read
sharing my first open source project

a CLI for downloading and syncing your X bookmarks locally so your agent can access them. it's free

› npm install -g fieldtheory
› login to your X account in a chrome tab
› ft sync (done!)

bonus:
› ft viz
› ft classify
♥ 4.3K · ⟲ 271 · 👁 556.4KView on X ↗

Ruben Recommends Superpowers Brainstorming Skill for Creative Work

brainstorming — obra/superpowers

Ruben replies to Paul Solt that he uses the brainstorming skill from the obra/superpowers repository on skills.sh. The skill is meant to be used before creative work such as building features or modifying behavior, to explore user intent.

Original post · 1 min read
@PaulSolt I use brainstorming all the time

skills.sh/obra/superpowers/brainstorming
skills.shbrainstorming — obra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent,…
♥ 119 · ⟲ 4 · 👁 34.1KView on X ↗

Analysis Traces Rohit Sharma's Late-Career Batting Evolution

The Man Who Rewrote His Own Obituary

A cricket analyst examines how Rohit Sharma reworked his batting between 2021-23 and 2024-26 after bowlers began exploiting his weaknesses, citing strike rates, Runs Above Average and pitch maps.

Original post · 6 min read
X ArticleThe Man Who Rewrote His Own Obituary
We've seen that most batters tend to hit a saturation or a dip around the age of 35. And honestly, that's not surprising as it happens with many cricketers. Some choose to retire because the game catches up with them, and some just settle into whatever's left.
As a Rohit Sharma fan for a long time, it was disappointing to see that there was always this expectation, created widely throughout the world, that the same thing would happen with him. And that expectation wasn't wrong either, because the problems were there. He was struggling in the powerplay, with left arm seamers, against slow left arm bowlers. After playing 200 T20s, there is also enough data, enough evidence in your technique and your methodology, for bowlers to finally start figuring you out really well.
And most of the time, players don't really evolve much at that point in their careers. Bowlers know what strategies work against them, the weaknesses become well established.
But what Rohit did was genuinely different.
He looked at that reality and found a way to flip the script in a way that very few batters have ever managed. What happened to Rohit between 2021-23 and 2024-26 is, for me, one of the most remarkable technical and psychological evolutions you'll ever see from a batter at the back end of a legendary career. And the numbers make it very clear.
The Obituary That Was Being Written
Think back to the years 2021-23, Rohit was still a formidable force on the field but behind the scenes, a strategy was developing in dressing rooms across the globe.
Using your slow left-arm orthodox bowlers attack him from around the wicket, keep it on the stumps, and deliver those good-length balls. Guess what? It actually worked.
During that period, Rohit faced left-arm fast bowlers in the powerplay with a strike rate of 103.02, but his Runs Above Average (RAA) was a concerning -7.85. That negative figure is crucial as it shows he was underperforming compared to the average batter against those bowlers. The plan was clearly effective.
Now, when it came to slow left-arm orthodox bowlers, he managed to score 95 runs off 77 balls, with a strike rate of 123. There were control issues, dot balls were piling up, and the bowlers had a clear idea of where to target him. The pitch maps from that time tell the whole story. Good length, right on the stumps. Four dismissals in that one area. Bowlers were lining up to exploit that corridor. When you can dismiss one of the best powerplay batters in the world by repeatedly bowling the same delivery, you keep doing that, right?

The Decision Most Champions Never Make
Most elite batters, when they sense vulnerability, retreat into their strengths. They tend to leave the problem areas alone and hope nobody exploits them.
Interestingly, Rohit made the difficult choice of outsmarting the bowlers, which would later prove to be crucial in India's World cup win in 2024.
The centrepiece of that rebuild? The sweep shot.
Against spin bowlers in 2021-23, Rohit’s sweep was a minor part (18% of runs) of his arsenal. By 2024-26, it accounts for 28.19% of all his runs against spin in the powerplay. That’s a 10% jump. But the biggest revamp was the control percentage: 83.33%, which was around 65% before.
You do not play an attacking shot that contributes nearly a third of your runs against quality spin bowling with 83% control by accident. That takes hours in the nets, tweaking the trigger movement, adjusting the head position, recalibrating the risk-reward every single time.
The SLA Problem and how Rohit solved it
Let’s zoom in on slow left-arm orthodox bowlers, because this is where the story gets genuinely interesting and worth studying.

Against slow left arm bowlers alone, the sweep shot has given him a huge boost. His SR of 167 is 44 points higher than an avg batter against these kinds of bowlers.

8.2% runs in the square leg region in 2021-23 and that number has skyrocketed to 28% of runs in 2024-26. His reverse sweep and late cuts using the pace of the ball also have increased the runs he scores against these bowlers in third man region.
Now add this: that good-length delivery on the stumps that used to be his weakness? The one where four dismissals were logged in that single grid?
Rohit against spin when the ball was bowled on a good length on stump line: He used to play 15% of those balls in the square leg region in 2021-23. During 2024-26 this number has risen to 21.3% which is 8.3% more than an average Right-handed batter against spin on that very line and length.

The 2 key variations of left arm orthodox are both taken to the cleaners now by him. His performance against those 2 types of deliveries has risen significantly.

In 2021-23, when Rohit faced left-arm spin deliveries, he scored 106 runs off 100 balls. Strike rate of 106. Dot ball percentage of 40%. Balls per boundary of 8.33.
By 2024-26: 70 runs off 38 balls. Strike rate of 184.21. Dot ball percentage collapsed to 15.79%. Balls per boundary down to 3.… continue on X ↗
♥ 773 · ⟲ 279 · 👁 296.2KView on X ↗