Alex Martin

Who are the absolute best vibecoders/shippers on 𝕏? I want to hire you: • I don't care if you never went to an Ivy league • I don't care if you never worked at Blackrock or Bain • I don't care if you never learned to code I care that you SHIP, I care that you BUILD My DMs are open!

@claudeai I JUST SPENT THE LAST 12 HOURS COMPLETELY RUNNING MY FABLE 5 LIMIT TO ZERO NOOO

We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.

@mathemagic1an Well that's completely insane, do you think it's feasible from a cost perspective?

What if we could cancel out urban noise pollution? Place a set of speakers that act like noise-cancelling airpods for all pedestrians simultaneously You can actually do this - and it paves the way for an "acoustics foundation model" Simulation code + proposal 👇

@DxnkLabs I'd use Opus 4.8 on high reasoning, it's quite token efficient and Opus works really well with it

Heres the github! Just send this to claude code and it will do the rest. github.com/amart-builder/…

I built a Claude Code plugin that takes you from zero to master vibecoder in 3 days And I just open sourced it It runs on a technology called "AI Adaptive Learning" and it works insanely well It's the same technology used by Alpha school, a k-12 in Texas where kids go to school just 2 hours a day and have 98th percentile test scores So I adapted it to take a total beginner all the way to vibecoding the same way that Boris Cherny does (founder of Claude Code) Here's how it works: 1. It starts with a diagnostic exam. Not a quiz you click through. A real adaptive exam. Miss a question and it drops easier. Nail one and it climbs. In minutes it knows exactly what you know. 2. It builds a curriculum just for you. Only the units you're missing, in the right order. Already know HTML? Skipped. Total beginner? You get the full path. That's why it's fast: you never sit through a lesson you don't need. 3. You can't fake your way through. Every unit ends with a mastery check. Pass and the next units unlock. Fail and it finds your exact misconception, reteaches that one thing, and tests you again. There is no moving on at 70%. 4. It never lets you forget. Everything you master goes into spaced review, and it brings each concept back right before you'd lose it. Same science as flashcard apps, except it writes the reviews for you. 5. It's a game. XP, ranks, streaks, badges. You climb from total beginner to a top rank you can only earn by shipping real work that an AI evaluator signs off on. The whole thing lives inside Claude Code. Your tutor is the same AI you build with, so every lesson ends with you actually building something. The plugin is free. Link in the first reply. Comment below to help push this to more builders! Maybe tag someone you think benefit from this.

@Zephyr_hg a lot of my clients’ employees are scared i’m going to automate them out of a job it’s pretty tough convincing them i’m actually enabling them to pull in 2x sometimes 10x the revenue for their company Fearmongering has had quite the impact

Sam Altman at TED: "You can say, oh man, it's doing everything I do. What's going to happen to me? Or you can say... okay, now there's this new tool." In a 47-minute conversation he pushes the choice further than the clip everyone shares. Both views are now measurable. One group's rate holds flat. The tool-takers keep landing on the winning end of every salary scan this year. The second view turns out to be five learnable skills. Watch the talk, then read the five below.

@morganlinton so odd that Fable 5 is worse on medium than low and high any thoughts on why? benchmark test issue or a weak spot in the model?

Woke up to see the results of my sweeping effort level benchmark comparing Grok 4.5, GPT 5.6, and Fable 5. This benchmark now looks at my eval suite across low, medium, and high effort. My eval suite has no puzzles, and no math problem, and no meaningless, "can it one shot this!" tests. It's all real engineering problems that represent work engineering teams like mine would throw at these models on a regular day. And I think this proves two things to me: 1. Grok 4.5 is a very good model, it is likely going to become my go-to as a core model in the execution layer 2. Moving from Medium to High effort doesn't increase accuracy. This is flawed thinking from the previous generation of models. With this new generation of models, Medium is very likely the sweet spot. I will be sharing more details of these results later today or tomorrow. But now I want to go for a trail run! 🥾🌲 Might be back on my computer today, or might just spend the whole day adventuring in Tahoe, we'll see.

I saw this went crazy viral! So figured I'd build it: I turned .@businessbarista's list into a personal AI tutor. It's a Claude Code plugin. It starts with a short quiz, skips what you already know, then teaches everything on the list one idea at a time: 3Blue1Brown, Karpathy's Zero to Hero, Dwarkesh's chalkboard lectures, @_raghavdixit_ vectors series. I built this same engine to teach myself to code and it worked insanely well. The rule: you can't move forward until you can explain the current idea in your own words. No watching a 2 hour video and pretending. By the end you can explain exactly how LLMs work, start to finish. Made for people who have never written a line of code. There's also a builder mode where you code a mini GPT yourself! Free and open source. Link in the first comment.

So apparently GPT 5.6 Sol might just randomly delete all of the files on your laptop 🚨 Here's how to safeguard yourself against it Paste this prompt into codex: Protect this machine from accidental permanent file deletion by Codex. Implement this as defense in depth, not merely as a written instruction. Requirements: 1. Inspect the current official Codex documentation for hooks, command rules, sandboxing, and approval settings before making changes. 2. Detect the operating system and identify its recoverable Trash or Recycle Bin command: - On macOS, use `/usr/bin/trash `. Do not add `--`, because macOS `/usr/bin/trash` treats it as a filename. - On Linux or Windows, verify an installed, reliable Trash or Recycle Bin mechanism before using it. - If no reliable recoverable mechanism exists, stop and explain what must be installed or configured. 3. Add a durable global Codex instruction: - Codex must never permanently delete a file or directory. - All removals must go to Trash or the Recycle Bin. - Permanent deletion requires the user to explicitly change this safety policy first. - Destructive Git operations such as `git clean`, `git reset --hard`, bulk checkout, or bulk restore must also be blocked. 4. Install a global `PreToolUse` hook that returns a hard `deny` before execution when it detects: - `rm`, `unlink`, `rmdir`, or `shred` - `find -delete` - `rsync --delete` - destructive Git cleanup or reset commands - common Python, Node, Ruby, or Perl deletion APIs - file deletion through `apply_patch` - delete-like MCP filesystem tools - nested destructive commands inside `bash -c`, `bash -lc`, `sh -c`, or `zsh -c` 5. The hook must direct Codex to the verified Trash or Recycle Bin command instead. Do not silently rewrite dangerous commands, because flags and shell expansion could be misinterpreted. 6. Add user-level Codex command rules that mark direct permanent-deletion commands as `forbidden`. Preserve existing rules and configuration. 7. Set safe global defaults when compatible with the existing configuration: - `sandbox_mode = "workspace-write"` - `approval_policy = "on-request"` Do not override managed policies or create conflicting permission configurations. 8. Preserve all existing configuration. Inspect files before editing, make narrowly scoped changes, and create recoverable backups of any configuration files that must be replaced. 9. Validate everything without risking real data: - Parse the hook and configuration files. - Feed simulated tool-call JSON into the hook. - Confirm dangerous examples are denied. - Confirm ordinary commands and the Trash command are allowed. - Test the Trash mechanism using a newly created disposable temporary file. - Run an end-to-end Codex test using `rm` with no arguments inside a read-only sandbox. Never test `rm` with a real target. - Confirm the command was blocked before shell execution. 10. Explain any limitations honestly. A Codex hook is a guardrail, not an absolute operating-system security boundary. 11. Check whether system backups are configured. Do not enable or modify backups without permission, but report clearly if no backup destination exists. 12. At completion, provide: - The files created or changed - The validation results - The exact one-time steps required to review and trust the hook - Any restart requirement - Any remaining risks Do not permanently delete anything while completing this task. If cleanup is necessary, move it to Trash or the Recycle Bin.

@brian_armstrong @bchesky @jgebbia @nathanblec Net worth staying at BNB vs starting Coinbase? 😂 might be a 10 fig diff

Had a great time at Airbnb and learnt a ton from @bchesky, @jgebbia, and @nathanblec. Most of the time this advice is correct. But when I read the Bitcoin whitepaper I realized there was a huge opportunity, and the world needed an updated financial system. Couldn't wait.

@morganlinton It has to, unless their servers are melting, or if opus 5 is coming out tmr

Prediction, Fable stays in subs, announced tomorrow.

@morganlinton this is making me change my mind to just use medium also feels like choosing effort level should go away asap, model should just choose based on the task

I firmly believe that with this new class of models, most engineering teams will use models with Medium effort the most. We used to live in a world where models weren’t as capable unless we used them with High or xhigh effort. But with the new class of models we have from AI labs, that has changed. So if you’ve had a habit of, or assumption that, to get good code out of models you need to use High effort, it’s time to rewire your brain. Everything changed in the last month.

@yasu0x1 yeah he was fantastic, actually internally i tell people i work with never to edit things that AI write for them as long as the ideas are right just wasted time for preference

@LoopOnChain hiring managers who flag AI writing are filtering for the wrong thing did the kid know how to use the tool or not?

Hot take: I do not care if your writing “sounds” AI Interviewed a kid the other day and this was his email, a huge em dash smack dab right in the middle no doubt email was written by AI …who cares? Why would I ever shame someone for using the best technology in the world?

@itsolelehmann What would you classify as the “hardest problems?” I always struggle to find the right balance with effort level

@morganlinton this is so far from what i expected 😂 Uhhh is grok 4.5 great then? I have used it with my 𝕏 premium to research twitter but not to build anything

@Hiraweb3 meaning Fable will follow your instructions even if it thinks they are the incorrect instructions you have to be very precise about your prompts

@LoopOnChain what happens when the process is broken but the output looks fine?

I asked Fable how it wants to be prompted. Sharpest line, pasted exact: "Given process, I'll comply, even when the process is what's wrong." That is the whole argument in the model's own voice.

@WidmerLuiz yeah this is a great idea, having that backup is so important

@LoopOnChain I run everything in a git repo for safety and version control. Even so the one that used to make these kinds of mistakes was 5.3 codex which took some commands very literally.

@eeleeyaa oh wow, very interesting. yeah maybe this is a harness issue seems like the team is working on it, they said they’ve never seen anything like this before

@LoopOnChain Its a codex problem not a model problem. This was with 5.5 , two days ago