sharing ai learnings when using it for my brand: https://xitter.cf/t.co/7Hna29m0Xt 👇 see how i use ai (open-source)
Joined March 2010
- Tweets2.8K
- Following400
- Followers1.1K
- Likes3.8K
Pinned Tweet
here's how i automated my operations manager job
this was originally was only for the $25k @Cloudflare github competition but i wanted to share it here for folks interested in me rambling about ai!
this is like the second amendment... but with ai!
Today we're releasing abliterated-model-large-v2.
Based on GLM-5.3, which is #3 on Terminal-Bench 4.0 (behind only Opus 5 and Fable), with 2× the cyber exploitation of 5.2.
We abliterated and hosted it so it does the offensive cyber, red teaming, and agent testing work other models refuse to do.
- US-hosted
- FP8
- 1 million context window
- Zero input/output prompt retention
Live now. 🧵
Steal this memory prompt so your ai can remember things across sessions (just paste this into your ai of choice):
> 4 stores, 1 job each: plans, specs, memory, wiki
> Plans: one folder per change with proposal, design, and tasks
> Specs: what shipped, frozen once the plan is archived
> Memory: short facts from each session, kept outside git
> Wiki: lasting how-tos, curated in the repo
> No fact lives in two places
> Each fact is one line
> Fact types: user, feedback, decision, reference, open question
> Each fact is tagged with change name, folder, author, date, and session
> Facts are never edited, only replaced by a newer one
> Replaced facts stay searchable
> Transcripts are kept with secrets scrubbed, so every fact traces back
> Personal facts stay private, decisions are shared with the team
> Session start: a hook loads a short digest of open questions, preferences, and recent decisions
> First edit in a folder: a hook shows that folder's warnings
> One map file links folders to tags
> Before big work, one search runs across facts, wiki, and specs
> Results cite the originals, and Claude reads them
> Each step (plan, build, ship) loads its own open questions
> Checkpoints write facts from the session
> A background run catches sessions you never saved
> Every write shows near-duplicates: add, replace, or drop
> Open questions must name who checks them next
> Struggles get logged and feed a self-improvement routine
> Unanswered questions auto-close after 30 days
> Facts that name a file get that folder's tag
> A regular pass merges duplicates and settles contradictions
> Lasting facts move to the wiki, and stale ones retire
> Planning reads memory and wiki before asking you anything
> Building loads facts for every file the plan touches
> Shipping archives the plan as a spec and saves decisions
> Closing a chat writes what it learned to the wiki
> Memory is a hint: when it disagrees with the code, the code wins
> Search by keyword first, and add embeddings only if it misses
> A test set of questions in CI keeps search from breaking
Or get the full source from: github.com/matthewwong525/Wo…
For those of you trying to incorporate AI into your business.
There’s no single magic AI app or workflow that will magically make your business AI native (yet).
But there’s a simple thing you can do:
Just do everything within your ai chat bot. No external ui allowed.
It’ll feel SUPER SLOW at the beginning and you’ll likely need to switch up some of your current infrastructure (e.g. subscriptions)
But once you get past that hurdle, you’ll absolutely feel the acceleration. It’s frickin crazy
Also, i want to mention, i open-sourced my entire workflow. If you’re curious check it out here: wongstack.com
Just told my employee that our goal for the end of the year is to hit RSI.
A couple months ago, i would have never believed that this was possible.
Matthew Wong retweeted
Introducing SKY-001, the first Engineering World Model.
On CAD and BIM tasks, it beats GPT-6 Astra and every frontier model we tested:
-> 51-125x faster
- > 150-210x cheaper
-> Higher IOU than all frontier models
SKY-001 lets you dream a physical object before you build it.
18 months of work. Built by the team out of Maluuba's deep learning lab (acq. by @MSRResearch).
The first ‘AI General Engineer’ will be the backbone for Autonomous Manufacturing.
so many companies are trying to get agents into common platforms (e.g. slack/imessage) we use.
But have we ever stopped to consier:
is slack/imessage the optimal UI to communicate with ai agents?
Matthew Wong retweeted
SI labs will never solve alignment.
they’re trying to solve a philosophical and theological problem as an engineering one.
they’ve assembled some of the greatest engineers on the planet, but put socially dysfunctional nerds with an incredibly shallow understanding of human nature in charge of defining what superintelligence should value.
humanity has spent thousands of years trying to align humans and always failed, we can’t even agree on what is good and now we expect a handful of engineers to solve the same problem for superintelligence? lmao, the hubris
you can’t align something with humanity if you don’t understand humanity in the first place.
at this point, you’re probably better off developing ASI and asking it how to solve alignment.
Matthew Wong retweeted
Turn any existing Agent into the world's best personal shopper for you.
The new shop skill: shop.md/
Add the skill & connect your Shop account & finally find the thing that you always wanted
Matthew Wong retweeted
Hivecoding is the most powerful use of Ai that i have ever used...
25 opus 5.5 agents + top chat messages = hivecoding
Everyday i live stream and 100s of agentic engineers type fixes and improvements in my chat.
most people in my chat are real devs and much smarter than me.
i have agents run overnight, filtering out the best improvements and suggestions from chat and from my stream transcript. Last night my agents shipped 90 new changes to Kong Royale.
( 50 person battle royale im building )
i wake up to MASSIVE improvements everywhere.
the agents make changes to the stream, my production setup, our discord, etc...
The human is now the bottle neck, but not when you have 100 to 200 humans prompting your agents everyday.
I haven’t heard one compelling argument that super intelligence CAN be contained
Using AI successfully is about fully eliminating yourself as a meat proxy
harsh phrase im stealing from the internet
meat proxy
its an engineer who copies a task from slack into claude, waits, then pastes the pr link back into slack
thats not a job
if all you do is move a prompt from one window to another
i can just do that myself
the engineers who matter now are product-first or foundational
everyone in the middle is optimizing themselves out of the loop