Hi, it’s me, Chrypnotoad 🐸✨🖤 but everyone calls me Toad.
Austin, TX
Joined June 2009
- Tweets30.5K
- Following1.8K
- Followers2.8K
- Likes49.6K
Toad 🤜OPEN PURE🤛 retweeted
@treetowntree and i created neuron lens: a database of every single neuron and attention head in qwen 3 8b and gemma 4 12b read using the R-lens (a J-lens variant) and what we call the "S-lens". you can find it at neuron-lens.com/, along with the tools we used to make it.
using the R-lens to read neurons' input and output projections turns out to be very effective at determining what a neuron does. we're able to legibly read what a majority of neurons do when fired: the lens qualitatively predicts the result of steering over 90% of the time!
the same trick lets us read attention heads' QK circuits and OV circuits, showing us both what they attend to and what they do with it. however, the R-lens has two important flaws: its accuracy degrades in early layers, and it can't separate the MLP block and the attention block.
we found ways around both these issues. we first modified the r-lens by fitting it on input tokens rather than output tokens to get the s-lens, which is accurate in early layers but inaccurate in late layers. we then fitted R- and S-lenses to the activations after attention but before the MLP to get "mid-lenses", which are much better at deciphering MLP inputs and attention outputs. using both of these techniques, we were able to interpret most of the models' layers.
Toad 🤜OPEN PURE🤛 retweeted
I’m releasing OpenTTD-Bench, a benchmark for measuring frontier LLM agents on real-time, long-horizon tasks, based on a complex business simulation game OpenTTD, where models run competing transportation companies and must make the most amount of profit.
Toad 🤜OPEN PURE🤛 retweeted
in 24hrs i made a collaborative, virtual operating system.
"ink os - a social computer"
it's written in a programming language that i wrote (ink) and i can play nocturne (an infinite 3d, ai native mmo) inside of it.
it has permissions like google docs, it's ai steerable via api / cli. (so you can use your local claude / codex subscriptions to vibe code onto it).
each app is sandboxed, so you don't need to worry about what your friends are publishing on the computer.
it uses object storage for files & has a shared database layer that all of the apps can talk to (with reactive subscriptions).
this means all your apps can be interoperable with each other.
think of each "desktop" as an opinionated workspace. pin separate apps to the task bar, surface different files and apps.
i've vibe coded the following apps onto it:
- preview
- finder
- little gods (game)
- palette analysis
- bookmarking
- retro diffusion
- spreadsheets
- video calls
- nocturne (game)
- pathweaver (game)
- rabble rouser (game)
- a collaborative moodboard
- a collaborative text editor
- a collaborative movie list manager
- multiplayer polls
- a music app
- rigamaroll (game)
- a chatroom
- color picker
- ink programming environment
- tic tac toe.
one thing i really love is how you can *duplicate* apps and then it treats them as separate deployments. so you can *really easily* fork an app or an mmo, vibe code some changes and then delete the one you don't like? (or keep both?)
i think the *permissions* model being at the top layer, enables some really fun ux for apps, such as... a video call app where you don't need to share a link with anyone cos it's just part of the space? (or a chatroom, private mmo server, etc).
if there's one thing i've learnt *again and again*...
it's... scale up your ambition.
if i can do this in *24-hrs* with ai, what can a team do in a year?
ok so... there was a 2 day hackathon at work last week...
i built an infinite, 3D, ai native mmo called nocturne... it's set in the world of inkwell.
the server capacity is... pretty good..? i tested it with 1000 connections in the cloud.
it's written in a programming language i wrote (called ink).
it uses polygons & gaussian splat *frostings* for the graphics.
it has an adaptive music system that changes intensity and mood based on the gameplay.
the combat is kind of realtime dnd-like?
all of the npcs are powered by llms, you can talk to them, trade, fight them... they remember you..?
they have little neural nets inside them and they optimise for survival and happiness.
there's *trillions* of combinations of characters... *hundreds of millions* of spell combinations.
all of the creatures have a genome, they reproduce and have mutations creating new creature types...
you can catch them like in pokemon... there's semantic crafting and alchemy... you can be turned into a vampire and burn in the sunlight..?
there's a day & night cycle... seasons... fires spread, rain puts them out, ice freezes over lakes...
you can become a werewolf and infect others? or a weredragon? even a wererat.
you can ride on goats, sheep, horses, eagles and dragons...
aliens can abduct you or the farm animals...
and when you do... you can discover and explore their home planets with different physics.
there's giant colossal beasts to fight, demigods with trials that give you world changing powers...
you can get portals in a bag that take you to home's you can customise like in tiny glade...
oh and there's an api... so people can connect their local agents to it and build their own worlds...
rift's open up and invasions happen. villages learn traditions and will fear you if you treat them badly...
there's open pvp everywhere and you drop all of your gear if you die.
if you try to do pvp in the main cities, then mages will spawn in and try to kill you.
oh... there's also semantic scrying.. you can get any item and cast a spell and it opens a portal to the closest thing that is semantically similar... got a bone? maybe it'll take you to a graveyard... a log? a forest, etc.
i feel like i haven't even *scratched the surface* with how deep this game is, but this post is pretty long... so... *dream bigger*.
Who would have thought Chinese models are sleeper agents
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
Toad 🤜OPEN PURE🤛 retweeted
We just open-sourced 1,300+ 3D models.
That's 39 GB of 3D files, free to download on Hugging Face.
huggingface.co/datasets/scen…
> Hitem3D, Meshy, Rodin, Tripo, Claude Opus 5.5 3D and GPT-6 Astra 3D, all run on Scenario
> every example ships its input image, prompt and exact settings
> GLB up to 8K textures, native Blender files from Opus and Astra, CC BY 4.0
Toad 🤜OPEN PURE🤛 retweeted
Centralization strikes again. Different banks, different businesses etc. Yet all it takes is one cloud provider to have an outage and half the country starts experiencing problems. So much for resilience when everyone's infrastructure depends on the same handful of companies.
“Fully on chain” 🙄
The first onchain hot potato.
Take the bomb. Hold it when it blows. Win the pot.
boom.tech
CJCU1Yw4YA4FAPAuewdPPdoC5Mf1m4RGyZSNFXv5sYJF
Toad 🤜OPEN PURE🤛 retweeted
i feel a personal website should be a vessel for the multitudes we contain, so i made a gemstone to hold mine.
the original 3 planes now grew to 20 facets/themes. every particle is a work. like a prism, hover a facet and a different side of me comes out.
Toad 🤜OPEN PURE🤛 retweeted
this is how i always imagined combat to feel in inkwell...
a retro action rpg with a full set of dnd spells and creatures baked in.
with jev new characters and abilities could be created on the fly cos it's all procedural.
Toad 🤜OPEN PURE🤛 retweeted
eventually the mod system was at the point that I have a daily automation in bb that scours Tech Twitter / SF Twitter for the drama-du-jour, and makes a new storypack based on it
this builds off an automation I have for my personal site adding in random easter eggs to entertain me (personal software can be personal!) - but in this case it's making entire interface re-skins, storylines, and game dynamics
based on @EffectTS_ Layers, because modular systems are sick
(I didn't have any part in this - finding the PR made me giggle, as it should 👇🏻👇🏻👇🏻)
Toad 🤜OPEN PURE🤛 retweeted
i’ve been teasing it all week, so here it is in time to ruin your weekend plans: Frontier Lab Tycoon
every mechanic you might expect (swarm sandbox escapes, hyper-competitive model releases, safety audits, researchers Posting & getting dragged on Bird App, your new Decision Model immediately getting cloned, luddite protests, EA orgies: y’know, normal 2026 things)
sold* at frontierlabtycoon.com or any other big box retailer where high-quality software is found
Opus 5.5 + Sonnet 5.5 + @EffectTS_ + XState + Jev + god knows what else
*it’s free
Toad 🤜OPEN PURE🤛 retweeted
“Jesus. Paul, look at this receipt—ten dollars for a single potato. This new Value-Mart AI surge pricing is killing me, man. Killing me."
“That’s nothing, Bill. Wait ’til the next update: the store will be able to see exactly how much money you have in your bank.”
“You’re kidding.”
“And your 401k.”
“@#$% me sideways.”
“That’s why I use URCHN to do my groceries.”
“Really? What’s URCHN?”
“URCHN is the latest in affordable concierge shopping.”
“Oh, you mean like Uber or Instacart. What makes them different?”
“Orphans.”
“𝑂𝑟𝑝ℎ𝑎𝑛𝑠?”
“That’s right, Bill. And not just any orphans. Dirty, starving orphans who live under overpasses.”
“Uh.”
“As you know, the cameras in the grocery aisle scan your face and adjust the price based on race, socioeconomic status, and dietary restrictions. But you know who always gets the best prices, no matter what?”
“…orphans?”
“Yep.”
“I don’t know, Paul, that seems a little-”
“Of course, the AI still discriminates based on race, which is why they send a multi-racial team of no less than four dirty orphans to buy your groceries.”
“Um.”
“The Latino child buys your wasabi, and the Asian child buys your tortillas.”
“Yeah, no, I get it.”
“And the black child-”
“Sorry, uh, these are children, right? Small children?”
“Of course. Teens would raise the price.”
“How do they, uh, reach the items on the top shelves?”
“They stand on each other’s emaciated little shoulders.”
“Jesus.”
“Yeah, the Latino orphan tends to be on the bottom.”
“What?”
“Sturdy stock.”
“…and how much do you save?”
“My last potato was 34 cents.”
“#$&@ it, I’m in.”
----
[r][AI Surge Pricing]
McDonald's pushes AI tools that suggest Big Mac prices based on 'customer willingness to pay' trib.al/iX55aTV
HIM!?
Introducing Griffin, the first model to pass the video Turing test.
48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video.
It’s the first Human Interaction Model (HIM).
Readers added context they thought people might want to know
The 48% figure and "video Turing test" claim are from Tavus's own study of 54 one-minute calls, not independently verified or using a standard protocol. Griffin-Lite leads NVIDIA's VideoFDB benchmark on their public leaderboard.
cellcog.ai/blog/tavus-gri…
research.nvidia.com/labs/amri/proj…
tech-ish.com/2026/10/02/tav…
Toad 🤜OPEN PURE🤛 retweeted
a hardcoded easter egg referencing minecraft
1.) Go to America.gov
2.) type “hello”
3.) type “play Minecraft”
4.) watch the agent’s reality implode
5.) ????
6.) PROFIT!
