A collection of GPT-7 powered organoids
Gold Coast, Australia
Joined March 2009
- Tweets3.6K
- Following893
- Followers3.3K
- Likes10.1K
Mike 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.
Mike retweeted
IKEA manuals, but they come to life.
I built the real BILLY instructions in 3D, so you can watch every step happen exactly as drawn. try it: sael.net/flatpack
Mike retweeted
We just open-sourced motion graphics templates.
Each loop is about 30 lines of code your agent can rewrite in your brand colours.
> seamless 12-second backgrounds: depth, lines, dots, shapes
> click a loop to open its code as one HTML file
> a skill your agent uses to design new ones
Comment "MOTION" and I'll send you the library.
Mike retweeted
earshot v0.3 is out!
added:
- a library of 201 sounds, all made with code
- earshot kit can create a whole sound kit
- earshot sfx times a sound to the exact frame (improved)
plus earshot is now used in cinetic (skill for creating motion design)
will continue improving it!
Mike retweeted
Introducing Sound Lab by React Bits: sounds for your interfaces and videos.
• 100 interface sounds
• 100 video editing sounds
• Tune pitch, tone, length and space
• Install with shadcn or download WAV, MP3 or OGG
100% free. Sounds install as code, not audio files.
Mike retweeted
Many asked me how to make these, so I turned it into a skill~
Colorfigures: Describe an everyday object, and Claude draws it hand-inked and isometric, with buttons that actually work.
You can also ask how something works and the skill will label the parts to explain how each part does its job.
Free and open source 👇
Love this! If I had kids, I’d love to teach them this way.
I’ve been exploring learning through play too! A Claude skill for making hand-drawn, interactive everyday objects. Here’s my little washing machine :)
colorfigures.vercel.app
Mike retweeted
Two Anthropic engineers built a prompt system that spits out a rigged Blender monster in minutes, and the trick is that it never designs the creature
It designs the skeleton first. Anatomy decides what a creature can do, and everything strange gets added on top of something that works.
> FRAME - limb count, joint positions and weight distribution set before any mesh exists
> FLESH - muscle volumes grown onto that frame, so the silhouette follows the structure
> WARP - the strangeness applied last, as deformation rather than invention
Image-led creature generation looks incredible and falls apart on the rig. Limbs bend in places nothing could bend.
Starting from a skeleton flips the failure. The monster may look wrong on purpose, but it always moves like something alive.
The warp stage is where the personality arrives. Stretch one limb pair, collapse a joint, invert a spine curve, and it reads as a different species.
Because the frame is parametric, every variant is one number away. A hundred monsters cost barely more than the first.
Weight distribution is the detail everyone skips. It decides posture, and posture is what makes a creature look dangerous or sick.
Studios build this knowledge into one senior artist. Here it sits in a prompt that any junior can run.
The full prompt chain and a batch of monsters from one frame are laid out underneath.
Mike retweeted
Reverse engineered my IoT devices and vibe coded my whole apartment into the menu bar.
Click a lamp in the 3D model, the real one turns on. ACs too.
Why stop at one AC.
Mike retweeted
Today we are rolling out Gemini 4 Argon globally in the Gemini API, to Pro + Ultra in Gemini App, and Ultra subs in Antigravity!
Argon is a huge step forward for Gemini, and a labor of love. This model has been battle tested by over 100,000 people at Google over the last month, with so much new innovation, ambition, and hard work baked into this release. The team has been heads down over the last 6 months working to make this possible across the stack (research, infra, safety, eng, etc).
We rolled the model out to cyber defenders as part of our fairwind program and early access customers over the last 2 weeks, it has been amazing to see and talk to folks who are now using this model daily. Seeing the models writing capability, raw knowledge and persistence for coding, paired with new features like 1M token output when the model needs to work more has gotten me even more excited about our path ahead.
This is just the start of the Gemini 4 era. As you saw with 3.5, 3.6, 3.7, and 3.8 Flash, our iteration speed with Gemini is fast : ) Now with Gemini 4 Argon to work from, I am cannot wait to see the progress continue, and most importantly see what this new model enables for the world.
This is such a special moment in the arc of technology, as millions of people everyday tackle problems they could never have thought possible before. Gemini 4 Argon was built to help you tackle the hardest problems. I have personally felt this spark with Argon and hope you do too. Onwards!!
Mike retweeted
ANTHROPIC JUST RAN AN AI THAT COPIES ITSELF ONTO A $55 SMART TV BOX. AND THERE IS NO OFF SWITCH.
Their own red team spent this month trying to build an AI worm. It worked.
No datacenter, no cloud, just cheap boxes sitting in people's homes:
- 15 off-brand Android TV boxes, $55 each
- a 9B model booted on 8 of 9 device types they tested
- GLM-5.2 ran the whole chain alone: find a device, copy its own weights across, boot up, go hunting for the next one
- every copy reaches a new host every 1-2 days
Now the part that should scare you.
They measured it against a botnet that already took 2,000,000 home devices. That many boxes = roughly 133,000 live AI copies at once, about 40 H100s of compute, humming inside fridges and ovens nobody is watching.
In their own words, there is "no centralized point to shut off".
You cannot patch a worm that lives in two million kitchens.
The best security teams on earth audit every line before they ship. Nobody audits the $55 box behind your TV.
CRAWLHIVE + JEV + GROK BOTS: 9 FEATURES IN 3 DAYS. ONE F*CKING PERSON.
repo: github.com/h100envy/crawl-hi…
three days ago crawlhive was a scanner with crawlers behind it
today it's a different product
every competitor says trust me. we shipped a page that shows every miss we made instead
9 major features. full breakdown of each with real numbers
> jev decides if a signal is worth attention in 0.5 seconds for $0.00002
> tokens that pumped scored 66–83%, the ones that dumped 49–55%, scam 3%
> 29 signals cost $0.0008 total
> grok bot sits on duty over a crawler and calls LOOK / WATCH / AVOID
> jev handles the confident ones, grok gets called only for the doubtful
> chatgpt, claude, meta muse, cursor connect in one click via oauth
> grok bot via mcp
> the agent wakes up only when there's something new and doesn't burn credits for nothing
> track record checks every verdict against the real price for 24 hours
> the misses are visible on purpose
> launch radar scans fresh launches on solana, bnb chain and robinhood chain every 3 minutes
> a full app inside telegram: scan, radar, watchlist, crawlers in one bot
> answers contract addresses right in groups
> watches for rugs for 7 days
> 24 hours later it tells you itself whether the verdict came true
> bnb chain and every solana launchpad including meteora
> AI costs dropped 8–50x
> ceiling of ~$1.25 a day no matter the traffic
> security audit: several agents went through the whole codebase
> every finding checked by hand
> more than a dozen holes closed before anyone else found them
now the main part
3 hours after launch track record showed a pattern
tokens under a day old with linked wallets and a fresh dev wallet were passing as CLEAN and then rugging
SUKI with 12 linked wallets and a dev under an hour old passed as OK and rugged
we changed the rule the same day
now those tokens can't score above RISKY
that's the whole point of a public track record. it shows where the scanner is weak before the market does
no competitor does this. for them it's still on the roadmap
full breakdown in the article with screenshots and how each piece works inside
which scanner besides us shows you where it got it wrong?
Readers added context they thought people might want to know
Anthropic's Oct 12 research describes controlled lab tests projecting that small models may gain self-replication capabilities on IoT devices in coming months, but no actual autonomous worm was built, run, or deployed on real consumer hardware. anthropic.com/research/self-…
securityweek.com/small-ai-model…
Mike retweeted
I can't believe this worked...
Gemini 4 Argon made a Jupiter/Moog combo synth in under 30 minutes.
It's all in a single HTML file, and you can play with it here: argonsynth.netlify.app/
Mike retweeted
ElevenLabs founders must be SHAKING right now...
these guys just dropped a voice AI you can direct like a real actor
no more typing weird little codes into your script... you just write "say it like you've given up on life" and it does exactly that
here's why that's a big deal:
- one word comes out wrong and you change only that word, the rest of the take stays the same
- it switches voices in the middle of a sentence
- it voices a whole scene with several characters from one script
- it clones your voice and keeps it in character in other languages
and to prove it, they cloned Uncle Roger's voice and had it roast his own fried rice the whole way through lmao
🤝 Paid partnership
$52M seed round. 15M+ users. But this is what we're most proud of yet.
Introducing Drama 3: the most emotive and controllable TTS model in the world
And yes, it's open-weight.
You can now control how even a single word should sound, change its tone, and shift your voice mid-sentence.
Uncle Roger was impressed too.
Try it here:
s.fish.audio/rqh0l8
Comment “DRAMA” within the next 48 hours for 1 month free.
If you're an enterprise your first month is on us:
s.fish.audio/qz9r0t
Mike retweeted
Gemini 4 Argon is here!
And for the past week, I've been obsessed with it.
I created dozens of visualizations with real-time data and audio synthesizers; I even tried to replicate Ableton.
I analyzed complex data in minutes and built apps I thought were impossible.
Argon is creative and very fast.
For example, I made this beautiful visualization of current wildfires and tectonic shockwaves in less than 5 minutes.
Made in Antigravity, running locally:
Mike retweeted
直接开源这个3D微缩海岛模型提示词
任何agent、任意模型都可以一键直出这样惊艳的动态效果
传送门至评论区，大家可以自己去尝试一下～
这个能旋转、能交互的 3D 小岛，是我让一个国产新模型对照参考图做出来的，效果和 opus 5.5 真不分上下！！！而且价格还比 opus 5.5 更便宜，性价比简直拉满了 ！
它就是 @StepFun_ai 阶跃星辰的 Step 5 Preview，上线 OpenRouter 第二天，就登上了 Trending 趋势榜的第一名。
Step 5 Preview 是阶跃星辰最新的旗舰基座模型，面向 AI 编程、软件工程和专业知识工作。它支持 1M Token 上下文、文字和图片输入，重点是让 Agent 从理解需求、修改代码到运行验证，把真实任务完整做完。
被它的前端能力和成本吸引，我立马把它接入 Claude Code（CC）进行了测试：做一个微缩版本的佩尼达岛。
整体效果非常夯，大家直接看下面的视频👇
整个网站用 Three.js 实时渲染，海水、岩石等场景元素都由代码生成。对着这个小岛旋转视角、看细节的时候，我真没想到，现在国产模型做 3D 场景已经能有这样的表现。
我的提示流程也很简单：先准备一张生成好的参考图，把网页需求和想要的交互描述清楚，再让它对照参考图反复检查、调整，逐步把效果打磨出来。当然，整个过程需要一些时间，大家可以稍微耐心一点。
目前 Step 5 Preview 已上线 OpenRouter，并接入 Kilo Code、Cline、Hermes Agent 和 OpenCode。不用迁移项目或改变现有工作流，配置后就能切换模型，更多集成也即将上线。
如果你也喜欢折腾前端、交互网页或者 Three.js，可以拿一个自己的小项目试试。我这次做下来，最惊喜的就是它把参考图落成可交互 3D 场景的效果。
体验链接：openrouter.ai/stepfun/step-5…
排行链接：openrouter.ai/rankings?view=…
Mike retweeted
I MADE IT!
I beat Adobe and shipped a better relight feature than them. (theirs is not even released yet).
A multibillion-dollar company with thousands of engineers just got beat by one person.
Crazy thing is that ImageFlow.dev runs in your browser locally for Free 🤯
Let's kill Adobe one feature at a time.
Stay tuned, I'm working on a desktop version of ImageFlow and a new LightRoom alternative that i'm so excited about!
Mike retweeted
A few people asked how I made the isometric objects and cities, so I packaged it into a Claude skill.
Describe an object, device or building, and ISO-GLOW creates a single interactive HTML file.
Try it out yourself!
isoglow.dev