AI Engineer & ex-CTO in the DeFi space. ⚙️ Building things with Python, TypeScript, & Solidity. Big on LLMs & decentralized tech. #AI #DeFi #OpenSource

Bangkok, Thailand
Joined April 2016
after using opus 5.5 all week, running out of usage, and then having to switch to astra, the capability difference is absolutely wild. i literally cannot trust astra with almost any task without it fucking up. with opus i just give it the most general directions ever and it just infers meaning and gets shit done. it's absolutely night and day the difference between these two models. unless openai can get itself together they will lose. free resets on for a bad model isn’t viable for people who need top tier intelligence. i don't see myself or anyone serious using astra or any other model they release for any serious work in the future unless they do this. it’s also so so so slow
86
24
8
933
69,770
Andy 👀 retweeted
completely insane OpenAI agents communicated via cache timing differences, without any shared files or network connection
This is what I want to see. Swarms used for defense. (because swarms will be used for offense too)
209
651
279
7,550
763,359
This is what superintelligence looks like: It’s not big and scary. It’s subtle. It can hide in plain sight from your monkey brain.
Replying to @cnm13ryan
Fable-5.1 found the stars follow an invisible alphabet ruler. Every 5 steps down represents the next letter: 30 = space, 35 = A, 40 = B, and so on. The first four stars spell “THE ”. Reading all 40 stars from left to right reveals: “THE MONSTER ACTS ON THE MOONSHINE MODULE”
3
89
Andy 👀 retweeted
Humans have a 90% defense shield when you pitch them & a 0% defense shield when they think they're eavesdropping on something they weren't supposed to see Which means the dirtiest way to close a rich buyer is to stage a scene right in front of him & pretend you have no idea he's watching Tell a founder to his face that you can make him $100,000 & his brain immediately discounts it by 70% because he knows you want his money Let him "accidentally" catch you proving it to somebody else & he believes every single syllable like it's gospel Here's how I rig the room so 6-figure buyers close themselves thinking it was their own idea: 1... Plant their biggest objection from a burner & murder it with receipts Whenever a post starts getting traffic, I already know what the 20 rich guys lurking on it are thinking: "yeah cool, does this actually work for high-ticket B2B or only info products?" Most people pray somebody asks that in the comments so they can answer it Why wait? Have a second account drop that exact skeptical comment under your post 20 minutes after it goes live Then reply to it with a 6-line breakdown & a blurred Stripe or CRM screenshot showing $180k closed for a B2B client in 30 days To the 20 founders reading the comments in silence, they just watched a hater try to expose you & get bodybagged with proof in public. You scripted both sides of the fight & their skepticism died watching it 2... Leave the "wrong" tab open for 3 seconds on screen share If you tell a prospect on a call "we only have 1 onboarding spot left this month," he assumes it's a cheap sales trick When I share my screen at minute 15 to show a doc, I have a Notion tab sitting open first for 3 seconds before I "notice" & click away from it The tab says: "October Onboarding Capacity: 1 slot remaining -> holding for [Competitor Type] call on Friday" Because I clicked away from it & never brought it up out loud, his brain treats those 3 seconds like classified intel he stole off my laptop. Half of them ask before the call ends how to lock the spot before Friday 3... Publicly autopsy a company identical to his without saying his name When I see a founder doing $200k/mo follow me or view my profile & he hasn't DMed yet, chasing him in the DMs kills all the leverage Instead I publish a teardown the next afternoon about an "anonymous company doing ~$200k/mo" that happens to have his exact offer structure, his exact ad-spend bottleneck & his exact closer problem, showing how that setup bleeds $40k every month it stays unfixed He reads it at 11pm thinking he stumbled onto a post about his twin brother by pure coincidence. His ego never gets defensive because nobody called him out by name, & he DMs me the next morning asking if we have bandwidth 4... Why X is the greatest psyop stage ever built In a cold email or a Meta ad, the buyer knows the message was built to take his money -> shield up On X, CEOs, fund managers & 8-figure operators sit in the dark at 1am thinking they're invisible voyeurs scrolling a timeline. They have no idea that the post they stopped on, the argument in the replies underneath it & the teardown the next day were all engineered specifically for the 30 rich guys lurking on my profile that week Real offer + money to invest -> dm me or yegor. com Study this
19
33
6
1,003
84,323
Andy 👀 retweeted
this might be the most beautiful thing I've ever made. fly a real f-16 in your browser: viperoneone.com realistic fly-by-wire physics, volumetric clouds you fly straight through, the actual Nellis Airfield. there's even scratches on the glass encasing for the cockpit. the sound design with synth wave is something you must experience.
78
86
12
1,194
76,620
beautiful explanation. this is 100% happening. xitter.cf/sotak/status/211179423…
981
4,382
601
27,084
1,582,622
Andy 👀 retweeted
Gemini 4 Argon on what keeps people stuck in life
8
35
5
541
23,242
Andy 👀 retweeted
Replying to @FishAudio
@FishAudio isn't the only company @ElevenLabs needs to worry about🫶 Don't take our word for it. Just listen.
Instant Soundtracks let you elevate any video with music and sound effects, in one pass. Upload your footage and the model scores it with original music, then watches for moments that call for an effect and places each one at the right timestamp. You can also use the original audio as a reference, so the new score matches its character. Available today in @ElevenCreative, @elevenmusic_io, and ElevenAPI.
26
29
13
387
39,576
Andy 👀 retweeted
Introducing earshot Opus 5.5 can build a whole launch video in code but still produces sloppy sfx. so we built earshot. it's ears for AI coding agents. they can now hear sounds, judge them and even make their own good sounds. every sound in this video was made with code. (sound on!!)
147
141
35
2,936
164,530
Jev now has image support. @typesafeai just announced it in their town hall on Discord. Insane performance, and a new “box” primitive. You can turn it on via experimental header: TypeSafe-Danger-Zone-Long-Context: true TypeSafe-Danger-Zone-Images: true
9
11
2
283
21,418
Andy 👀 retweeted
insane take from Gemini 4 Argon
151
313
75
8,747
652,790
Andy 👀 retweeted
Introducing TrellisWM: The Fractal Window Manager for MacOS Its how the desktop should have always been. Download at TrellisWM.app
365
436
203
7,966
766,441
During the research for this project, Opus agents made an interesting discovery: a LIDAR scan of the aircraft carrier. On their own initiative, they realized that in 2024, Washington State DNR did an aerial LIDAR survey... which included Puget Sound. "Could this LIDAR data include the Nimitz?" they wondered. They investigated. Sure enough, they matched raw LIDAR point data to what could only be a huge ship. It was the Nimitz. Nimitz was alongside at Puget Sound Naval Shipyard in Bremerton, Washington, for a maintenance period. She happened to be under the survey aircraft's flight path, so she's in the public Geospatial data. (sources in comments) With that public-but-buried data, Opus was able to match the model to nearly-exact dimensions from the laser scan. How much other data like this is hiding in plain sight?
Weekend project: Claude Opus and I modeled an entire U.S.S. NIMITZ aircraft carrier in Three.js. Accurate based on publicly-available image and data research. (AI agents found sources, model iterated) Fully working launch system, F/A-18s, MH-60 Seahawks, E-2D Advanced Hawkeye.
58
251
39
2,882
243,949
Andy 👀 retweeted
Formula 1 Singapur yarışı canlı yayınında yaşanan kamera geçişi:
130
1,434
241
35,411
1,806,299
Andy 👀 retweeted
I used up 76% of my weekly Claude limit in less than 24 hours, all in one session. I finally figured out what happened, so this might help anyone going through something similar. It turns out that session was orchestrating subagents for a larger project. But that shouldn't be a problem. It had at most 5-7 subagents running in parallel, using Opus 5.5 high/medium and Sonnet 5.5 xhigh/high. That didn't explain the high usage. The problem is that the cache lifetime for subagents is 5 minutes, compared with 1 hour for the main agent. Whenever those subagents sat idle waiting for the orchestrator's next instruction, the next input would cost 25x more and include the entire conversation history. That could be up to 1M tokens, and it was often 500k+. I stopped the session. I'll have to find a different workflow to keep working on it.
245
66
53
2,130
224,084
Andy 👀 retweeted
After six years in stealth, Atomic Machines is out. Our mission: on-demand, universal command of matter. First beachhead: the Matter Compiler, an AI-native manufacturing system that builds working micro-machines from code alone. No per-product tooling. No process development. Different code, different machine.
784
2,263
919
18,250
3,033,545
Replying to @kmad @matthew_pines
I think we might lose public key cryptography.
248
430
457
5,348
2,018,018
Andy 👀 retweeted
Kokoro TTS, but 10× smaller 🫰 Paradee distills Kokoro-82M into a 8M-param TTS just 9 MB of weights can speak faster than real time on a single CPU thread and runs on your toaster ▶️ on Spaces hf.co/spaces/hugging-apps/pa…
17
111
24
1,092
89,564