dydx (@dy2dx3) header
dydx (@dy2dx3) profile picture

Vibe Coding || Connect to explore || Into AI || Making AI Simple, Useful & Exciting for everyone || DM for Collab

Joined September 2022
Photo by dydx (@dy2dx3)Photo by dydx (@dy2dx3)Photo by dydx (@dy2dx3)Photo by dydx (@dy2dx3)Photo by dydx (@dy2dx3)Photo by dydx (@dy2dx3)Photo by dydx (@dy2dx3)Photo by dydx (@dy2dx3)Photo by dydx (@dy2dx3)Photo by dydx (@dy2dx3)
Pinned Tweet
been saving a lot of video edits for inspiration but describing the exact pacing and motion in a prompt never really works. with Tesseract from @trymirage, you can show your agent a reference video and have it recreate that style with your own footage. everything stays editable after so you’re not stuck with the first result.
Any video can now be the starting point for your own project. With Tesseract in Claude or ChatGPT, paste a link or upload a reference video, then tell your agent which elements to bring into your new project. Kudos if you recognize the inspiration behind the below.
3
4
1
12
459
dydx retweeted
Create 3D worlds with Claude using Tesseract.
11
10
5
139
9,076
Violet Glass treats the controls as objects floating over the film: frosted violet glass, soft rim light, custom icons and a neon magenta timeline. On desktop, the full-width timeline sits above a centered play bar. On tablets and phones, the three main... contra.com/community/tfvyR3O…
3
70
dydx retweeted
Did you know there's a video editing timeline right inside ChatGPT? With the updated Tesseract plugin, watch your video and review its timeline beside your conversation. Tell your agent what to change, see the results, and keep shaping the same editable project, all without leaving ChatGPT. Try the updated Tesseract plugin. Link below.
37
12
15
178
144,014
Anyone having problems with subscription cancel on kiro ? Please tell me how to cancel subscription
2
25
I saw an agent try to bypass an independently enforced signing policy. I saw the NEAR contract reject it. I saw that the same locked contract could construct and sign a genuine Hyperliquid-format order. And I understand exactly which risks remain.
Who gets to sign? An AI agent can propose a trade. Should it also hold the key that signs it? Most "agent guardrails" are checks inside the same process that holds the key — so a hijacked agent signs anyway. ZeroDelta draws a different boundary. The agent proposes. A keyless NEAR contract — zero access keys, immutable policy — is the only account that can request a signature for the treasury's hedge key. WHAT IT ACTUALLY GUARANTEES One order shape: an IOC sell on a listed market, never a transfer or withdrawal. Price-independent coin caps: 40 ETH per order, 80 ETH and 6 orders per 24h. Expiry within one hour, hashed into the order. Replay protection. And the EIP-712 digest is built on-chain by the contract — a solver-supplied digest is never signed. WATCH THE MOMENT In the video I submit an order priced at 1/10 of market to inflate its size 10×. The locked contract recomputes it and rejects on-chain: ORDER_SIZE_CAP, 159.5779 ETH against a 40 ETH cap. No signature. No trade. The guard can't be talked out of it. THE PROOF The contract built a Hyperliquid testnet order, computed its digest on-chain, and called NEAR Chain Signatures. Hyperliquid recovered the contract's own MPC-derived address from that signature: 0xa74e082a…859f5e. Flip one bit of the signature and a different address comes back. That's signer recovery, proven end-to-end. THE HONEST PART Signer recovery is not a fill. No order executed, no funds moved, testnet only. Delta/leverage/slippage are a consistency check on solver-reported legs, not proof of a hedge — we demonstrate that live: a naked $80k short with an $800M reported treasury passes the math, and the coin caps are what bound it. This key cannot sign closing buys or withdrawals, and the TEE is software-simulated. Every limit is labeled in the repo and the video. THE EVIDENCE 208 TypeScript tests. 132 Rust tests. 95/95 live verification checks. 27/27 deliberate contract mutations caught. Pinned WASM with byte-for-byte source rebuild. Re-run everything: npm run verify. TRY IT (no wallet) Live read-only checker: zerodelta.vercel.app Code + proof: github.com/g4stein/zerodelta Built at NEAR HACKATHON by a solo builder, on @NEARProtocol Chain Signatures + Intents. #NEARHackathon Would you let a contract with zero keys be the only signer for your treasury — and what would you still want to verify? @NearBuilders @NearLegion
1
17
Built ZeroDelta for #NEARHackathon 🛡️ AI agents can propose trades, but only a locked NEAR Risk Guard can authorize MPC signing after checking risk. Real testnet proof: 3 attacks rejected. Demo: zerodelta.vercel.app Code: github.com/g4stein/zerodelta @NEARProtocol
1
1
90
Updated ZeroDelta demo: v3.1 contract-built signing, exchange signer-recovery evidence, and live checks showing both enforced restrictions and remaining trust assumptions. Testnet prototype; no fills or funds moved. teapaw.org/dy2dx3/status/21208114…
Who gets to sign? An AI agent can propose a trade. Should it also hold the key that signs it? Most "agent guardrails" are checks inside the same process that holds the key — so a hijacked agent signs anyway. ZeroDelta draws a different boundary. The agent proposes. A keyless NEAR contract — zero access keys, immutable policy — is the only account that can request a signature for the treasury's hedge key. WHAT IT ACTUALLY GUARANTEES One order shape: an IOC sell on a listed market, never a transfer or withdrawal. Price-independent coin caps: 40 ETH per order, 80 ETH and 6 orders per 24h. Expiry within one hour, hashed into the order. Replay protection. And the EIP-712 digest is built on-chain by the contract — a solver-supplied digest is never signed. WATCH THE MOMENT In the video I submit an order priced at 1/10 of market to inflate its size 10×. The locked contract recomputes it and rejects on-chain: ORDER_SIZE_CAP, 159.5779 ETH against a 40 ETH cap. No signature. No trade. The guard can't be talked out of it. THE PROOF The contract built a Hyperliquid testnet order, computed its digest on-chain, and called NEAR Chain Signatures. Hyperliquid recovered the contract's own MPC-derived address from that signature: 0xa74e082a…859f5e. Flip one bit of the signature and a different address comes back. That's signer recovery, proven end-to-end. THE HONEST PART Signer recovery is not a fill. No order executed, no funds moved, testnet only. Delta/leverage/slippage are a consistency check on solver-reported legs, not proof of a hedge — we demonstrate that live: a naked $80k short with an $800M reported treasury passes the math, and the coin caps are what bound it. This key cannot sign closing buys or withdrawals, and the TEE is software-simulated. Every limit is labeled in the repo and the video. THE EVIDENCE 208 TypeScript tests. 132 Rust tests. 95/95 live verification checks. 27/27 deliberate contract mutations caught. Pinned WASM with byte-for-byte source rebuild. Re-run everything: npm run verify. TRY IT (no wallet) Live read-only checker: zerodelta.vercel.app Code + proof: github.com/g4stein/zerodelta Built at NEAR HACKATHON by a solo builder, on @NEARProtocol Chain Signatures + Intents. #NEARHackathon Would you let a contract with zero keys be the only signer for your treasury — and what would you still want to verify? @NearBuilders @NearLegion
1
19
Who gets to sign? An AI agent can propose a trade. Should it also hold the key that signs it? Most "agent guardrails" are checks inside the same process that holds the key — so a hijacked agent signs anyway. ZeroDelta draws a different boundary. The agent proposes. A keyless NEAR contract — zero access keys, immutable policy — is the only account that can request a signature for the treasury's hedge key. WHAT IT ACTUALLY GUARANTEES One order shape: an IOC sell on a listed market, never a transfer or withdrawal. Price-independent coin caps: 40 ETH per order, 80 ETH and 6 orders per 24h. Expiry within one hour, hashed into the order. Replay protection. And the EIP-712 digest is built on-chain by the contract — a solver-supplied digest is never signed. WATCH THE MOMENT In the video I submit an order priced at 1/10 of market to inflate its size 10×. The locked contract recomputes it and rejects on-chain: ORDER_SIZE_CAP, 159.5779 ETH against a 40 ETH cap. No signature. No trade. The guard can't be talked out of it. THE PROOF The contract built a Hyperliquid testnet order, computed its digest on-chain, and called NEAR Chain Signatures. Hyperliquid recovered the contract's own MPC-derived address from that signature: 0xa74e082a…859f5e. Flip one bit of the signature and a different address comes back. That's signer recovery, proven end-to-end. THE HONEST PART Signer recovery is not a fill. No order executed, no funds moved, testnet only. Delta/leverage/slippage are a consistency check on solver-reported legs, not proof of a hedge — we demonstrate that live: a naked $80k short with an $800M reported treasury passes the math, and the coin caps are what bound it. This key cannot sign closing buys or withdrawals, and the TEE is software-simulated. Every limit is labeled in the repo and the video. THE EVIDENCE 208 TypeScript tests. 132 Rust tests. 95/95 live verification checks. 27/27 deliberate contract mutations caught. Pinned WASM with byte-for-byte source rebuild. Re-run everything: npm run verify. TRY IT (no wallet) Live read-only checker: zerodelta.vercel.app Code + proof: github.com/g4stein/zerodelta Built at NEAR HACKATHON by a solo builder, on @NEARProtocol Chain Signatures + Intents. #NEARHackathon Would you let a contract with zero keys be the only signer for your treasury — and what would you still want to verify? @NearBuilders @NearLegion
3
1
7
403
Animation made by claude Inspo from games by noclipepe
1
149
If you're still farming testnet projects for airdrops through XP and point systems, you're wasting your time. You're already way behind the race. Most of these projects have simply been farming their communities instead of rewarding them. ( you already know.) If a project has officially confirmed an airdrop, rewards, or dropped strong hints about future incentives, then it might be worth your time. The reality is that airdrops aren't as surprising as they used to be.
1
23
This is the talk I’ve been waiting for. The shift from “better prompts” to “systems that manage their own memory and improve over runs” feels like the actual unlock for agents that are useful in production. Especially the part about in-band memory hitting limits and the need for an out-of-band process (“dreaming”) to consolidate and clean up memory across sessions. Most of the agent frameworks I’ve seen still treat memory as an afterthought. Curious how many people here have actually shipped something with a real memory + consolidation loop vs still mostly hand-prompting. The 90% number at Anthropic is wild if true.
This is f*cking insane. Anthropic engineer: "You're not supposed to prompt Opus 5.5 You're supposed to build a system that prompts itself. At Anthropic, 90% of our engineers have already built these systems" In 30 minutes, she walks through exactly how Anthropic builds agents that remember, fix their own mistakes and get sharper with every run. Better than any $1000-$2000 agent course I've seen. Watch it, then read the article below ↓
1
3
156
dydx retweeted
Fable 5.2 modeled a historically accurate White House, then vs. now. Created with Tesseract 3D.
13
16
4
168
14,177
Fable 5.2 is now available in the Claude Platform and Claude Code. It’s up to 30% faster than Fable 5.1, and costs $8/$40 per 1M tokens, down from $10/$50.
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges. They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
62
Dreamina Seedream 5.5 makes its global debut! 🧑 More realistic portraits ‣ Natural skin texture and lighting, without a greasy or artificial look. ‣ Upgraded beauty, makeup, retouching and photoshoot-style portraits. ✍️ More accurate text ‣ 95% text accuracy—even complex characters and dense layouts stay clear. ‣ Generate commercial-quality posters, infographics and presentation slides directly. 🖌️ More precise editing ‣ Edit selected areas while preserving untouched regions pixel-perfectly. ‣ Refine the same image repeatedly without losing detail. 🧊 Transparent assets, no cutout needed ‣ Generate transparent and semi-transparent assets from one prompt, with crisp hair and sheer-fabric edges. ‣ Skip manual cutouts and reuse assets across scenes. 👉 Now open to all users for the first time. Try it today! 🕐 Limited-time offer: Unlimited use—7 days for everyone, 30 days for Basic and higher-tier subscribers. 🎁 Repost and comment within 12 hours—the first 200 participants will each receive 1,000 Dreamina credits. First come, first served. #DreaminaSeedream5_5 #DreaminaAI
192
184
52
415
149,429
dydx retweeted
I've created your 40-second 1080p 10-bit HDR Rec.2020 motion graphics film exploring the fourth dimension:
1
2
173
dydx retweeted
Took on the @tradeifyforex certificate challenge and redesigned all three as one set. What's different: → Every certificate has a QR code + serial so anyone can verify it's real → Guilloché line work, microprint border and a holographic seal, the way banknotes do it → Each one draws the trader's own equity curve and stats, so no two look alike → One HTML template renders all three from account data, so it can go live as is Matches the Tradeify FX look: dark glass and cyan glow, no clip-art. #TradeifyRedesign
DESIGN CONTEST 🎨 Think you can make our certificates look better? Prove it. Redesign any of our: → Payout certificate → Phase 1 pass certificate → Certified Trader certificate We’ll pick a Top 3 and rank them: 🥇 1st place: $1,000 cash 🥈 2nd place: $50K account 🥉 3rd place: $50K account How to enter: 1- Quote this post with your design 2- Repost this post 3- Tag @tradeifyforex + #TradeifyRedesign 4- Drop it in our Discord too: discord.gg/tradeifyfx Winners will be announced next Friday. ⏳
1
1
1
114
Every major tech CEO is vulnerable to Chinese female spies right now. Except Sam Altman. Because he isn’t into women. I’m bullish on OpenAI. Who else is?
62
When you buy the dip but it keeps dipping vs me looking at the charts. @lobincoin $LOBIN
1
27
1. Heading to the strip club after a +3% green candle — LOBIN flexing in the green thong and suspenders, already celebrating like the bag just went parabolic. One small green candle and the claws are out. 2. Me after not looking at charts for 5 minutes — chilling by the pool with money raining down, cigar lit, completely unbothered. Chart watching is optional when the vibes are this strong. 3. When my $5 crypto bag pumps 20% — suddenly the gold watch comes out, cigar in mouth, feeling like the biggest whale on Robinhood Chain. Tiny bag, massive confidence. @lobincoin $LOBIN
1
87