avatar
Hedda🐽
27,298 Twitter followers
Follow
小豬🐽|Team @TheUnderdogsHK @createra_ |Ambassador @Scallop_io @Glypherio|@XCircle1314 @zjubca 玩合約用BingX|理財資金作保證 https://bit.ly/hedda
Posts
avatar
Hedda🐽
03-20
Five tech giants squeezed through the same door within five days—the AI ​​world is in an uproar! You must know that! On the 18th, Stripe, Tempo, and Paradigm jointly released the payment protocol MPP, and the Tempo mainnet. On the same day, Visa established Crypto Labs and released the Visa CLI. On the 17th, Mastercard acquired stablecoin infrastructure BVNK for $1.8 billion. Did you notice Coinbase's x402 protocol upgrade and WLFI's release of its AgentPay SDK last night? They're all vying for the same thing: AI's bank cards. Over the past three years, every layer of the AI ​​industry chain has been controlled. - The model layer is the playing table for OpenAI, Anthropic, and Google. - The computing power layer is monopolized by NVIDIA - The application layer, from programming assistants to search engines, has become a red ocean, but the battle at the payment layer has already begun. Stripe × Tempo built an open MPP protocol, and Coinbase x Cloudflare built X402. Last year, Google launched the AP2 consortium to develop global standards for AI Agent payments, and HashKey was among the first compliant digital financial institutions in the Asia-Pacific region to enter the field. I remember their Dr. Xiao Feng saying something like this: "If an AI agent is to begin operating independently of humans and creating economic value, it will definitely need an account. Currently, the only suitable form seems to be a digital wallet, rather than an account opened for the agent through a bank." Traditional bank accounts have limitations on the number of accounts that can be opened, unclear liability, high payment costs, and low efficiency. When 5 billion AI agents simultaneously conduct high-frequency, micro-amount cross-border settlements globally, traditional systems will be instantly overwhelmed. AI and blockchain are two sides of the same coin, and that's exactly what HashKey is doing. With custody, exchanges, public chains, RWA, and stablecoins all in hand, HashKey is beginning to dominate other crypto projects, especially considering they went public on the Hong Kong Stock Exchange last year. The AI ​​economy needs the same thing: a clearing language, settlement channels, and trusted compliance anchors designed for agents. HashKey has already established a presence in these three areas: the Hong Kong VASP license serves as a compliance anchor; the public blockchain and stablecoins act as settlement channels; and joining Google's AP2 consortium to jointly develop standards with PayPal, Circle, and UnionPay International. Is HashKey the SWIFT of the AI ​​era? This article is a collaboration with @HashKeyGroup and features @tiezhucrypto.
WLFI
1.45%
avatar
Hedda🐽
02-15
Thread
Anyone who's used OpenClaw knows that different AI models used with it make a world of difference 🦞 My problem is that OpenClaw can't use the Claude Max solution, only the Claude API. It's very intelligent but extremely expensive, costing me 50u upfront. But once you've used Claude, you can't go back. My impression is that in terms of intelligence: Claude > Gemini > Kimi > GLM. I haven't tried the others. (GLM 4.7's inability to read Traditional Chinese is driving me crazy; Gemini has a 300u free credit.) I think the same way as Benson @BensonTWN: use Claude Code as the brain of OpenClaw. ▪️ My architecture: Telegram → Claude Code CLI (Opus 4.6) → Response to Telegram ▪️ Benson's architecture: Telegram → OpenClaw Gateway → Simulate TTY → Claude Code CLI (Opus 4.6) → Response to Telegram Mine doesn't have this. Gateway setup is simple, but Benson is more advanced, possessing both the complete OpenClaw ecosystem and the native capabilities of the Claude Code CLI. Features include semantic memory search, task scheduling, sub-proxies, and proactive messaging. My bot has a major flaw: it forgets everything after each reboot. It can only remember who it is by reading the archive. Benson requires maintaining both the OpenClaw and Proxy layers, potentially necessitating adjustments when OpenClaw updates. My architecture only relies on Anthropic's official Claude Code, avoiding compatibility issues. Aside from the memory loss problem (which has been resolved), other functions like programmatic computer control are identical! I've seen many people using Session Tokens in Anthropic and getting easily banned, even with the Max solution. Benson and I use the Claude Code CLI, which has excellent security. Yesterday I figured it out from @Tz_2022 ⬇️ A gateway can have multiple agents. An agent can have multiple sessions. A session can create multiple sub-agents. Now I have two agents and one independently running agent on one of my gateways, that's three agents! Anyway, both methods are very convenient! Choose whichever you prefer. / I want to become a professional lobster mom 🦞 twitter.com/Rav_Hedda/status/2...
OPUS
1.3%
loading indicator
Loading..