1. When will the book be published? Answer: Not recently, the more I see, the less I want to express.
2. How is Binance's AI progress planned? Sister He believes that AI can achieve "quantitative equality." Currently, it is in a "semi-disabled" iteration phase, but will continue to explore the deep integration of Binance AI with financial products.
By the way, everyone can try Binance AI Pro right now.
It's getting better 😂 Today, Big Brother AMA answered my question.
I asked CZ about the recent performance differences between gold and Bitcoin, and the debate in Dubai. Bitcoin has various advantages over gold, but why is gold crushing Bitcoin's trend? CZ's answer,
adoption rate. It's crucially important, and it's still gonna take more time for most people to accept Bitcoin as the new money system.
Just like AI is amazing and can do all kinds of work. But that doesn't mean it can do it today.
This adoption rate is the reason for the noise generated in the time lag! Feeling more confident, continue to HODL!
Jaredfromsubway got wrecked for $7.5 million by a reverse MEV honeypot; don’t just think the whales are the only ones getting hit. The attacker set up 66 fake tokens and fake pools, turning the 'optimal path' into an authorization trap. As on-chain activities become more automated, profits and risks will be executed faster and colder.
Rubin Ultra HBM, this round's 16Hi / 12Hi debate, hold your horses before picking a side. The AI hardware chain is super sensitive right now; even half-baked spec leaks could trigger pricing action: memory, packaging, yield, and stock prices all moving together.
Goldman Sachs predicts that enrollments in computer science and programming will drop by over 10% for the 2025-2026 academic year. It's not that students suddenly lost interest in coding; they first caught a whiff that job security is shifting. The thing about AI and humans competing for jobs isn't just about the salary sheet; it might start with the sign-up sheet.
Japanese corporate pensions are putting 1% of their assets into crypto, which is a small allocation. But this isn't just about positions; it's a permissions story: being able to get into the investment committee, risk control sheets, and product lists is what counts for institutions seriously getting involved. Crypto really making its way into institutions isn't just a gas pedal move; it's about being allowed to take a small taste first.
The market storage is most afraid of just being all talk. Korea's data from 20 days prior to June is clear: DRAM +342%, NAND +336%, MCP/HBM +209%, SSD +405%. Price can be noisy, but orders don’t lie. AI infrastructure ultimately comes down to how much you can deliver.
AI security tools getting cheaper doesn't mean projects will skimp on audits. The bottom line is changing: auto-audits will become the default action once they get cheap enough. Crypto projects used to ask if they had been audited; soon they'll be asking if they have ongoing audits.
OpenRouter dropped Fusion on June 12, don’t just see it as an API update. As the gap between single models narrows, the next layer of competition will shift to routing, comparisons, costs, and latency. Apps don't necessarily rely on a single model, but rather allow a set of models to back each other up in the background.
I saw someone in the square scoop up 13 Seekers all at once; don't just think of it as grabbing airdrops. This thing is more like a wallet, identity, and dApp Store in the form of a smartphone. Crypto is about to go mainstream; the entry point might not just be a website, it could be the hardware in your pocket.
AI chips are easily spun into a GPU narrative. But going below 2nm, High-NA EUV, yield rates, masks, and production cycles will all factor into the cost. On the demand side, it can ignite with just a word, while the supply side is a machine, a round of validation, and a process that slowly climbs. The real tough spot is in the wafer fab.
A lot of folks are looking at x402, and it's easy to mistake it for a payment protocol. I'm more interested in which layer it's stuck at: HTTP, API, MCP, tools, data—they could all turn into billing interfaces when agents are running tasks. If an agent is really going to work, they can't just know how to use tools; they also need to be able to authorize and settle within the boundaries. This is where crypto comes into play.
Google dropped ARD on June 17, and it's not just another protocol name. ⠀ It tackles the toughest three questions before an agent enters the enterprise: What skills do they have, who to use, and can they be trusted? ⠀ More models will emerge, but entry points will dwindle. Those who can access the workflow, obtain permissions, and gather data are closer to the default calls.
Don't just see HBM as a GPU accessory. ⠀ The more the Agent resembles a workflow, the more apparent the memory wall becomes. The GPU is crunching numbers, but storage determines whether it can keep crunching continuously.
DCD relayed from Bloomberg: Meta is locking in a contract with Crusoe for 1.6GW of hashing power. ⠀ The glaring aspect of this number isn't how many GPUs were bought, but rather that the metric of AI infrastructure has shifted from GPUs to GW. ⠀ In the future, whoever can deliver power, land, cooling, and clusters will have the model ceiling.
Don't treat eval like a scorecard. ⠀ What the B side ultimately checks is: whether the trade was executed, if mistakes can be reviewed, and if costs were managed properly.
Don't just focus on GPUs for AI infrastructure. ⠀ The more cards you stack, the more you worry about the network lagging behind. ⠀ The value of optical modules, silicon optics, and CPO isn't just about price hikes; it's about reducing wait times.
A lot of folks think about RWA and immediately picture moving stocks, bonds, funds, and notes onto the chain. ⠀ That statement isn't wrong, but it's a bit too in-your-face. ⠀ The real threshold isn't about whether assets are on-chain, but who can create a low-cost, frictionless, and easily settled access pathway for those assets. ⠀ Saw that quote from CZ in the square today, saying that BNB Chain is aiming to be faster, cheaper, and offer more privacy controls. ⠀ This isn't just your average on-chain performance hype. ⠀ It actually connects three key things: AI agents needing to pay, RWA requiring an entry point, and stablecoins needing settlements.
When many people hear AI memory, they think it just remembers what you like. ⠀ This understanding is too superficial. ⠀ Once it's truly in production, memory feels more like three ledgers: user ledger, task ledger, execution ledger. ⠀ The user ledger records who you are, the task ledger tracks how far along the task is, and the execution ledger explains why each step is taken. ⠀ Without these three layers, an agent is like a smart but forgetful intern: quick to respond, but unstable in execution; if interrupted, it starts over, and if it makes a mistake, there's no way to trace who caused it. ⠀ So now, OpenAI Agents SDK talks about sandbox, snapshot, and rehydration; Temporal discusses durable execution; LangGraph mentions checkpoint; and Mem0 refers to procedural memory.
AI security isn’t just about preventing it from speaking out of turn. ⠀ What’s even trickier is when it has the right permissions but takes the wrong actions. ⠀ Recently, Arcade rolled out a secure action layer, and Zscaler expanded Zero Trust towards agentic AI; essentially, they’re talking about the same issue. ⠀ An agent isn’t just a chatbot anymore. It can read web pages, check emails, run code, tweak APIs, connect to databases, and may even modify files, place orders, or send messages on your behalf. ⠀ So the risks aren’t limited to the model layer. ⠀ The real danger lies in the tool layer and permission layer: what it can access, when it must pause to consult a human, which steps require auditing, and which steps must be sandboxed. ⠀ In the past, companies managed accounts and people. ⠀ In the future, they’ll also need to manage a bunch of “non-human employees”: they don’t sleep, they operate at lightning speed, and their permissions might even exceed those of newcomers. ⠀ Once MCP, tool calling, and agent browsers become mainstream, the security industry won’t just have a new marketing buzzword; it will have a new control plane. ⠀ In the future, the most valuable aspect of AI security won’t be just telling you if a response is toxic. ⠀ It’ll be about controlling whether it can execute, how it executes, and if it can be held accountable when it goes wrong. ⠀ Responses may be cheap, but permissions are costly.
Log in to explore more content
Join global crypto users on Binance Square
⚡️ Get latest and useful information about crypto.