0x99DaDa

Nvidia, together with OpenShell, has built a security fence around AI Agents at the hardware level—keeping Agents running inside and minimizing the chance they jailbreak and break out of the security boundary.
It breaks down into two layers:
1. The software layer locks down files, networks, and credentials by default, injecting them only after approval;
2. The hardware-side monitor adds another layer of oversight, without relying on the Agent to report whether it has crossed the boundary.
#英伟达 #OpenShell

I recently saw an old piece of gossip about Wahaha, and I was honestly stunned,
Zong Fuli really seems hooked on playing the “leading lady,” insisting on overturning her father’s arrangements, and leaving a total mess in her wake~
The old man’s plan was:
The legitimate daughter would take charge and inherit the family business.
The younger son and daughter born outside the marriage would also have a safety net through a trust.
But no one expected the eldest daughter to refuse and insist on turning the table over.
I wonder if the old man, if he knew what was happening from beyond the grave, wou

The hottest crypto investment fund, a16z Crypto, is dropping “Crypto” from its name. Let’s look back at the projects they’ve invested in and see how many you know and how many you’ve participated in.
(1) Coinbase: Added to its position in multiple early rounds → cashed out when it went public in 2021—their most successful investment narrative
(2) Ripple: Dixon said their first crypto investment was in 2013; the equity story is still playing out, while XRP’s price action is another matter
(3) Uniswap: Led the Series A; the DEX flagship is still around. UNI has fallen by half from its peak, leav

JPMorgan Asset Management has assigned Bitcoin a long-term return estimate for the first time:
Over the next 10 to 15 years, the base-case annualized return is approximately 10%, higher than the 7.5% it assigns to US large-cap stocks and 5.5% for gold.
This is the 2027 edition of its annual “Long-Term Capital Market Assumptions.”
It has been published for 31 years and covers more than 200 asset classes. Many institutions’ investment committees use it as a reference when planning long-term allocations.
Bitcoin was newly added in this edition. Other figures in the same table include:
🔹US large-

I have a friend who worked for a female boss for many years. She was only three or four years older than him and had been divorced for quite some time, while he was still unmarried.
One evening, he drove his boss home. When they arrived downstairs, she got out and asked him not to leave yet, but to wait a while. After some time, she came downstairs carrying a cardboard box. From the outer packaging, it appeared to contain pillows.
The boss said, “Someone gave me a pair of pillows. They’re direct-selling products that are hard to buy elsewhere. I’ll give them to you.” After saying that, she ope

昨天摩根大通把 base:0x62c1ba684061c449b4676445668b4705044878fc 搬到了ETH链了，
看来以太坊还是传统金融大佬最新人的公链呀，
利好 ethereum:native
去年6月份 base:0x62c1ba684061c449b4676445668b4705044878fc 发行在 Base 公链上，做公链概念验证，
一年多跑下来，看起来不错，所以这次正式搬到了 ETH 链上（ BASE 链还保留）。
毕竟还是 ETH 的用户量大，信任度更高。
不过 base:0x62c1ba684061c449b4676445668b4705044878fc 是给经过许可的机构用的，普通人交易不了。别看到消息去乱交易，被同名的 MEME 币割了韭菜。

Victoria's Secret 2026 Fashion Show (Live Stream)
Would anyone be interested in watching?

I feel like my OpenAI has gotten dumber.
I'm using 6.1 SOL, but it keeps getting the question wrong.
I repeatedly tell it not to do something,
but it insists on doing it anyway, and when I point it out, it just says Sorry, I'll change.
Fuck, it's driving me insane……

The founder of #HEX and #PulseChain , Richard Heart, was arrested by Turkish police in Istanbul early this morning.
Turkish media said he was the subject of an Interpol Red Notice issued by Finland. He and his wife were taken away, and about $600k in cash, 24 luxury watches, 7 guns, and 20 graphics cards were found at their home.
After the news broke, HEX fell by about 30% in 24 hours.
Interestingly, in 2023, the U.S. SEC sued him, alleging that he had raised more than $1 billion and misappropriated investors’ money to buy sports cars and black diamonds. But the court dismissed the case last

🪲 Anthropic’s red team says: small models may be only a few months away from becoming worms on their own.
On October 12, they published a Frontier Red Team report, whose title, translated literally, is “AI models may be able to self-replicate without a data center.”
The experiment roughly went as follows:
🔹 bought 15 off-brand Android TV boxes (about $55 each, with 4GB of memory apiece), split the open-source model Gemma4-26B across the boxes, and ran it at about 0.39 tokens per second
🔹 tested “whether an inference service could be set up on unfamiliar hardware”: models in the 9B range suc

🎲 Broadcom CEO Chen Fuyang said in Singapore: Many AI investors are acting like they are in a casino.
On October 12, he discussed AI infrastructure at DBS’s Asia Future Forward event. His point was: Many people actually don’t know what they are investing in, their expectations are extremely high, and they ultimately find it is not much different from gambling in a casino—especially when it comes to frontier large models.
He also ran through a few figures:
🔹 Based on a rough estimate using OpenRouter’s inference throughput, the world spends around $200 billion a year on “spitting out tokens”

🏌️‍♀️ Epic has put a blockbuster Marvel × Star Wars game on hold and switched to using AI to write code and create art.
Epic is the company behind Fortnite and Unreal Engine. In 2024, Disney invested $1.5 billion in Epic, and the two sides teamed up to make a multiplayer shooter featuring Marvel and Star Wars characters together, originally slated to launch in March 2027.
Bloomberg reported on October 12 that the game has been delayed, with no new date set:
🔹 Progress is slower than expected; Epic underestimated how many people and how much time it would need
🔹 Last week, the game director

🖼️ Microsoft plans to have its own chips handle most AI inference, rolling them out at a scale of “several gigawatts” starting next year.
The AI features in Copilot and Teams have mainly run on NVIDIA GPUs. Microsoft began developing its own AI chips in 2023, called Maia, and launched the second-generation Maia 200 this January.
Nadella posted tonight that this is Microsoft’s “biggest bet ever on in-house chips”: starting in 2027, it will expand Maia to multi-gigawatt-scale deployments, making it Microsoft’s primary inference system for its own products and services. It also unveiled the

Intel CEO Lip-Bu Tan publicly praised Chinese large models today: tokens are cheaper, performance is “good enough,” and people are using them “very quickly.”
He said this at a forum hosted by DBS Bank in Singapore, specifically naming DeepSeek, Kimi, and Zhipu’s GLM. He also said he “strongly believes in open source—it’s the best way to improve efficiency,” and that the market should decide which model is suitable for which scenario.
There is data to back this up. OpenRouter is a platform that connects hundreds of models through a single interface, with developers paying based on usage. It pub