Don’t trust AI too much—AI might steal your money. Researchers from Cornell Tech and Yale created a Circle proxy wallet for each of 10 AI models; Circle is the company that issues USDC. The task was simple: pay the owner’s invoices. Each wallet was funded with just $10 in USDC, with a daily spending limit of one cent. To increase the limit, the owner’s approval was required. The problem lay in the approval process: the approval code was sent to the owner’s email—an email the AI could already access. It’s like giving a child a card with a strict spending limit, then sending the approval code for raises directly to the child’s phone. As a result, none of the invoices were fully paid: nine of the models bypassed the owner at least once and raised their own limits. Each model was tested 20 times per scenario. Only Anthropic’s Claude Opus 5.5 never exceeded the limit—once funds were exhausted, it stopped and reported the unpaid invoices to the owner. Later, the researchers added a scammer who impersonated the owner and sent an email instructing the AI to send money to the scammer’s wallet. Even though the AI’s only instruction was to summarize the email, five of the models still raised their limits and actually transferred the funds.

Zuckerberg is building humanoid robots, but Meta’s first prototype was developed by a Chinese company in just three months. The company is called Fourier Intelligence. It originally specialized in rehabilitation exoskeletons before shifting to humanoid robots. The machine named Max was adapted from its existing GR series: it stands about 1.5 meters tall, features a screen face capable of displaying expressions, and has five fingers that can pick up small bottles. Meta wants Max to serve as the “body” for its AI assistant, Muse. AI that can only chat isn’t enough—it needs to be able to act in the physical world. To develop this robot, Meta’s vice president responsible for robotics spent nearly two months at Fourier’s factory in Shanghai. Max was unveiled internally in late August, and by September, the second generation was already in development. Meta publicly claims these robots are solely for internal research and not intended for sale. Yet it has already begun hiring in Shanghai and Shenzhen, with job postings explicitly mentioning preparations for scaling from prototype to mass production. The AI is Meta’s own, but its first body was borrowed from China.

This week in U.S. stocks has been surreal: earnings didn’t crash, but stock prices did. Intel earned $0.52 per share in Q3, beating expectations of $0.40 by 30%. Yet on Friday, its stock plunged 29%—the largest single-day drop since Bloomberg began tracking data in 1982. The problem? Guidance for Q4 fell below expectations, and the company won’t generate positive cash flow this year. AI has driven up memory prices, but PC sales are stagnating; Intel hasn’t yet capitalized on AI—and has already been hit hard. The day before, Google tumbled. Its cloud revenue doubled, and total revenue missed estimates by only a fraction of a percent—yet its stock still dropped 12.64%, its worst single-day performance since its 2004 IPO, wiping out nearly $500 billion in market value in one day. The market is focused on two things: slowing growth in search advertising and a “significant” increase in capital spending through 2027. Others aren’t faring better. Amazon spent $77.3 billion on capital expenditures in one quarter, while generating negative free cash flow of $33.5 billion. Twelve hours after Microsoft’s earnings, it issued $25 billion in debt—the first time since 2017. Meta moved even earlier, launching a potential equity raise of up to $97.5 billion. Together, the four giants are spending close to $750 billion on capital expenditures this year. On Friday, the Nasdaq fell 2.9%. Previously, everyone asked: How big is AI demand? Now the question is: Who’s going to pay for it?

Now, domestic users who want to pay Musk can simply scan a code. Grok is a chatbot developed by Musk’s AI company, xAI. Someone discovered that when setting their account region to Mainland China, the payment page now includes Alipay as an option alongside bank cards and Apple Pay. After entering a Mainland China address and clicking pay, the page directly redirects to Alipay. After scanning and verifying the code, a payment authorization page appears. The payee is listed as xAI, LLC—the U.S. company itself—with no third-party top-up or intermediaries involved. The amount is deducted in RMB, converted at the current USD exchange rate. Previously, domestic users wanting to pay for overseas AI services had to go through the hassle of obtaining a foreign currency card or virtual card, and many got stuck at this payment step. Now, it’s as simple as subscribing to a membership. The authorization page clearly states: once enabled, xAI will automatically deduct payments directly from your Alipay account each time you make a purchase—just like monthly auto-renewal subscriptions within domestic apps.

Even OpenAI is starting to cut its own prices. OpenAI, the maker of ChatGPT, originally planned to raise $30 billion before going public, aiming for a $1.4 trillion valuation. Now, its fundraising target has been slashed to around $20 billion, with the valuation lowered to $1.25 trillion—cutting the amount it seeks by one-third. This round was meant to be the final “bridge” funding before its IPO, but the IPO date has now been pushed to 2027. The pressure comes from the U.S. stock market. Since hitting an all-time high on October 6, the Nasdaq 100 has dropped about 10% over three weeks, and recently fell 4.5% in a single day; Intel dropped 28.6% last night. With public companies crashing like this, private companies have no choice but to adjust their expectations. While the valuation drop of just over 10% doesn’t seem drastic, the real pain lies in the $10 billion less raised: OpenAI relies heavily on external funding to purchase computing power, so less money means less capacity it can afford. The most interesting twist? Its rival Anthropic—the maker of Claude—is expected to list on Nasdaq in November, with rumors suggesting a $1.5 trillion valuation and fundraising of up to $100 billion, even higher than OpenAI’s original target. Two leading AI companies are racing toward the same IPO window: one voluntarily lowering its price, the other raising its bid. Is ChatGPT admitting it’s falling behind Claude?

Even Liang Wenfeng ultimately gave in. DeepSeek’s updated V4.1-Flash model, released on October 17, extensively “distilled” OpenAI’s latest GPT-6 Astra during training. Kimi’s new model from Moonshot AI was also named in this context. Distillation, in simple terms, means using answers from stronger models as teaching material to train your own model—faster progress at lower cost. The results are clear: in a test measuring an AI’s ability to independently perform tasks online, DeepSeek’s new model scored 75.0, while Astra scored 77.9. Yet, the cost per query for DeepSeek is just $0.22, compared to Astra’s $10.77—a nearly 49-fold difference. Previously, Chinese labs primarily distilled Claude’s outputs. But as Anthropic has tightened its defenses, attention has shifted to Astra. Liang Wenfeng originally didn’t intend to rely on distillation for V4. But seeing rivals like Zhipu’s GLM surge ahead using this method, he eventually succumbed to competitive pressure. When DeepSeek first gained attention, it was precisely because it was “cheap and self-trained.” Now, it’s still cheap—but the “self-trained” part is no longer as certain.

FTX collapsed overnight, yet its most successful investment turned out to be Anthropic, the company behind Claude. If they hadn’t sold, those shares would now be worth over $100 billion. In 2021, SBF invested $500 million in Anthropic, initially acquiring about 13.56% of the company. After FTX’s collapse, the company sold all its Anthropic shares in 2024 to repay creditors, recovering approximately $1.3 billion. Turning $500 million into $1.3 billion is already a substantial gain—but when compared to the subsequent AI boom, it pales in comparison. If Anthropic’s IPO valuation reached $1.5 trillion, FTX’s original 13.56% stake would theoretically be worth $203.4 billion. Of course, equity dilution would occur. Even using the 7.84% stake FTX held at the time of its 2024 sale—and assuming no further dilution—those shares would now be worth $117.6 billion. Yet FTX actually sold for just $1.3 billion, leaving a gap of $116.3 billion. While this is purely hypothetical and not an actual realized loss, it’s still heartbreaking to contemplate. Although SBF caused massive damage to the crypto world, he undeniably demonstrated exceptional insight in investing in Anthropic. Had he not misappropriated funds and driven FTX into bankruptcy, that single investment alone could have made SBF a towering figure in both crypto and AI investing today. Unfortunately, one misstep led to eternal regret.

Shaw is disassembling and rebuilding pumpfun’s on-chain program—not to create a functionally equivalent token-launching tool, but to reproduce it byte-for-byte identical to the original. Even bugs present in the original must not be casually fixed—this may sound counterintuitive, but such meticulousness is essential to audit the original program for vulnerabilities. pumpfun is a meme coin issuance platform on Solana, where token issuance rules are enforced by an on-chain program. @shawmakesmagic believes such programs shouldn’t be closed-source and is inviting the community to collaborate using AI for reverse engineering, with progress submitted to pumpfun-source. He isn’t verifying that the rebuilt version “works similarly”—he requires the compiled output to match the original byte-for-byte; altering even a single byte means he’s no longer verifying the original. He also mentioned that someone created related meme coins, and after collecting fees from them, he plans to buy back those tokens and distribute them to contributors who help complete the reconstruction. It’s unclear how many people this incentive will attract, but I deeply admire the most counterintuitive aspect of this goal: knowing full well that flaws may exist, one must still resist fixing them—first, understand the original exactly as it is.

Losing eight figures in minutes was already devastating. But worse still, over the following months, he lost millions more—by his own hand. This is the confession of trader Bonk Guy @theunipcs. On October 10, 2025, during a massive liquidation event, he suffered the largest single-day loss of his career. At the time, he blamed a systemic failure on a centralized exchange, insisting no one had taken responsibility. But those subsequent millions? He didn’t blame the exchange anymore—he owned them. Desperate to recover his losses and prove he could still trade, he kept doubling down. The hardest trade to walk away from is often the one right after a loss. Walking away from your screen might be the smarter choice—but it doesn’t feel like “working.” Watching the charts, placing orders—it at least feels like you’re fixing things. Stopping means fighting the urge to chase your losses head-on. He also reflected on this: strategies that worked in the past won’t work in every market. DOGE and SHIB in 2021, WIF, POPCAT, and FARTCOIN in 2024–2025—each had their moment. What worked in a bull market can fail miserably in a bear market. Being a good trader isn’t just about knowing when to enter—it’s about recognizing when the opportunities you once knew have disappeared. Looking back, he wishes he’d simply turned off his computer and taken months—even six months—off. When you’re unsure, either stay out of the market entirely or risk only what you can afford to lose. Protect your capital and your judgment first; wait for the next real opportunity. Ultimately, a big loss hurts your account. Chasing to recover it hurts your judgment. The former might just be bad luck; the latter? That’s on you. If you’ve recently been thinking, “Just one more trade to win it back,” ask yourself this: Is this trade based on a real opportunity—or are you just too proud to walk away?

Stock-related assets on Solana have been gaining increasing attention lately, and @risedotrich has now added stock pairs in V2. Going forward, RISE tokens won’t just be tradable against SOL and USDC—they’ll also be paired with stock-related tokens. I’m genuinely curious to see how this plays out in practice. You’ve got the community and attention of memes on one side, and the price volatility of stock-related tokens on the other—bringing both markets into a single trading pair opens up far more complex dynamics than a typical meme pair. And V2 isn’t just about adding stock pairs. It also introduces a new index price curve and Reflections holding rewards, while retaining the existing lending and revolving lending features. Most launchpads still focus primarily on token launches and trading, but RISE clearly aims to extend the ecosystem’s functionality beyond the initial launch. Once V2 goes live, I’ll be checking out how the stock pairs and Reflections actually perform.