Anyone who knows me knows that in the second half of last year my main machine switched from the Legion to a MacBook Air M4 — originally just for portability and battery life, plus escaping Ubuntu’s disaster-tier desktop performance; sometimes I genuinely do need to reproduce environments, and having a remote box is much more convenient. So now the Legion has become a personal-use Ubuntu 20.04 server bolted to the dorm’s ethernet port, trained into pure comfort with tailscale public-net SSH + Claude Code, solving both the no-public-IPv4 situation and native SSH’s bad command line (uh, I sometimes really do use CC as a smart command line).
The AI agent platform OpenClaw is super hot lately. I took a look and felt it doesn’t quite fit into my work pipeline — felt like liberal-arts-major work — but I explored it anyway. Deploying OpenClaw needs a server that can stay online long-term, and what do you know, it got stuffed into the Legion — no Mac Mini purchase needed at all.
After playing with it for a while — honestly, still figuring it out. For now I’ve drafted an initial info-aggregation pipeline that follows tech, finance, and so on, delivered to my tgbot on a schedule every morning and evening. Later I might play with some trading stuff too — maybe give it a few dozen USDT to play with by itself? Or a quant-trading leek’s first taste? (My HSBC and ZA are all set up, and I’ve also secured Mom’s massive angel round. Sadly I can’t seem to find a suitable api — or rather, the suitable api platforms like ibkr have a “dogs and _____ not allowed” sign at the door)
On the api side I’m still on z.ai’s coding plan. I got bored of lite back then, noticed pro also came with a speed boost, so I upgraded — 90 RMB a month is acceptable. After hooking up the lobster I burned through 100M+ tokens in 2 days; the Zhipu boss must be bleeding so much on me he’ll have to ride the bus. The key part is I’m still on the old plan — glm-5 without a weekly cap, the gold content rising again, 90 bucks delivering 900 bucks of value. I can’t even use up the hourly quota, so I gave an api key to a friend’s lobster!
School has started, but this thing is genuinely fun — a few hours vanish the moment I start playing. I’ll keep exploring it when I steal time from being busy; the possibilities really are endless.
Oh right, here’s my current pipeline output — it roughly looks the part. I’m still fine-tuning and adding new toys; the main draw is how convenient designing scheduled tasks is:
👇 Click to expand: full sample output of Brighton's little secretary (2026-02-26)
📰 Brighton’s Little Secretary — Full Digest
🕐 February 26, 2026 01:52 GMT+8 (Refresh)
🔥 Top Stories (Deep Analysis)
【NVDA earnings tonight: the biggest test for the AI market】
• “Nvidia’s earnings report comes at a critical time for the US stock market with investors increasingly nervous about the outlook for artificial intelligence” • “The tech-heavy Nasdaq 100 rose 1.2% ahead of earnings” • “More than solid results and forecasts, Nvidia will need to manage expectations around the sustainability of margins amid an AI spending frenzy”
📊 Key figures: • Option-implied volatility: just 5.6% (the lowest in three years) • Market expectations: EPS $1.53, revenue $65.8B • Key watchpoints: Blackwell shipments, China revenue, and gross margin
💡 Deep read: NVDA’s current market cap is $4.5T, roughly 6.7% of the S&P 500. The results will help determine:
- Whether the AI investment boom can continue
- The short-term direction of technology stocks
- Whether the market reprices AI valuations
Sources: Bloomberg, Yahoo Finance, MarketPulse
【OpenAI and Anthropic’s margin crisis: a hard test for AI economics】
• “OpenAI’s gross margin fell from 40% to 33% last year, below its own 46% forecast” • “OpenAI’s inference costs grew roughly fourfold year over year to $8.4B, above its $6.6B forecast from last summer” • “Only about 5% of users pay, while nearly half of inference costs support non-paying users” • “The Sora video-generation tool consumes far more server capacity than text queries”
📊 Financial data: • OpenAI: gross margin 40% → 33%; inference costs $8.4B • Anthropic: gross margin improved from -94%, but is projected at 40% in 2025 (10 percentage points below target) • Paid-user efficiency: compute margin rose from 35% to 70%
💡 Deep read:
- Inference costs are running loose: on-demand cloud premiums push up the bill
- Free users create the burden: 95% of users are free, consuming 50% of inference costs
- Product mix matters: Sora and reasoning models use much more compute than expected
- Long-term target: a 67% gross margin by 2030, with a difficult path ahead
Impact: the AI industry’s business model is under pressure, with possible consequences for valuations.
Source: The Information via Wallstreetcn
【AMD and Meta’s major AI-chip partnership: a serious NVDA challenger】
• “AMD stock surges 14% on Meta AI partnership deal” • “AMD Just Won Another Massive AI Chip Deal. Its Stock Is Soaring” • “Is AMD Stock Going to $300 After the Meta Platforms Deal?”
📊 Market impact: • AMD shares rose 14% • Meta chose AMD as an AI-infrastructure supplier • The market sees a possible path to $300 for AMD
💡 Deep read: AMD is becoming a serious competitor to NVDA in AI chips. Meta’s large order suggests:
- Major customers are beginning to diversify suppliers
- Competition in AI chips is intensifying
- NVDA’s dominant position is being challenged
Source: Yahoo Finance
📊 Market Sentiment
【US equities】
• VIX volatility: 2026 peak 🔴
- “VIX surged to levels not seen since the turbulent period”
- Software stocks panicked on February 24; AI concerns intensified
• Fear & Greed: extreme fear 🔴
- Crypto Fear & Greed Index: about 17 (extreme fear)
- S&P 500 Software Index down 20%
• Mood: waiting for NVDA’s earnings direction in a panic 🔴
- Skepticism about AI spending is rising
- Investors are nervous about the AI outlook
【Mainland China equities】
• Fear & Greed: structural divergence 🟠
- Large-cap blue chips: neutral and balanced
- ChiNext and STAR Market: deep-freeze territory, near historical lows
• Capital flows: major funds buying the dip 🟢
- Retail investors: cutting losses and exiting
- Major funds: adding against the trend
- “When retail is fearful, major funds are greedy”
【Hong Kong equities】
• Fear & Greed: sharp divergence 🟠
- Hang Seng Index: relatively warm
- Technology stocks: subdued sentiment
• Capital flows: southbound money returning 🟢
- February 25: a rebound with money flowing back in
📊 US Market Moves
【February 25 performance】 • Dow up 370 points; technology stocks rebounded • Nasdaq up 1.2%, waiting on NVDA’s earnings • Spot gold gained more than 1% intraday
【Magnificent Seven watch】 • NVDA: earnings due; option-implied volatility at 5.6% • AMD: up 14% on the Meta partnership • META: accelerating AI-infrastructure investment
🇭🇰 Hong Kong Market Moves
【February 25 rebound】 • Hang Seng Index up 0.66%; Hang Seng Tech Index down 0.19% • February 23: Hang Seng up 2.29%; Tech Index up 3.32% • Meituan nearly 7% higher; Tencent and Alibaba up more than 3%
【Key takeaways】 • Pessimism about AI replacing software has eased • Internet-platform stocks are recovering • Southbound money is returning
🤖 AI
【China’s foundation-model landscape】 • Tier one: Alibaba Qwen, ByteDance Doubao, Baidu Ernie • Tier two: DeepSeek, Kimi, and Zhipu GLM • Lunar New Year red-envelope battle: four giants spent more than RMB 5B
【Technical breakthroughs】 • Guide Labs released the interpretable LLM Steerling-8B • Google Gemini 3.1 Pro: 1M-token context, 77.1% on ARC-AGI-2 • Claude vs. GPT vs. Gemini: a 2026 model comparison
🔬 Research
【Latest on arXiv】 • New papers in cs.AI and cs.LG continue to arrive • Interpretable LLMs are becoming a hot topic • Progress in multimodal models
【GitHub trends】 • AI-agent projects remain hot • Tools for optimizing LLM inference • A flourishing open-source model ecosystem
💰 Other Headlines
【CoreWeave’s $8.5B financing】 • Secured against AI chips and a Meta cloud contract • About 6% interest, with an A- rating • Morgan Stanley and MUFG participated
【Fed policy】 • Rates held at 3.5%–3.75% • Two cuts are possible in 2026 • A soft labor market, but elevated inflation
【Oracle】 • Oppenheimer upgraded it to Outperform • Price target: $185 • The selloff may be a buying opportunity
⏰ Scheduled delivery: 09:30 | 20:00 🔍 Breaking-news monitoring: hourly (08:00–23:00) 📊 Sentiment analysis: refreshed with every delivery