$181 Million for a Pollock. Is the Art Market Sending Signals?
Christie's May 18 auction was headlined by a $181M Jackson Pollock — the fourth highest price ever. A Rothko also went for $95.4M. A Brancusi fetched $107.6M (second highest ever for sculpture).
Those masterpieces were obvious outliers, but by the end of the evening, $1.1B in art had been sold.
Pollock died in 1956. The number of paintings he left behind is limited. Same for Rothko, Warhol, Basquiat. When one of these works trades at the top of the market, the remaining supply gets thinner and more contested.
That scarcity is what Masterworks was largely built around. Their acquisition committee — former specialists from Sotheby's and Christie's — uses over 3 million data points to identify which pieces to buy.
29 exits have delivered net annualized returns like 16.5%, 17.6%, and 17.8% on those held longer than a year, not including those unsold.
To join Masterworks, my subscribers skip the waitlist by clicking this unique link.
*Investing involves risk. Past performance is not indicative of future returns. See important Reg A disclosures at masterworks.com/cd.
Beginners in AI
Good morning and thank you for joining us again!
Welcome to this daily edition of Beginners in AI, where we explore the latest trends, tools, and news in the world of AI and the tech that surrounds it. Like all editions, this is human curated and edited, and published with the intention of making AI news and technology more accessible to everyone.
THE FRONT PAGE
Robinhood's New AI Agent Buys Concert Tickets and Trades Stocks for You

TLDR: Robinhood is letting AI agents trade stocks and swipe a virtual credit card on your behalf, opening up a kind of automated finance that used to be reserved for hedge funds.
The Story:
Robinhood announced on Wednesday that customers can now hand their stock trading and shopping over to AI agents. You set up a separate trading account, drop in some cash, and connect an AI agent that can read your portfolio, build strategies, and place orders without asking you each time. There's also a new virtual credit card tied to Robinhood Gold, which lets agents buy stuff for you. Think grabbing concert tickets the second they go on sale, or buying a product the moment the price dips below what you said you'd pay. Users can set monthly spending limits and require manual approval on every purchase, or let the agent run free. Robinhood plans to add options, crypto, futures, and prediction markets to the agent platform later. The company's own disclosures warn that AI agents can misread instructions, act on stale data, or behave in ways you didn't expect, and you could lose your whole investment.
Its Significance:
This is a big deal because it puts pro-level automation in the hands of regular people. Before now, only big firms with quants and trading desks could run software that watches markets all day and acts in milliseconds. Now anyone with a Robinhood account can do something similar. The flip side is real. An AI agent can be tricked. Researchers have shown that bad actors can plant hidden instructions in websites, emails, or even product listings that hijack an agent into spending or trading in ways the owner never wanted. A Deloitte survey found that agentic AI is already moving faster than companies can monitor. Crypto is heading in the same direction, with services like MoonPay Agents and AWS Bedrock AgentCore Payments giving AI software the ability to hold crypto wallets and pay other agents using stablecoins. If you try this, set tight limits and require approval on big moves until you trust how your agent behaves.
QUICK TAKES
The story: Chinese lab OpenBMB just released MiniCPM5-1B, a one-billion-parameter model that fits in about 500 megabytes and runs on a phone. It supports tool calling and Model Context Protocol out of the box, which means agents can do real work without ever calling a cloud server.
Your takeaway: On-device AI keeps your data on your device. No prompts sent to a third-party server, no monthly fees, no internet needed. As these small models get better, expect more of your daily AI tasks to happen privately on your phone instead of in someone's data center.
The story: UMG and TikTok just signed a new multi-year licensing deal that promises to find and remove AI-generated songs that copy artists without permission. The two had a public fight back in 2024 when UMG pulled its catalog from TikTok over AI and pay disputes, so this is a big shift in tone.
Your takeaway: The music industry is no longer treating AI knockoffs as a small annoyance. Expect more deals like this from Spotify, YouTube, and Meta, all built around permission, credit, and payment for the actual artists. If you're a creator using AI tools to remix music, the rules are tightening fast.
The story: YouTube is making its AI labels much more visible and will start adding them on its own when it detects "significant photorealistic AI use" in a video. The platform also recently opened its likeness-detection tool to all creators 18 and over, so people can spot videos using their face without permission.
Your takeaway: This makes it harder to pass off AI-generated content as real. Repeated failure to disclose can lead to forced labels you can't remove, plus strikes against your channel. If you make videos with AI tools, get used to checking the disclosure box.
TOOLS ON OUR RADAR
🎙️ AntennaPod Free and Open Source: A completely free and highly respectful podcast player for mobile devices that lets you discover new shows and download audio episodes without any advertisements.
📝 OnlyOffice Free and Open Source: A highly compatible and completely free office productivity suite containing everything you need to create text documents financial spreadsheets and stunning visual presentations safely offline.
🧠 TiddlyWiki Free and Open Source: An incredibly unique and highly customizable personal notebook that stores all of your written thoughts and digital ideas inside a single interactive web page file.
📺 Yattee Free and Open Source: A beautiful alternative media player for Apple devices that allows you to watch your favorite online videos completely free of annoying promotional advertisements and tracking algorithms.
TRENDING
Microsoft's MAI-Image-2.5 Lands at #3 on the Arena Image Leaderboard - Microsoft's new in-house image model can render text inside images more reliably and handles lighting, scale, and spatial relationships better than its predecessor. It trails only Google's Gemini 3.1 Flash and OpenAI's GPT-Image 1.5.
Sleep Startup Sond Exits Stealth With $7M From Bose's Former Head of Sleep - Sond is building "Dreambuds," AI-powered sleep earbuds that learn your patterns and adapt over time. The co-founder previously ran sleep products at Bose, sold his last startup to Bose, and holds a PhD from MIT.
StepFun's New Voice AI Tops Every Benchmark and Reads Your Sighs - Shanghai-based StepFun says its StepAudio 2.5 Realtime beats GPT Realtime 1.5 and Gemini Live on five voice tests, including reading vocal tone, speed, and emotion from your audio. Note that those are the company's own numbers, so independent testing is still pending.
China Now Requires Some AI Workers to Get Approval Before Traveling Abroad - Per a Bloomberg report, China is restricting overseas travel for senior AI researchers at firms including Alibaba and DeepSeek. Beijing is treating top AI talent as a national security asset, mirroring how the U.S. has tried to restrict chip exports.
GuideGeek Brings AI Travel Planning Into WhatsApp, Instagram, and Messenger - GuideGeek, made by Matador Network, lets you chat with an AI travel assistant inside apps you already use. It claims about 98% accuracy on suggestions thanks to human travel experts checking its work, and is now powering official tourism bots for places like New Brunswick, South Africa, and New Zealand.
New Research Says AI Probably Isn't Conscious — and the Old Tests Don't Work - Researchers from Bradford University and RIT applied math tests used to detect consciousness in the human brain to a GPT-style model. The signals showed up even when the AI was deliberately broken, which means the tools we have can't tell real awareness from clever mimicry.
Gecko Robotics CEO Says AI Should Catch Infrastructure Failures Before People Get Hurt - Jake Loosararian, CEO of Gecko Robotics, told CNBC his wall-climbing robots collect data on bridges, power plants, and Navy ships, then use AI to spot weaknesses before they break. The company has a Navy contract supporting an 80% fleet readiness goal by 2027.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
📖 Answer 6 questions. Become the subject of your own Wikipedia entry — complete with infobox, citations, and a Controversies section.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Wikipedia of You — generate a faux-Wikipedia article about the user. Persist to localStorage key 'wikipedya_v1'.
Aesthetic: pixel-perfect Wikipedia — off-white (#f6f6f6), Linux Libertine/Source Serif 4 for headings and serif text, Source Sans 3 for body, Wikipedia blue links (#0645ad), gray borders (#a2a9b1). Sidebar infobox with title bar, circular initials avatar, dotted dividers. Triple-double border under article title. Faux nav bar at top with Article/Talk/Edit/History tabs.
Form: 6 fields — name, what you're known for, places, eras, real stuff, controversies. Field help text under each label in italic. Blue "Publish Article" submit button.
System instructions to the model: write the entry in dry faux-encyclopedic Wikipedia voice. Take their life seriously while finding the absurd, the petty, the human. Make up plausible citations like [1] [2] throughout the prose. The Controversies section is essential and playful. Use their actual details. Return raw JSON: article_title, subtitle (italic), lead_paragraph (with bolded name and citation markers), sections array (exactly 5: Early life and education / Career / Notable beliefs and habits / Controversies / Legacy, each with paragraphs array), quote (text + attr), quote_section (which section to embed it in), references array (6-8 fake but plausible refs), categories array (4 items), infobox object (known_for, eras, affiliations, views, enemy).
Render: faux-Wikipedia article with infobox sidebar (initials avatar, 5 metadata rows), lead paragraph with bolded name and [N] superscript citation links, table of contents box, 5 sections with bottom-bordered serif h2s, embedded quote-box inside the matching section, references section with ^ N. prefixes, categories footer.What this does: Type your name, what you're known for, your formative places, your life eras, your real habits and beliefs, and your controversies. Get back a full Wikipedia-style article: title, italic subtitle, lead paragraph with bolded name and [1][2] citation markers, sidebar infobox with initials avatar and Known For / Eras / Affiliations / Notable Views / Sworn Enemy fields, five sections (Early Life / Career / Notable Beliefs and Habits / Controversies / Legacy), a faux quote inside a Controversies pull-box, fake references list, and Wikipedia categories. Saves to localStorage.
What this looks like:

Attio - the AI CRM for modern businesses.
Attio is the AI CRM that keeps you ten steps ahead.
Ask Attio anything. Where should I focus? What deals are at risk? Search, update, and create across your customer data.
Ask more from CRM. Ask Attio.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Trade stocks and make purchases on your credit card through dedicated agent accounts on Robinhood, within limits you set.
❌ Still Can't: Be trusted to ignore hidden instructions buried in websites, emails, or product listings. AI agents can still be tricked into doing things you didn't ask for.
✅ AI Can Now: Run agentic tasks like tool calling and Model Context Protocol on a smartphone, fully offline, with a 1-billion-parameter model that fits in 500MB.
❌ Still Can't: Match large frontier models on the hardest reasoning or knowledge tasks. Small on-device models still trade some capability for privacy and speed.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING

A young woman goes on a getaway weekend with her boyfriend and his friends and discovers, during the worst possible moment, that she isn't human at all but a leased AI companion whose personality settings her boyfriend has been adjusting from his phone. To say anything more would spoil the movie, but this was fairly well liked when it came out.
Thank you for reading. We’re all beginners in something. With that in mind, your questions and feedback are always welcome and I read every single email!
-James
By the way, this is the link if you liked the content and want to share with a friend.
Some * designated product links may be affiliate or referral links. As an Amazon Associate, I earn from qualifying purchases. This helps support the newsletter at no extra cost to you and Amazon makes a tiny hair less.





