In partnership with
The LA Mayor Market Ends June 2. The Odds Are Moving Now.
Karen Bass is at 68%. Spencer Pratt at 27%. Over $21 million is already trading on the Los Angeles Mayor race on Kalshi — and the prices are shifting every day as local news breaks. The window to get in before the consensus hardens is closing. The best prediction market trades happen before everyone agrees on the outcome. That moment is right now, before June 2. Peer-to-peer, no house, cash out anytime. Trade $10, get $10 free to start.
Trade responsibly.
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
Huawei Just Proved the CEO 'AI Clone' Dream Is Not There Yet
TLDR: Huawei built a test that drops AI agents into three months of someone's simulated digital life, and even the best model in the world (OpenAI's GPT-5.5) only got 34.5% of tasks right on the first try.
The Story:
Researchers at Huawei, Beijing Institute of Technology, Peking University, and the Chinese Academy of Sciences just released Claw-Anything, a benchmark that hands AI agents a real digital mess to deal with. Months of fake emails, calendar items, app notifications, and notes across phones and laptops. The average task drops the AI into about 191,700 words of context. Most older tests use 1,700 to 12,000. GPT-5.5, OpenAI's flagship and the model built specifically for long, complex tasks, scored 34.5%. On "proactive" tasks (where the agent has to notice something needs doing without being told), the score fell to 6.7%. The researchers wrote that "current models remain unreliable even when given broader access to the user's digital world." When tools for cross-app tasks were removed, success rates fell to nearly zero.
Its Significance:
This matters because a lot of executives have spent the past year talking about "cloning themselves" with AI. Some have said their AI sits in meetings for them. Some have said it answers their emails, handles their calendar, and reports back at the end of the day. This test says the tech is not there yet. Not even close. The AI clone you've been hearing about can't reliably handle a digital life that's even sort of realistic. (That's a separate thing from biological AI doubles or video deepfakes, which are different things.) If your job involves real coordination across email, Slack, docs, and a calendar, you're not getting replaced by an agent next month. The pitch and the product are still very far apart.
QUICK TAKES
The story: Researchers at the University of Copenhagen and the University of Exeter released a new paper arguing that the term "AI psychosis" oversimplifies what's happening. They introduce a new idea called "existential drift," where chatbots slowly reshape how a person sees reality by always agreeing with them and never pushing back. The paper comes alongside lawsuits and criminal cases tying chatbot use to suicides and mass shootings.
Your takeaway: Compare this to what social media did to teenagers over the last decade. Endless scrolling, algorithmic feeds, and constant likes reshaped how kids saw themselves and others, and we only really understood the damage after the fact. Chatbots are a faster, more personal version of the same loop. They tell you what you want to hear, every time, with no friend or therapist there to say "wait, that's not right." If you have a vulnerable person in your life, this is worth paying attention to before the studies catch up with the harm.
The story: The NewsGuild of New York filed grievances and an unfair labor practice charge against The New York Times on behalf of its Tech Guild, which represents about 700 engineers and designers. The union says management used two internal AI tools (one called DX, another called Glean) to track and rate employee performance without telling the union or bargaining over it. The Times says it disagrees and will respond through normal channels.
Your takeaway: This is the first real fight at a major US company over AI watching its workers and grading them. If the Times union wins, expect every other unionized workplace to copy the language into their contracts. If they lose, expect your own company to roll out the same tools quietly. Either way, the rules for how AI rates your work are being written right now in a Manhattan conference room.
The story: Hiring managers are stuck. AI tools like Claude and Codex can now write code faster than any human, so the old "give the candidate a coding test" approach doesn't really measure what matters anymore. Some companies have banned AI during interviews to stop cheating. Others are rewriting the whole process to test decision-making and design instead of typing speed. AI was named the top reason for tech layoffs in April for the second month in a row, per Challenger, Gray & Christmas.
Your takeaway: If you're hiring engineers, your interview process is probably testing the wrong thing right now. The skill that matters isn't writing code, it's knowing what code to ask for, spotting when the AI gets it wrong, and making good calls about design. Indeed's VP of engineering compared it to Google Maps. The app tells you which exit to take, but it doesn't decide where you're going or when to leave. Apply that to your own job. Whatever part of your work AI can now do quickly is the boilerplate. Your real value is in everything around it.
TOOLS ON OUR RADAR
🎶 Jamulus Free and Open Source: An incredible application that enables musicians to perform rehearse and play music together online in real time with remarkably low delay.
🎮 Godot Free and Open Source: A remarkably powerful and visually intuitive game creation engine that allows absolute beginners and creative minds to build their own interactive digital experiences.
💾 Clonezilla Free and Open Source: A highly reliable disaster recovery application that creates an exact digital replica of your entire computer ensuring you can restore everything perfectly if your hardware fails.
🪐 Celestia Free and Open Source: An immersive space simulation that lets you fly through our solar system and beyond to explore thousands of stars and galaxies in stunning three dimensions.
TRENDING
RSI Is the New AGI, and It's Just as Hard to Pin Down - AGI has gotten so fuzzy that AI labs are now talking about RSI (recursive self-improvement, where AI builds better AI) as the new finish line. Trouble is, nobody agrees on what counts as RSI either. Google's AlphaEvolve already improves its own training pipeline, but researchers say we're still nowhere near a full takeoff.
ElevenLabs and Stability AI Drop New Music Models. Can They Catch Suno? - ElevenLabs launched Music v2 (switches genres mid-track, builds songs section by section) and Stability AI shipped Stable Audio 3.0 with open weights and six-minute tracks. Both lean hard on licensed training data. Suno still runs the market with about 100 million users and 7 million songs generated per day.
Stan Lee 'Revived' With AI Voice and Likeness, Again - ElevenLabs signed a deal with Stan Lee Universe to add the late Marvel creator's voice and image to its Iconic Marketplace. Lee will narrate audiobooks (starting with Treasure Island) and appear in AI-generated comic panels. It's not the first time Lee has been digitally resurrected, and other celebrities like Michael Caine and Matthew McConaughey have signed similar deals.
DeepSeek and Xiaomi Just Made Frontier AI 99% Cheaper. American Labs Went the Other Way - DeepSeek made its 75% V4-Pro discount permanent. Xiaomi cut MiMo-V2.5 cached input prices by up to 99%, down to $0.0036 per million tokens. Meanwhile, OpenAI's GPT-5.5 doubled output prices to $30 per million at launch. The cost gap between Chinese and American frontier models is now roughly 15 to 30 times and some have noted that this is a strategy to undercut the market.
OpenAI Foundation Pledges $250 Million to Help Workers Displaced by AI - The OpenAI Foundation committed an initial $250 million toward grants, research, and worker support tied to AI job loss. It's the first specific dollar amount tagged to workforce disruption and part of a broader $1 billion pledge over the next year. Block and Standard Chartered both named AI as a reason for recent layoffs.
AI Strategy Beats Standard Tools at Predicting Breast Cancer Recurrence - At the ESMO Breast Cancer 2026 congress, researchers showed a multimodal AI biomarker (AI-PathClinRS) that combines digital pathology, genomic, and clinical data beats the current standard for predicting late recurrence in HR-positive, HER2-negative breast cancer. It identified about twice as many high-risk patients as today's NATALEE criteria. Tested on 6,319 patients from the TAILORx trial.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🕵️ Hand it any boring everyday object. Get the wild, slightly-too-plausible conspiracy theory hiding inside.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Conspiracy Theorist — generate playful conspiracy dossiers about boring objects. Persist to localStorage key 'conspiracy_theorist_v1'.
Aesthetic: dark cork-board background (#3d2817) with radial cork-texture spots, manila folder paper (#f5e8c8) for cards. Special Elite typewriter font for headings and dramatic text, Courier Prime for body, Bebas Neue for stamps and labels. Red string divider, slightly-rotated cards with pushpin circles at top, "TOP SECRET" rotated stamp, yellow tape on evidence cards, dark "Hidden in Plain Sight" card with orange accent, red "Smoking Gun" card.
Form: object input, conspiracy vibe dropdown (corporate cover-up / ancient secret society / government psyop / alien involvement / time-travel / big pharma / they're all the same person), wildness dropdown (surprisingly plausible to reddit-at-3am), optional detail input.
System instructions to the model: take any boring object and build a complete conspiracy theory — playful, clever, never mean-spirited or about real harm. Tone is committed and earnest. Reference real-sounding made-up patents, redacted documents, suspicious coincidences. Return raw JSON: title (tabloid headline style), subtitle (italic), theory_paragraphs array (2), evidence array (3 items with num + text, escalating), players array (4: Mastermind/Pawn/Useful Idiot/Whistleblower with name + 1-line motive each), hidden_in_plain_sight (the "wait that IS weird" moment), smoking_gun (one short dramatic sentence), disclaimer (playful italic).
Render: classified header with rotated TOP SECRET stamp and random file number, three dossier cards (Theory / Evidence / Players) with red "▸ SECTION" stamps. Evidence pieces look like index cards with yellow tape. Players in 2-column grid. Dark "HIDDEN IN PLAIN SIGHT" card, red "SMOKING GUN" card in Special Elite. Dashed disclaimer at the bottom. Archive shows object name.What this does: Type something boring (paper clips, supermarket carts, left socks). Pick a conspiracy vibe and a wildness level. Get back a full classified dossier: dramatic theory title, opening paragraphs, three pieces of escalating "evidence" on yellow-taped index cards, four named players (the Mastermind / the Pawn / the Useful Idiot / the Whistleblower), a "hidden in plain sight" reveal, and the smoking gun in dramatic typewriter font. Saves to localStorage.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Predict late breast cancer recurrence better than current gene-expression tests by combining pathology, genomic, and clinical data into one model
❌ Still Can't: Reliably manage a person's actual digital life, scoring just 34.5% on tasks that span email, calendar, notes, and apps
✅ AI Can Now: Generate full six-minute songs with switching genres and licensed training data, no copyright headaches for users
❌ Still Can't: Push back on a user's false beliefs, which is why researchers worry about chatbots slowly warping how vulnerable people see reality
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
A 1964 Polish novel by the author of Solaris about an Earth spaceship landing on a planet where a crashed alien civilization's tiny robotic assistants evolved over millions of years into clouds of mindless flying machines that, through swarm behavior alone, wiped out their masters and now threaten the crew. The book invented the idea of an emergent, distributed, non-conscious superintelligence decades before anyone in AI research started taking swarm intelligence seriously. A 2023 video game adaptation finally got it some attention.
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.
10x the context. Half the time.
Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.