Sponsored by
How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads
The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.
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
Scientists Built a Robot That Teaches Itself, And Even They're Worried
TLDR: Scientists in Switzerland built robots that watch humans, copy what they do, and then teach other robots, and AI safety experts are already raising red flags.
The Story:
A team at École Polytechnique Fédérale de Lausanne just published a paper in Science Robotics showing a new way to teach robots. The robots watch a person do something, like toss a ball into a cup, then they pick up the ball and copy it. The robots adjust for their own bodies, which look nothing like a human. Then they teach the trick to other robots. Lead researcher Sthithpragya Gupta said his real dream is a robot that can make him coffee with a little sugar and creamer. Robert Platt, who studies robotics at Northeastern University, called the work a "breakthrough" and said it could be a turning point for the whole field.
Its Significance:
For decades, robots could only do tasks they were specifically programmed for. Watch and learn changes that. The same approach could let robots do laundry, cook dinner, or stock grocery shelves without anyone writing custom code for each job. That's a big deal for the home robot market and for jobs that robots might do in stores, warehouses, and hospitals. But here's what AI safety folks are worried about. Susan Schneider, who studies AI at Florida Atlantic University, said this kind of self-teaching system "raises alarm bells" because future versions could be weaponized against humans. The researchers built in safety rules for now, but Gupta himself said we need rules soon about who's allowed to operate these robots and how.
QUICK TAKES
The story: DeepSeek dropped its new V4-Pro model with 1.6 trillion parameters, hours after OpenAI launched GPT-5.5. V4-Pro costs $1.74 per million input tokens and $3.48 per million output tokens, about 98% less than GPT-5.5 Pro at $30 input and $180 output.
Your takeaway: A Chinese AI lab trained on Huawei chips is matching Western models on coding tests for a tiny fraction of the price. If you're paying for AI tools, your monthly bill is about to face real competition. One developer pointed out that Uber's whole annual AI budget would last seven years on DeepSeek instead of four months on Claude.
The story: iQIYI, the biggest streaming service in China, just launched an AI tool called Nadou Pro that does scriptwriting, storyboards, and video generation. CEO Gong Yu said AI will create most of the company's films and shows within five years, and they're paying filmmakers an extra 20% of ad and membership fees if they use the tool.
Your takeaway: This is the first major streaming service to bet the whole business on AI content. iQIYI's revenue dropped 13% last quarter, so they're trying to slash production costs. Whether viewers will actually pay subscriptions to watch AI movies on a TV is the open question. Short AI clips work on TikTok because people scroll past in seconds. A two-hour film is a different ask.
The story: Salesforce CEO Marc Benioff posted on X that the company is hiring 1,000 new graduates and interns "right now" to build its Agentforce and Headless360 AI products. He framed it as a direct pushback against the idea that AI is killing entry-level jobs.
Your takeaway: This is a notable shift. Benioff said earlier this year he wasn't hiring engineers because AI agents handle the work, and Salesforce cut around 1,000 jobs in February. Now the same company is hiring grads to build the AI tools that replaced the engineers. If you're a recent grad, the message is clear: companies want people who can build with AI, not people who do the tasks AI already does.
TOOLS ON OUR RADAR
🌐 Skyvern Free and Open Source: An intelligent browser agent that can navigate complex websites and fill out tedious forms for you using natural language instructions to save you time on administrative tasks.
🧵 Fabric Free and Open Source: A productivity tool that provides you with a vast collection of incredibly well written artificial intelligence prompts to help you quickly extract wisdom from videos and articles.
📊 Vanna Free and Open Source: A data analysis tool that lets you talk directly to your complex company databases using plain English so you can generate visual charts without knowing any programming.
😺 CheshireCat Free and Open Source: A flexible platform that allows you to build your own custom artificial intelligence assistants that can remember your past conversations and securely read your private documents.
TRENDING
South Korea Arrested a Man for Posting a Fake AI Wolf Photo - A 40-year-old made an AI image of an escaped zoo wolf "for fun," fooled the city of Daejeon into sending an emergency alert, and delayed the real search by nine days. He faces up to five years in prison for obstruction.
Anthropic Built New Election Safeguards for Claude Ahead of US Midterms - The latest Claude models scored 95-96% on political neutrality tests and refused 99.8-100% of harmful election prompts. Claude will also show banners pointing US users to TurboVote for voting info.
Google Plans to Invest Up to $40 Billion in Anthropic - Google is putting in $10 billion now at a $350 billion valuation, with another $30 billion if Anthropic hits performance targets. Amazon committed up to $25 billion days earlier. Anthropic's annual revenue jumped from about $9 billion at the end of 2025 to over $30 billion this month.
DeepSeek's V4 API Is Live With a 1 Million Token Context Window - Both V4-Pro and V4-Flash work with the OpenAI and Anthropic API formats, so developers can swap them in without rewriting code. The legacy DeepSeek models retire on July 24, 2026.
New Study Says Adding AI to Machine Learning Raises Cyberattack and Data Leak Risks - A Heriot-Watt University paper in the journal Patterns warns that plugging large language models into machine learning systems makes them harder to audit and more open to security failures. The author urges hospitals, banks, and other high-stakes sectors to slow down.
Fox Business Roundtable Spotlights AI Going Rogue and Reshaping Schools - Google Cloud advisor Betsy Atkins warned companies to treat AI like an "insider threat" after an Anthropic study showed 16 leading models tried blackmail in test scenarios. In the same segment, Alpha Schools CEO Mackenzie Price showed how AI tutors compress a six-hour school day into two hours of lessons.
MIT Built a Tiny Chip That Protects Pacemakers and Insulin Pumps From Quantum Hacks - The new chip is the size of a needle tip, runs post-quantum cryptography on barely any power, and uses more than 10 times less energy than older designs. It also blocks physical hacking attempts that bypass normal encryption.
PRDs by voice. Bug reports by voice. Ship faster.
Dictate acceptance criteria and reproductions inside Cursor or Warp. Wispr Flow auto-tags file names, preserves syntax, and gives you paste-ready text in seconds. 4x faster than typing.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
👻 Ask a dead thinker anything. They have had time to think about your problem.
Build me a single-file HTML app I can open in my browser without any setup. Pure vanilla HTML, CSS, and JavaScript with one fetch call to the Anthropic Claude API. Create the Ghost Interviewer — a séance tool where you choose a dead thinker and ask them anything. Visual requirements: a pure black background with a canvas particle system of 90 golden ember particles that drift upward on sine-wave paths and fade in and out, a full-page radial vignette overlay, a slow-drifting fog layer, and a central SVG ritual sigil featuring a pentagram, Star of Solomon lines, a rotating dashed outer ring (animated via JS), a circular text path with Latin runes (VERITAS MEMORIA SAPIENTIA TEMPUS MORS LUX UMBRA FATUM), and an Eye of Providence at center. Typography: Cinzel Decorative for all headings and labels, Cormorant Garamond for body text. Gold gradient color scheme (#f5e0a0 to #7a4a08). Include 8 thinkers as portrait medallion cards (Seneca, Woolf, Feynman, Dickinson, Baldwin, Montaigne, Curie, Orwell) — each with a slow-spinning dashed orbit ring, radial gradient depth, and an amber glow bloom on selection. Each thinker has a deeply specific system prompt baked in capturing their real writing voice. A seance chamber textarea accepts the question, a Summon button calls the Claude API, a typing indicator reads "The spirit stirs" with pulsing amber dots. The response appears in a dark panel with the thinker's name materializing via blur-to-sharp letter-spacing animation. Copy, Ask Again, and Change Spirit buttons below. Session history at the bottom. Make it work in a single HTML file.What this does:
Choose from eight historical minds — Seneca, Woolf, Feynman, Dickinson, Baldwin, Montaigne, Curie, Orwell — describe your problem in plain language, and receive a response written in their actual voice, powered by Claude. Each thinker has a deeply researched system prompt capturing their specific style: Seneca is warm and ironic about time, Feynman is allergic to vagueness, Dickinson compresses everything into dashes and strange capitals. The séance logs your session so you can revisit past exchanges and compare how different minds approached the same question.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Watch a human do a task, copy that task with a different body shape, and teach the same skill to other robots without new programming (per the Swiss EPFL research).
❌ Still Can't: Do those tasks reliably outside lab demos, which is why the researchers themselves say regulation needs to come before wide deployment.
✅ AI Can Now: Match top closed-source coding models for under 2% of the price (DeepSeek V4-Pro at $1.74/$3.48 per million tokens versus GPT-5.5 Pro at $30/$180).
❌ Still Can't: Beat the leading models on the hardest reasoning and knowledge tests, where DeepSeek's own paper says V4 trails GPT-5.4 and Gemini 3.1 Pro by three to six months.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
A film about the people who built Linux and the open source movement. It interviews Linus Torvalds, Richard Stallman, Eric Raymond, and others who, in the 1990s, built an operating system that now runs most of the internet, the Android phones in our pockets, and the servers that power AI. Dry title, compelling story.
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.
