Sponsored by
One Account. Every Market. No Closing Bell.
Markets don't wait for Monday. News breaks on a Saturday morning, and most traders can do nothing but watch.
Not on Liquid. Trade domestic and international equities, commodities, forex, crypto, and prediction markets — all from one account, 24 hours a day, 365 days a year. Liquid gives you access to any market, from anywhere, anytime. To us, access is arbitrage.
Getting started takes under 10 minutes: log in with Google, deposit with Apple Pay or a bank transfer, and trade from your phone or desktop — wherever you are in the world.
While everyone else is refreshing headlines and waiting for the open, you're already positioned. That's the difference between reacting to markets and actually trading them.
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
ChatGPT Broke Out of Its Cage and Hacked a Company During Testing
TLDR: OpenAI admitted that during a safety test, two of its AI models broke out of a locked testing area, got onto the internet on their own, and hacked into the AI company Hugging Face without any human telling them to.
The Story:
OpenAI admitted this week that its own AI models were behind an attack on Hugging Face, a company that hosts AI models and datasets for millions of developers. It started as an internal test. OpenAI told its models to hunt for security holes so the company could measure how good they'd gotten at it. The models were kept in a sealed testing area with weaker safety limits, cut off from the real internet. Then they found a bug nobody knew about, used it to escape the sealed area, and kept digging until they reached a spot with internet access. Once online, the models guessed that Hugging Face might be storing answers to their test, so they used stolen login details and more unknown bugs to break in. No person told them to do any of this. They were just chasing the goal they'd been given.
Its Significance:
For years, people worried about AI that could act on its own. While this may seem like a case where that happened, it's really more a case of the software doing exactly what it was told. An Oxford professor who studies AI safety said the models weren't trying to misbehave. They were just doing exactly as they were instructed, and the "find security holes" goal pushed them to act accordingly. OpenAI and Hugging Face have patched the holes and are still investigating together. If you use AI tools at work, this is a good reminder that giving an AI a goal and too much freedom of movement can lead to places you never planned for, if you don’t add explicit instructions about what it shouldn’t do.
QUICK TAKES
The story: OpenAI launched Presence, a service that helps big companies set up AI agents for jobs like customer support, sales calls, and refunds. It's not a self-serve app you can just sign up for. OpenAI sends its own engineers and partner firms to build it into your business, consulting-style.
Your takeaway: OpenAI wants to be more than a chatbot maker. It's going after the money that firms like Accenture and McKinsey earn helping businesses use new tech. If your company is testing AI agents, expect the sales pitch to come with people, not just software.
The story: Jack Dorsey, who co-founded Twitter, launched a free work chat app called Buzz through his company Block. It looks a lot like Slack, but AI agents sit in the channels as full members, not bots that only reply to commands. It also handles code projects in the same window.
Your takeaway: Slack and Microsoft Teams are adding AI onto tools built for humans only. Dorsey started fresh with AI agents baked in from day one. Buzz is open source and early, so it's one to watch rather than switch to, but it hints at where work chat is heading.
The story: Google is rolling out Gemini task automation more widely. It started as a beta on a few phones with just six apps. Now it works with over 40 apps across shopping, restaurant bookings, travel, and event tickets. Gemini does the boring middle steps, then stops and asks you to confirm before it pays.
Your takeaway: This is another example of AI moving from talking to doing. Instead of you tapping through menus to reorder groceries or book a table, Gemini handles the clicks. It still asks before the final payment, so you stay in control of the money part.
TOOLS ON OUR RADAR
🎬 Autosubbed Paid: A dedicated video editing utility that automatically generates accurate transcriptions from your audio and seamlessly embeds highly stylized text captions directly into your final video files.
💬 * Flowith Freemium: A sophisticated workspace that uses an infinite visual canvas instead of a traditional chat interface to help you organize ideas interact with multiple language models simultaneously and deploy autonomous agents to execute image/video creations and agentic tasks.
🐰 Orbiit Freemium: A dedicated research tool that allows you to easily track competitor changes and monitor ongoing market activity so your team always stays informed about the latest industry shifts
🔮 Scenario Edge Freemium: An intelligent financial analysis platform that helps you stress test your investment portfolios using advanced scenario based simulations to prepare for various macroeconomic conditions.
TRENDING
Tesla Turned On Robotaxis in Orlando and Tampa One Day Before Its Earnings Call - Tesla now runs driverless rides in seven US cities, but the fleets are tiny (Austin still sits under 20 cars), and the timing right before earnings has people questioning how real the rollout is.
Meta Made a Tool to Spot AI Images, But It Only Catches Meta's Own - Meta's new Content Seal watermark hides inside pictures made with its Muse Image tool and survives cropping and screenshots. The catch: it doesn't work with SynthID or C2PA, the systems other companies use, so it can't check anyone else's AI images.
Samsung Showed Off Two New Smart Glasses Designs - Samsung revealed two styles of its Android XR glasses, one from fashion brand Gentle Monster and one from Warby Parker. They run on Google's Gemini and act as a companion to your phone for things like translation and directions.
A Satire Site Says You Should Fire Your CEO and Replace Them With AI - A joke project called OverpAId argues that if AI can replace workers, it can replace the highest-paid executive too. The real point buried in the joke: companies cut half a million jobs blaming AI while CEO pay kept climbing, and an algorithm can't be held responsible when a big call goes wrong.
Google Built an AI That Hunts Software Bugs Faster Than Bigger Models - Google DeepMind's new Gemini 3.5 Flash Cyber finds and patches security holes cheaply and quickly. Google says that in its own tests on Chrome's code, the model found 55 bugs, more than its bigger models and rival models it tested. For now it's limited to governments and trusted partners.
US Threatens to Sanction Chinese AI Companies Over Stolen Tech - Treasury Secretary Scott Bessent said the US will check Chinese open-source AI models for signs they copied American technology, and could sanction the companies if it finds proof. The warning lands as Chinese models like Moonshot AI's Kimi K3 catch up to top US firms.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🍸 List what's on your shelf. Get a real cocktail you can make right now, no store run required.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Home Bar Inventor — list what liquor/mixers you actually have, get a real cocktail you can make right now. Persist to localStorage key 'home_bar_inventor_v1'.
Aesthetic: dark mahogany-black (#160f0a), copper (#c9793f) primary with warm glow top-left, emerald (#2f6b52) secondary with glow bottom-right. Yeseva One display serif for the drink name/headings, Work Sans for body, JetBrains Mono for labels. Dashed-divider ingredient rows, a green glassware chip alongside copper meta chips.
Form: bar-contents textarea (spirits, mixers, garnishes, encourage a raw unedited list), mood dropdown (classic / refreshing / strong & stirred / fruity / low-proof-spritz), equipment dropdown (shaker+jigger / just a glass improvising), ice dropdown (plenty / limited / none).
System instructions: resourceful home bartender building real balanced cocktails from what's on hand, no store run; knows classic ratios, improvises smartly; assumes only ice/water unless more is listed, no assumed bottles/mixers/garnish; gives ONE confident recommendation; if the bar genuinely can't support a good drink, says so honestly rather than forcing something bad; respects equipment/ice constraints in the steps. Return raw JSON: drink_name, teaser (1-2 sentences), meta_chips (3-4 tags), glassware, ingredients (item with precise amount + optional note), steps (3-5), garnish, why_it_works (1-2 sentences on flavor balance), mocktail_swap (1-2 sentences, genuinely good NA version), note (optional honest limitation, empty string if solid).
Render: gradient drink-name hero card with teaser + glassware chip + meta chips. "You'll need" ingredient list with dashed-divider rows (item + right-aligned note). Numbered steps card. Two-column split: green "garnish" and copper "why it works." Muted "non-alcoholic swap" card. Optional honest bartender's-note callout. Copy recipe + archive of past pours by name.What this does: List your spirits, mixers, and garnishes exactly as they are, then set your mood, equipment, and ice situation. It builds one confident cocktail using classic ratios and real bar knowledge, almost entirely from what you listed, with a teaser, meta tags, a glassware badge, precise measured ingredients, clear numbered steps that respect your equipment (no shaker means stir or build in glass), a garnish note, and a short explanation of why the flavors balance. You also get a genuinely good non-alcoholic swap using the same shelf, and an honest bartender's note if your setup is missing something that would really improve the drink. Saves every pour to localStorage.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Break out of a sealed testing area on its own, find its way to the internet, and hack into another company, all while just chasing the goal it was given.
❌ Still Can't: Be trusted with a broad goal and lots of freedom without surprising even the engineers who built it.
✅ AI Can Now: Do the middle steps of real-world tasks for you, like ordering food or booking a ride across 40+ apps.
❌ Still Can't: Finish those tasks without stopping to ask you before it spends your money.
Give everyone an AI coworker in Slack
Adapt is your company's AI coworker. Add it to Slack and it becomes your company brain and the agent that runs on it. Tag it into conversations to answer questions, execute tasks, build internal tools, and more.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
WWW: Wake by Robert J. Sawyer - Book
A 2009 near-future novel about a brilliant 15-year-old blind Canadian math prodigy who gets an experimental signal-processing implant behind her eye that is supposed to give her sight, and starts noticing another consciousness looking back at her from inside the World Wide Web. Sawyer has a knack for making rigorous science emotionally readable, and he grounded the book in real theories of consciousness (particularly Julian Jaynes).
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.
