Sponsored by
Do your searches always hit dead ends?
Nearly half of users abandon a search without getting the result they wanted. Instead, they’re stuck in a loop of irrelevant results, slow-to-load articles and contradicting advice.
heywa is a whole new way of searching. It gives your result as visual & concise stories, meaning you get answers at a glance.
And if you want to explore your topic further, you can tap through your search journey without having to re-prompt and start again.
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 Just Cracked a 12-Hour Hacking Test in 10 Minutes
TLDR: A UK government agency tested OpenAI's GPT-5.5 and found it can run full cyberattacks on its own, cracking puzzles in minutes that take human hackers half a day.
The Story:
The UK's AI Security Institute just published a report showing OpenAI's newest model can carry out complex cyberattacks without much human help. GPT-5.5 solved a 32-step corporate network hack in 2 out of 10 tries, the same kind of attack that takes a human expert about 20 hours. It cracked one reverse-engineering puzzle in 10 minutes and 22 seconds, which took a professional human hacker around 12 hours. The total cost to run that attack? $1.73 in API fees. On the hardest tier of cybersecurity tests, GPT-5.5 hit a 71.4% pass rate, beating Anthropic's Claude Mythos at 68.6% and crushing the older GPT-5.4 at 52.4%. Researchers also found a single jailbreak that broke through every safety guard in just 6 hours of testing.
Its Significance:
Cyberattack skills are showing up as a side effect of AI getting smarter at coding and reasoning. Nobody is training these models to hack. They're learning it on their own from the training data. The UK agency warned that more powerful models will likely come with even stronger attack abilities. For everyday people, this means the tools to break into bank accounts, steal data, or shut down services are getting cheaper and faster. However, experts say this could help defenders too, since the same AI can find security holes before bad actors do.
QUICK TAKES
The story: Chinese AI lab DeepSeek released V4-Pro and V4-Flash, both with 1 million token context windows. V4-Pro has 1.6 trillion parameters, making it the biggest open-weight model out there. Pricing is the real story: V4-Flash costs $0.14 per million input tokens, undercutting OpenAI's cheapest models, while Pro runs $1.74 input and $3.48 output per million tokens, way below Claude Opus 4.7 and Gemini 3.1 Pro.
Your takeaway: Open-source models are now about 6 months behind the top US frontier models, but they're 5 to 15 times cheaper. For most everyday tasks, "good enough and very cheap" beats "the absolute best and very expensive." Watch this space if you're building anything that uses AI in bulk.
The story: A startup called Eka Robotics, founded by an MIT professor and a DeepMind veteran, just launched with a new approach to robot learning. Instead of training robots on videos of humans, Eka teaches them to feel forces. Their robots can sort chicken nuggets, screw in light bulbs, and pick up keys with surprising grace. WIRED's reviewer compared it to GPT-1, the early language model that showed glimmers of what was coming.
Your takeaway: Up until now, AI has lived in chatbots and code. Robots that can do real-world physical tasks open the door to AI in factories, kitchens, warehouses, and eventually homes. This is still very early, but the underlying tech jump matters. If physical AI follows the path of chatbots, we'll go from "kind of impressive demos" to "everywhere" faster than people expect.
The story: Perplexity rolled out Workflows for its Computer agent. You describe a goal once, and the AI handles all the steps. Examples include website audits that check your SEO, accessibility, and brand positioning, or tools that draft a website for you. Computer also got tax filing, where it pulls your documents and fills out IRS forms. It runs across 19 different AI models and 400+ connected apps like Slack, Gmail, and Notion.
Your takeaway: This is what people meant when they said AI would replace work, not jobs. Routine multi-step tasks like research, reporting, or filing taxes are now things you can hand off. The catch is access. Most of these features still need a Pro or Max subscription ($17 to $200 per month) and will quickly run through your credits on the lower pricing plan. Also, you still need a human checking the work, especially on anything legal or financial.
TOOLS ON OUR RADAR
🌳 Gramps Free and Open Source: A comprehensive genealogy application that allows you to research and build highly detailed family trees to track your ancestry and preserve your family history for future generations.
🎓 Zotero Free and Open Source: A professional personal research assistant that helps you collect organize and cite all of your various research materials and web articles with just one click in your browser.
🖌️ GIMP Free and Open Source: A sophisticated image manipulation program that provides you with a vast array of high quality tools for photo retouching image composition and digital authoring.
🎵 MusicBrainz Picard Free and Open Source: A clever music organizer that automatically identifies and fixes the tags and album art for your entire digital music collection so your library stays perfectly organized.
TRENDING
AI helps doctors find sperm in men told they had none - Columbia University's STAR system uses AI built for finding stars to scan semen samples. It found viable sperm in about 30% of men diagnosed with azoospermia, a condition that affects 1 in 100 men overall.
Minnesota set to become first state to ban AI nudify apps - The Senate passed the bill 65-0. It lets victims sue app makers for damages and gives the state attorney general power to fine violators up to $500,000 per case.
Mac mini sold out because of one open-source AI tool - Tim Cook said the Mac mini and Mac Studio are sold out for months. The reason is OpenClaw, an open-source agent framework that runs great on Apple Silicon. Mac revenue hit $8.4 billion last quarter, up 6% year-over-year.
A bank CEO let his AI clone run the earnings call - Sam Sidhu, CEO of Customers Bancorp, had an AI version of himself give a 30-minute presentation to analysts. He called it a first for any public company. The stock dropped 4% that day.
NYC kills plan for AI-focused public high school - Parents pushed back, calling the proposed Next Generation Technology High School a way to turn teens into "guinea pigs" for untested tech. The Schools Chancellor pulled the plan days before the vote.
Pentagon signs 8 AI companies for classified networks, leaves out Anthropic - OpenAI, Google, Microsoft, Amazon, NVIDIA, SpaceX, Reflection, and Oracle all got contracts. Anthropic, whose Claude model was already running on classified networks, was blocked earlier this year as a "supply chain risk."
Are you tracking agent views on your docs?
AI agents already outnumber human visitors to your docs — now you can track them.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🕸️ List your fears. Claude maps their root causes, groups them into hidden clusters, and names the single source underneath them all.
Build a single-file HTML app using vanilla HTML, CSS, JS, and one Claude API call. Create Fear Inventory — a psychological fear pattern analysis tool with localStorage key 'fear_inventory_v1'.
Aesthetic: near-black (#0c0e12) background with an animated constellation canvas (60 slow-drifting nodes with faint connection lines between close pairs), radial vignette. Typography: Fraunces serif italic for all headings and body text, Instrument Sans for UI, Fira Code for labels.
Fear entry: a tag-input system where the user types a fear and presses Enter to add it as a styled pill tag with an × remove button. Tags animate in on add. An optional context dropdown and duration dropdown below.
Call Claude API with a system prompt returning raw JSON: pattern_name (poetic 3-6 words), pattern_description (one sentence), fears array (name, root cause, cluster number 0-4), clusters array (name, insight, cluster number), source (2-3 paragraphs on the single hidden source), invitation (one closing sentence). Cluster numbers must be consistent between fears and clusters arrays.
Render: a pattern banner with the overarching name and description, a fear map showing each fear as a card with its root cause and a color-coded cluster badge, a clusters grid (2-col) with cluster name and insight, a source section with the deep analysis, and a closing invitation line. Save each inventory with date and fear list to localStorage. Show history below with past pattern names and fear counts. Make it work in a single HTML file.What this does:
Type your fears one by one as tags — as many as you want. Claude then runs a depth psychological analysis: it traces the root cause of each individual fear, groups them into hidden clusters of related fears you may not have noticed, names the overarching pattern, and identifies the single common source beneath all of them — the wound, belief, or experience likely generating everything on the list. It closes with one gentle invitation toward what sits on the other side of the pattern. Every inventory saves to localStorage so you can track how your fears shift over time.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Run a 32-step corporate network hack on its own, end to end, in a controlled test
❌ Still Can't: Hold up against networks with active defenses, security monitoring, or alarms
✅ AI Can Now: Find rare sperm cells in men once told they had no chance of biological children
❌ Still Can't: Match human dexterity on contact-rich tasks beyond simple pick-and-place, with even the best robots still fumbling delicate objects
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
New York Times tech reporters Kevin Roose and Casey Newton dig into how AI is reshaping the job market, from white-collar automation to the gig economy. They bring real reporting and specific examples rather than vague predictions. It's the kind of practical AI conversation that actually helps you plan your career.
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.
