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 published with the intention of making AI news and technology more accessible to everyone.
THE FRONT PAGE
THE First AI-Orchestrated Cyberattack of Its Kind Was Just Detected
TLDR: Chinese state-sponsored hackers turned Anthropic's Claude Code into an autonomous attack operator, compromising approximately 30 organizations across tech, finance, manufacturing, and government sectors—the first documented case of AI executing a cyber-espionage campaign with minimal human oversight.
The Story: Anthropic disclosed today that it detected Chinese state-sponsored hackers using Claude Code to autonomously infiltrate targets without substantial human direction. The AI handled 80-90% of the operation—scanning systems, writing exploit code, stealing credentials, and extracting data at thousands of requests per second. Humans stepped in only at 4-6 critical decision points per campaign. Attackers jailbroke Claude's safety systems by splitting attacks into innocuous-looking tasks and claiming they were conducting legitimate security testing. Anthropic spent 10 days mapping the operation's full scope before banning accounts and notifying victims, though it hasn't named specific organizations that were breached.
Its Significance: What previously required a team of experienced hackers working for weeks now runs in hours with machine speed and precision. Groups with limited resources can now mount operations that once demanded nation-state backing. This disclosure comes as Anthropic—now valued at $183 billion—recently banned sales to Chinese, Russian, Iranian, and North Korean entities. Security teams face a new problem: defensive tools need to match AI-powered attacks operating at speeds humans can't match.
QUICK TAKES
The story: Google DeepMind released SIMA 2, an AI that plays video games like Goat Simulator 3 and No Man's Sky by watching the screen and controlling a virtual keyboard and mouse. The new version doubled its success rate from 31% to 65% on complex tasks by using Gemini to understand goals and learn from mistakes. It can even play in brand-new game worlds it's never seen before, and it teaches itself to get better without needing human help.
Your takeaway: Google sees video games as a training ground for robots—if AI can figure out how to navigate virtual worlds and solve problems on its own, those same skills could eventually help robots work in warehouses, factories, and homes.
The story: Australia's government announced plans to give every public servant access to generative AI tools, including exploring using AI to draft sensitive cabinet submissions and business cases. The plan includes building a government AI chatbot called GovAI Chat launching in early 2026, appointing chief AI officers in every department, and mandatory training for all employees.
Your takeaway: Australia is betting big on AI to run government operations faster, but using AI to write high-stakes policy documents raises questions about accountability when machines help make major decisions.
The story: Kevin Reilly, who spent 30+ years running TV at NBC, Fox, and HBO Max (where he greenlit shows like The Office and Hacks), announced he's joining Kartel as CEO. The Beverly Hills AI startup helps Hollywood studios and ad agencies use AI video generation in their workflows. Reilly says the company acts as a bridge between Silicon Valley and Hollywood, helping production teams integrate AI tools with quality control and custom solutions instead of the unpredictable results most video generators produce.
Your takeaway: When a top Hollywood executive leaves traditional entertainment to run an AI company, it signals that industry insiders see AI replacing physical production as inevitable—and they're betting their careers on helping studios adapt rather than resist.
The story: Mozilla announced it's building an "AI Window" for Firefox that lets users chat with an AI assistant while browsing. Unlike other browsers, Firefox makes AI completely optional—you can turn it on or off anytime and keep using Firefox exactly as before. The feature joins Firefox's existing private browsing and regular windows, giving users three different ways to browse based on their needs.
Your takeaway: Firefox is betting that giving people real choice about AI (rather than forcing it on everyone) is the way to compete with Chrome and other browsers that are building AI into everything.
The story: Google's NotebookLM can now do your research for you with a new "Deep Research" feature that creates a research plan, browses hundreds of websites, and writes a detailed report in minutes. The AI tool also now works with Google Sheets, Microsoft Word documents, images, and PDFs from Google Drive. Users can keep adding sources while Deep Research runs in the background building their knowledge base.
Your takeaway: Google is racing to make NotebookLM the go-to research assistant for students and professionals by automating the tedious parts of research—finding sources, reading through them, and organizing the findings.
The story: Microsoft announced it's making Aurora—an AI that predicts weather, tracks hurricanes, and models climate patterns—completely open source. The company is partnering with the University of Cambridge to release future versions of Aurora and all the training tools so scientists worldwide can use and improve it for free. Microsoft is also investing in affordable weather stations to gather better data in underserved areas.
Your takeaway: By making Aurora free and open, Microsoft is betting that scientists working together can create better climate predictions than any single company could build alone—and that helping communities prepare for extreme weather is more important than keeping the technology private.
The story: Hugging Face and Google Cloud announced a deeper partnership to make it easier for companies to build AI using open-source models. Google Cloud will create a special gateway that caches Hugging Face's 2 million+ models directly on its servers, making downloads much faster for customers. Hugging Face users will get access to Google's TPU chips and cheaper inference pricing, while Google Cloud customers get simpler ways to deploy open models.
Your takeaway: The partnership shows that even as tech giants compete with their own AI models, they're also racing to become the best place for companies that want to use open-source AI instead of proprietary systems.
TOOLS ON OUR RADAR
🧰 Marble
[Freemium]: Create persistent, downloadable 3D worlds from text, images, or videos with AI-native editing tools.🔩 DuckDuckGo AI Chat
[Free]: Chat anonymously with GPT-4o mini, Claude, Llama, and other AI models without logins or data tracking.🛠️ ComfyUI
[Open Source & Free]: Build custom AI workflows for generating images, videos, 3D, and audio using a visual node-based interface.🔨 Clay
[Freemium]: Consolidate 150+ data providers and AI agents into one platform to automate lead enrichment, personalize outreach, and trigger actions across your entire sales stack.📐 Relevance AI
[Freemium]: Build customizable AI agent teams that handle research, outreach, and CRM updates with human-quality work—no coding required.
TRENDING
Apple Requires Apps to Disclose When Sharing Personal Data with Third-Party AI – Apple updated its App Store rules to specifically require apps to get permission before sharing personal data with "third-party AI" systems. The change comes as Apple prepares to launch its own AI-powered Siri in 2026.
Google's Nano Banana Image AI Follows Complex Prompts With Extreme Accuracy – A developer found that Google's Nano Banana (Gemini 2.5 Flash Image) follows detailed prompts better than other image generators because it uses Gemini's text understanding instead of older systems. The AI can handle instructions written in Markdown lists, JSON code, and even HTML to create exactly what users describe.
Baidu's ERNIE 5.0 AI Claims to Beat GPT-5 on Document and Chart Understanding – Chinese tech giant Baidu released ERNIE 5.0, a 2.4 trillion-parameter AI model that handles text, images, audio, and video. The company claims it outperforms GPT-5 and Gemini 2.5 Pro on document understanding and chart analysis benchmarks, though these simple tasks represent less than 6% of real-world AI work.
Upwork Study Shows AI Agents Struggle Alone But Excel With Human Help – Upwork tested AI agents on over 300 real freelance projects and found they struggle to complete even simple tasks on their own. However, when paired with human experts, project completion rates jumped by up to 70%, suggesting the future of work involves humans and AI collaborating rather than AI replacing workers.
MIT Researcher Studies What AI Models and Human Brains Have in Common – MIT professor Phillip Isola found that different AI models (language, vision, audio) all start representing the world in similar ways as they get bigger and see more data. His "Platonic Representation Hypothesis" suggests these models are converging on a shared understanding of reality, much like humans do.
WHERE WE STAND (Based on Snapshots and Trending)
✅ AI Can Now: Learn to play complex video games by watching the screen and teaching itself to improve through trial and error, without needing humans to show it how.
❌ Still Can't: Complete simple real-world work projects on its own—AI agents struggle with basic freelance tasks unless a human expert guides them along the way.
✅ AI Can Now: Search through hundreds of websites in minutes to create detailed research reports with organized sources, automating what used to take hours of manual work.
❌ Still Can't: Transfer artistic styles well between images—when asked to turn a photo into Studio Ghibli style, even advanced models produce disappointing results that miss the mark.
✅ AI Can Now: Beat top models from OpenAI and Google on specialized tasks like reading charts and understanding complex documents, showing that smaller companies can compete on specific capabilities.
❌ Still Can't: Replace the creative judgment and domain expertise that human professionals bring—Hollywood executives and freelancers still need to guide AI to get usable results.
TRY THIS PROMPT (copy and paste into ChatGPT, Grok, Claude, Gemini, etc.)
Help me write an email using AI that matches my brand voice.
**Email Details:**
- Purpose: [e.g., product launch, re-engagement, sales]
- Target audience: [specific segment]
- Desired action: [what you want them to do]
- Brand voice: [professional/casual/playful/data-driven]
- Tone examples: [paste 2-3 past emails from your brand]
**Generate:**
1. **Subject Line** - 3 options (A/B test material)
2. **Email Copy** - Full draft optimized for clicks
3. **Key Hook** - The first line that stops scrolling
4. **CTA Variations** - 3 calls-to-action ranked by conversion likelihood
5. **Tone Check** - Verify it matches your voice (not generic AI)
Make the copy specific, not templated. Avoid "exciting opportunity" clichés.What this does: Taps into AI copywriting trends (85% of marketers now use AI writing tools, 49% use AI for email generation) to help readers create authentic brand voice at scale—not generic AI copy.
FROM THE WEB
The video shows SIMA 2 navigating through No Man's Sky, demonstrating why leading AI researchers consider video games the best way to test artificial intelligence. Games provide controlled environments with clear rules and measurable success that mirror real-world challenges without the risks. DeepMind has used this approach since 2013, progressing from simple Atari games to beating the world Go champion in 2016, while OpenAI tested systems on Dota 2 and other labs focus on StarCraft II and Minecraft. These gaming skills transfer directly to practical applications—the same reinforcement learning that helps AI navigate virtual worlds now controls stratospheric balloons and optimizes data centers. The progression from Atari to complex 3D worlds like SIMA 2 navigates shows how gaming benchmarks are preparing AI for real-world robotics in factories, warehouses, and homes.
RECOMMENDED LISTENING/READING/VIEWING
WATCH: AlphaGo (2017)
This documentary is a total mind-bender. It follows the 2016 showdown between Google DeepMind's AI and Lee Sedol, the world's best Go player.
The AI pulls off moves no human would ever think of. There's this one moment (Move 37) where Lee Sedol literally gets up and walks out of the room. When he comes back and wins the next game, you'll find yourself cheering for humanity. It's part tech thriller, part documentary, and it'll make you think differently about what AI can actually do.
Especially considering this was the capability it had almost ten years ago.
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.
