In partnership with

The AI coworker built for teams in Slack

The era of solo player AI is over. Adapt is the integrated coworker that empowers every team to work AI native together.

Anyone can tag @Adapt in Slack to answer a quick question, schedule a task, prepare for a board meeting, or build the dashboard you’ve been waiting on for weeks.

Built for business users, powerful enough for engineers.

Get $100 free credits for your team

^{Free credits awarded when you use your work email}

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

China's New Free AI Model Nears the Top, and Prices Could Drop

TLDR: A Chinese company called Moonshot AI released Kimi K3, a free-to-download AI model that comes close to the best paid models from companies like OpenAI and Anthropic.

The Story:

Kimi K3 has 2.8 trillion parameters, which are the tiny parts that help a model think. It can read up to one million words at once, and it handles images and video, not just text. On a test that measures real work across 44 different jobs, K3 scored third overall. It landed just behind Claude Fable 5 and GPT-5.6 Sol, and it beat most other models. Moonshot calls it the first open model in the 3 trillion parameter range, and the full files anyone can download are set to arrive on July 27.

Its Significance:

Until now, the very best AI models were locked inside a few big companies, and you had to pay them every time you used one. An open model works differently. Anyone can download it, run it on their own computers, and change it to fit their needs, all without paying a fee for each use. Building a top model can cost hundreds of millions of dollars, so only a handful of rich companies could make one. If free models keep getting this good, smaller businesses, schools, and startups can use top-tier AI without a giant budget. And when people have a strong free option, the big companies can't charge whatever they want. More choices tend to keep prices from climbing too high.

There may be another reason behind a giveaway this big. Cheaper and free models from China can pull customers away from U.S. companies and drag their prices down, so undercutting American AI may be part of the goal. Anthropic has also accused China of using distillation methods(asking an AI model many questions to reverse engineer it) on their models.

1-on-1 Live Claude AI Crash Course Tutorial

A 1-hour beginner-friendly video call to get you comfortable with the Claude ecosystem — Claude Desktop, Claude Code, Cowork, Skills, Projects, Correct File set-ups, and Plugins. Real-world example...

$75.00 usd

QUICK TAKES

Baseball Bans AI Help on Dugout iPads During Games

The story: Major League Baseball told teams they can no longer use AI tools on their dugout iPads to help make calls during games. As many as a third of teams had built custom apps that suggested things like pitching changes and player swaps.

Your takeaway: The league wants people, not machines, making the big calls on the field. It's an early example of a workplace drawing a clear line on where AI is allowed and where it isn't.

Google DeepMind Shares Its Plan to Fight Biological Threats With AI

The story: Google DeepMind and Isomorphic Labs shared a plan they call bioresilience. It uses AI to spot new disease outbreaks faster, design medicines and vaccines, and stop bad actors from misusing AI for biology.

Your takeaway: The same AI that helps scientists find cures could also help someone cause harm, so DeepMind says it's adding safety checks and has built more than 15 partnerships with governments and health groups. It's a look at how AI companies are trying to handle real safety risks, not just new features.

Grok Can Now Do Jobs on Its Own, on a Schedule

The story: xAI added a feature called Automations to its Grok chatbot on July 16. You describe a job in plain words, pick a time or a trigger, and Grok runs it for you and reports back.

Your takeaway: This turns a chatbot into something closer to an assistant that works while you sleep. It can send you a news summary every morning or watch for a certain email and act on it, and you don't need to know any code.

TOOLS ON OUR RADAR

📻 AzuraCast Free and Open Source: A complete web radio management suite that allows you to easily start your own internet broadcasting station and manage your digital music playlists online.

✍️ Etherpad Free and Open Source: A highly customizable document editor that lets large groups of people type on the same page at the exact same time making it incredible for collaborative drafting.

🏫 BigBlueButton Free and Open Source: A specialized web conferencing system built primarily for online education that includes digital whiteboards breakout rooms and tools designed specifically for teaching.

🎞️ HandBrake Free and Open Source: An essential desktop utility that easily converts video files from almost any format into modern and widely supported file types while significantly reducing their overall size.

TRENDING

NVIDIA's New Nemotron 3 Embed Tops a Key Search Test - NVIDIA released a free set of models that help AI find the right information to answer a question. Its 8B version ranked #1 on RTEB, a test of real-world search accuracy, so AI helpers can pull up better facts and give better answers.

Google Renames NotebookLM to Gemini Notebook and Adds Code - Google's research tool NotebookLM is now called Gemini Notebook. It can now write and run code to analyze your data, and it's already used by more than 30 million people.

Google Search Can Now Start Jobs Inside Other Apps - Google's AI Mode in Search can now link to apps like Instacart, Canva, and YouTube Music. So a grocery list you make by chatting can go straight into an Instacart cart without leaving Search.

South Korea Is Building Its Own AI Model for Cybersecurity - South Korea plans to launch a homegrown AI model that finds security bugs by the end of 2026. The push came after the U.S. limited access to some advanced models, showing how countries want AI they can fully control.

A Researcher Poisoned an Open AI Model for Under $100 - A security researcher showed she could secretly plant a hidden backdoor in an open model using just ten training examples, making it write unsafe code on command. Bigger models were easier to trick, a reminder that open models are hard to fully check.

Google Vids Now Lets You Star in Your Own AI Videos - Google Vids can now build a digital avatar that looks and sounds like you from one selfie and a short voice clip. You can also make and edit videos by typing what you want, and every clip gets an invisible SynthID watermark so people can tell it's AI.

TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)

⚡ Tell it where your chapter ends. Get sharper ways to close it, minus the clichés readers have seen a thousand times.

Build a single-file HTML app with vanilla HTML/CSS/JS. The Chapter Cliffhanger Generator — describe where a chapter ends, get sharper non-cliché ways to end it. Persist to localStorage key 'cliffhanger_gen_v1'.

Aesthetic: near-black (#0b0a10), crimson (#e0355c) primary with red glow top-left and blue glow bottom-right. Zilla Slab serif (italic for prompts/diagnosis, upright for the ending passages) for headings and prose, Inter sans for body, JetBrains Mono for labels. Ending cards use a subtle red gradient background with a pill-shaped technique badge.

Form: scene-at-chapter's-end textarea (encourage concrete detail: who's there, stakes, what's just happened), optional current-draft-ending textarea, genre dropdown (literary/thriller/fantasy/mystery/romance/horror/YA), desired-reader-feeling dropdown (dread/shock/intrigue/heartbreak/urgency).

System instructions: chapter-ending craft coach who understands page-turning techniques (the turn, the question, the image, the reversal, the quiet gut-punch, the ticking clock) and actively avoids overused clichés (convenient interruptions, "little did she know", unearned fake-out gasps, it-was-a-dream, monologues cut off for no reason). Write ACTUAL prose passages tailored to the specific scene and characters, not generic templates. If a current draft was given, diagnose it honestly first. No em dashes. Return raw JSON: diagnosis (honest read on current draft, empty string if none given), endings (exactly 4, each a different technique: technique name + a 2-4 sentence prose passage + why with **bold** key craft move), avoid (a specific cliché trap this scene risks + why to skip it).

Render: optional diagnosis card (only shown if a current draft was given). Four ending cards, each with a pill-shaped technique-name badge, the prose passage in large serif type, and a bordered-off italic why line. Blue-violet "skip this one" caution card. Copy endings + archive keyed by scene snippet + genre.

What this does: Describe the scene at your chapter's end, optionally paste your current draft ending, then set genre and the feeling you want the reader left with. If you gave a current draft, it starts with an honest specific read on whether it's landing. Then it writes four genuinely different ending options, each using a real page-turning technique (the turn, the question, the image, the reversal, the quiet gut-punch, the ticking clock), written as actual prose tailored to your specific scene and characters, not generic templates, each with a one-line reason it works. A closing note flags the specific cliché trap this scene is at risk of falling into. Saves every chapter to localStorage.

What this looks like:

WHERE WE STAND(based on today’s news)

✅ AI Can Now: Run a near-top model on your own computer for free using open weights like Kimi K3, which also handles images, video, and a million words at once.

❌ Still Can't: Have its behavior fully checked or predicted, even when the weights are public, since a few bad training examples can hide a backdoor.

✅ AI Can Now: Write and run its own code to analyze your data inside a research tool like Gemini Notebook.

❌ Still Can't: Always show its work clearly enough for you to catch every mistake it makes.

FROM THE WEB

— # (#)

RECOMMENDED LISTENING/READING/WATCHING

Little Joe (2019) - Movie

An Austrian sci-fi drama directed by Jessica Hausner, about a corporate plant breeder in an English lab (Emily Beecham, who won Best Actress at Cannes for the role) who has designed a new engineered flower called "Little Joe" that is supposed to make anyone who cares for it happy. Coolly composed, and only moderately sinister film about the promise and menace of any technology that promises to optimize human wellbeing. Ben Whishaw is in it and is also excellent.

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.}