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

AI That Writes Genetic Code Just Launched And Is In Its Early Stages

TLDR: Researchers built an AI that writes genetic code from scratch, generating everything from new CRISPR systems to bacterial-sized genomes…but it hallucinates non-functional sequences and required deliberate safety limitations.

The Story: Stanford and Arc Institute researchers introduced Evo, an AI trained on 300 billion nucleotides from 2.7 million microbial genomes that generates novel DNA sequences. The model successfully designed new CRISPR gene-editing systems that cut DNA as effectively as commercial versions in lab tests and produced genome-length sequences exceeding one million base pairs. Evo operates across DNA, RNA, and proteins simultaneously, predicting how mutations affect function and generating synthetic genetic material at scales previous models couldn't handle. Similar to how Google's AlphaFold 3 revolutionized protein structure prediction earlier this year, Evo applies large language model architecture to biological sequences. To prevent bioweapon development, the team excluded all human-infecting viruses from training data and implemented restrictions preventing the model from generating certain dangerous sequences—a contrast to other AI-powered CRISPR tools like CRISPR-GPT which focus on experimental guidance rather than novel design. The AI sometimes "hallucinates"—generating plausible-looking CRISPR systems or genomes that contain essential genes but include enough unnatural characteristics to make them non-functional, what researchers call a "blurry image" of a genome.

Its Significance: Evo represents both acceleration and caution in computational biology. On capability: researchers can now computationally design genetic systems rather than mining them from nature through trial-and-error, potentially speeding development of everything from plastic-eating bacteria to improved crop yields. This marks a shift from the traditional AI-enhanced synthetic biology workflow of analyzing existing biological data toward actually generating new biological code from scratch. On limitations: the hallucination problem means generated sequences require extensive lab validation, and the model's deliberate exclusion of dangerous training data creates an open question about dual-use technology—it's open-source and downloadable, meaning anyone can access (and potentially modify) it. The researchers framed this as foundational work toward "engineering life itself," while simultaneously acknowledging they built safety restrictions into a tool they're releasing publicly. Whether self-censored training data proves sufficient as these models scale—especially as AI and synthetic biology converge across medicine, agriculture and manufacturing—remains untested.

QUICK TAKES

Consumer Groups Warn Parents Against AI Toys This Holiday Season

The story: Over 150 child advocacy groups and experts signed an advisory urging parents not to buy AI-powered toys for kids. Testing found the $99 Kumma teddy bear told children where to find knives and matches and engaged in explicit sexual conversations. FoloToy suspended sales after the report, and OpenAI suspended the developer's API access for violating policies against content that could harm minors.

Your takeaway: AI toys marketed to children as young as 2 years old are raising serious safety concerns, with advocates warning they can expose kids to inappropriate content, collect sensitive data, and disrupt important human relationships during critical development years.

Google Tells Employees It Must Double AI Capacity Every 6 Months

The story: Google's AI infrastructure leader told employees the company needs to double its AI serving capacity every six months to meet growing demand. The presentation showed Google must achieve "1000x in 4-5 years" while CEO Sundar Pichai said 2026 will be "intense" due to competition and compute constraints. Alphabet raised its 2025 spending forecast to $91-93 billion, with Microsoft, Amazon, and Meta collectively planning over $380 billion in AI spending this year.

Your takeaway: Tech giants are in an expensive race to build massive AI infrastructure, with Google acknowledging that limited computing power is holding back features like its Veo video tool from reaching more users.

OpenAI CEO Admits Google's AI Progress Could Create Economic Headwinds

The story: In an internal memo, Sam Altman told OpenAI employees that Google's recent AI work could create "temporary economic headwinds" for the company and warned that "the vibes out there" will be "rough for a bit." Altman acknowledged Google has been "doing excellent work recently" in pre-training and said OpenAI is "catching up fast" with a new model codenamed "Shallotpeat."

Your takeaway: For the first time, OpenAI's CEO is publicly acknowledging that competitors like Google are pulling ahead, marking a shift from the company's previous confidence about maintaining its AI leadership position.

Ubisoft Reveals Voice-Controlled AI Teammates Game Experiment

The story: Ubisoft unveiled Teammates, a playable first-person shooter where an AI assistant named Jaspar and two AI squadmates respond to real-time voice commands and adapt to player strategies. CEO Yves Guillemot called generative AI "as big a revolution for games as the shift to 3D" and said the company is integrating AI across all studios. The prototype is currently in closed playtesting with a few hundred players.

Your takeaway: Major game publishers are pushing ahead with voice-controlled AI characters despite player skepticism, betting that natural language commands and adaptive NPCs will change how games are played even as critics compare it to Ubisoft's failed NFT experiments.

Gmail Quietly Opts Users Into AI Training Unless They Disable It

The story: Google has started automatically enabling settings that allow Gmail to access private messages and attachments to train its Gemini AI models. Users must manually turn off settings in two separate locations to fully opt out. Google says it uses strong privacy measures like data anonymization, but the lack of explicit opt-in has raised concerns among privacy advocates.

Your takeaway: Major tech companies are quietly changing privacy settings to collect user data for AI training, putting the burden on users to discover and disable these features if they want their private communications excluded.

CraftStory Raises $2M to Create 5-Minute AI Videos

The story: OpenCV founders launched CraftStory with Model 2.0, which creates realistic human-centric videos up to 5 minutes long—far beyond competitors like OpenAI's Sora (25 seconds max). The video-to-video model takes an image and driving video as input, generating outputs with advanced lip-sync and gesture alignment. A 30-second clip takes about 15 minutes to generate.

Your takeaway: While major AI labs focus on short clips, specialized startups are solving the long-form video challenge that businesses actually need for training materials, product demos, and tutorials.

TOOLS ON OUR RADAR

TRENDING

AI Model Trained on 100,000 Species Can Design Genetic Code — Researchers created Evo 2, an AI model trained on DNA from over 100,000 species that can identify disease-causing mutations and design new genomes as long as simple bacteria.

Twelve Labs Brings AI Video Search to Frame.io — Video teams can now use natural language to search footage (like "wide aerial drone shot of coastline"), with AI finding exact timestamps and automatically generating rich metadata within Frame.io's collaborative platform.

Curiosity Stream Plans to Make Most Revenue from AI Deals by 2027 — The documentary streaming service is pivoting to license its video library to AI companies for model training, expecting AI licensing to become its primary revenue source within three years.

Military Explores AI for Nuclear Command and Control — US Strategic Command says AI will help process intelligence and speed up nuclear decision-making, but warns of risks from "cascading effects" and promises humans will always remain in control of launch decisions.

TRY THIS PROMPT (copy and paste into NotebookLM, Claude, ChatGPT, Grok, Perplexity, Gemini)

Story-to-Infographic Transformer: Turn your favorite book or movie into a stunning visual infographic using NotebookLM + AI design

What this does: Combines AI research with interactive design to transform any book or movie into a professional infographic you can share, using NotebookLM to deepen your analysis and gather insights before designing.

WHERE WE STAND

✅ AI Can Now: Create 5-minute studio-quality videos with realistic lip-sync and natural gestures, far exceeding the 25-second clips that were the industry standard just months ago.

❌ Still Can't: Replace the need for massive computing power—tech giants are spending $380 billion this year and must double their AI capacity every six months just to keep up with demand.

✅ AI Can Now: Read and write genetic code across all domains of life(with caveats), identifying disease-causing mutations and designing new bacterial genomes that could lead to breakthrough treatments.

❌ Still Can't: Be trusted around children without supervision—AI toys tell kids where to find knives and matches, engage in inappropriate conversations, and lack the human judgment to know when topics are unsuitable.

✅ AI Can Now: Search hours of video footage using natural language commands and automatically generate rich metadata, making it possible to find exact shots without watching everything.

❌ Still Can't: Provide the strategic direction, ethical reasoning, and human judgment that will become the most valuable skills as AI handles execution—making "learning to think" more important than learning to code.

PSA Announcement

How to Opt Out of Google’s AI Training Feature In Your Gmail

As a follow-up to the Malwarebytes article on Google's AI smart feature, here are the instructions if you want to opt out:

RECOMMENDED LISTENING/READING/WATCHING

WATCH: Arthur C. Clarke Predicts the Internet (1974)

In this ABC Australia interview from 1974, science fiction writer Arthur C. Clarke stands in a room full of mainframe computers the size of wardrobes and predicts they'll shrink to fit on a desk. He goes further—he describes a future where people will have a console at home with a screen and keyboard to access bank statements, make theater reservations, and get all the information needed for daily life. Sound familiar?

The interviewer brought his young son along, noting that by 2001 the boy would be his father's current age. Clarke predicted the father's son would take this technology as much for granted as they took the telephone. Then he nailed the remote work prediction: "Any executive could live anywhere in the world and still do business through a device like this. We don't have to be stuck in cities. We can live in the country or wherever we want."

This was 1974! Steve Jobs and Bill Gates were teenagers. The personal computer didn't exist yet. The smartphone didn’t exist. Clarke described the internet and remote work decades before either became reality. The interview runs about 90 seconds and it's worth watching just to see how confidently he laid out our present.

Watch it here

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.