🧠 Meta Reads Minds, Google Finance Gets Scary Good, and HP Goes All In

Quick context: last year Meta showed off Brain2Qwerty v1, research that turns brain activity into typed text without any implant. This week they shipped v2, and it reads full sentences in real time.

Here's how it works. They trained the system on roughly 22,000 sentences from nine volunteers, each sitting for 10 hours inside an MEG scanner (a machine that reads magnetic signals from the brain) while typing. The AI learns to map those raw signals to words.

The numbers are the headline. Average word accuracy hit 61% across participants, and the best volunteer reached 78%, with more than half their sentences coming out with one word error or less. Accuracy also keeps climbing as you add more data, which suggests this only gets better from here.

This could give a voice back to the millions of people with brain conditions that block communication, without the brain surgery current methods require. Meta is releasing the full training code, and its research partner is opening up the v1 dataset.

Google Finance just came out of beta with a real upgrade, plus a new Android app. The big addition is portfolios, now rolling out globally.

You drop in your holdings by uploading a CSV, a PDF, or even a screenshot, or you can just describe what you own in plain English. From there you get one dashboard with performance, asset allocation, and a research tool you can interrogate like a chatbot ("what sectors am I underweight in?").

There's also a scheduling feature. You describe a task like "send me a daily pre-market briefing on big overnight crypto moves," and it runs in the background and delivers on your schedule. The new app adds a live news feed and AI "key moments" that explain why a stock actually moved.

P.S. this is a research and tracking tool, not financial advice. iOS users have to wait, the app is Android-only for now.

OpenAI announced that HP is scaling up its Frontier partnership, and the details are a useful reality check on where corporate AI actually is.

HP only started testing in February 2026. Within weeks, one engineer used the models to push through 122 pull requests across 43 different projects. The security team unlocked an estimated 82 hours of capacity a week using ChatGPT to remediate bugs and speed up analysis.

Why it matters: this is the version of AI adoption nobody posts hype threads about. It is less "magic chatbot" and more plumbing, with HP using OpenAI's tools (ChatGPT and Codex) wired into real workflows with governance around them. That is the boring-but-real path most companies are now copying.

Help shape Practically AI
Tell us what's working. Your feedback shapes what lands in your inbox, and it takes less than 2 minutes.

Quick context: Claude Code is Anthropic's coding agent that lives in your terminal, and "subagents" are little helper agents it spins up to handle focused side tasks like research or code review.

In the next version, those subagents run in the background by default, so you can keep chatting with Claude while they work. Want one in the foreground instead? Just tell Claude. Best for developers who hate watching a progress bar.

Gemini can now take your meeting notes inside Google Meet for Google AI Pro ($19.99/mo) and Ultra subscribers, plus eligible Workspace accounts. It transcribes the call, writes a summary with action items, saves it to a Google Doc, and emails you the recap. It works on web and mobile in eight languages. If you already pay for Google AI, this might quietly replace a separate notetaker subscription. Best for solopreneurs and anyone living in back-to-back calls.

We came across a video this week we cannot stop thinking about. An auto driver in Bengaluru, filmed by a local creator, casually walking through how ChatGPT has become part of his daily routine. He types his questions in English and asks for answers in Kannada, digs into everything from general knowledge to astrology, and even has sharp takes on where it gets things wrong. OpenAI's own account loved it enough to drop a thank-you in the comments, in Kannada and English.

Here's why it stuck with us. This is a person with no tech background, no fluent English, no insider vocabulary, using a frontier AI tool more thoughtfully than plenty of people who do this for a living. AI is the rare technology where your profession, your language, or your starting point doesn't decide whether you get to benefit from it. The only real requirement is the will to learn.

We believe that kind of access should belong to everyone, which is exactly why we show up in your inbox to make you 1% better at AI every day.

Help shape Practically AI
Tell us what's working. Your feedback shapes what lands in your inbox, and it takes less than 2 minutes.