☕️ Meta AI model predicts human brain reactions

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

🧠 Meta AI model predicts human brain reactions

📧 FBI director Kash Patel’s email hacked

🤝 Zuckerberg offered to help Musk with DOGE

👓 Meta plans Ray-Ban smart glasses for prescription wearers

🔍 Yahoo launches AI answer engine Scout

Plus: 🎁 10 other news you might like, 🧰 6 tools, and 📚 3 papers.

🧠 Meta AI model predicts human brain reactions LINK

Meta's FAIR lab built an AI model called TRIBE v2 that predicts how the human brain reacts to images, sounds, and speech, often matching the typical brain response better than any single person's fMRI scan.

TRIBE v2 was trained on over 1,000 hours of fMRI data from 720 subjects and predicts brain maps with 70,000 voxels, a huge jump from TRIBE v1, which covered just four subjects and 1,000 voxels.

The model still has significant limitations: fMRI only tracks blood flow with a seconds-long delay, three sensory channels are missing, and it treats the brain as a passive receiver without modeling decisions or actions.

📧 FBI director Kash Patel’s email hacked LINK

An Iranian government-backed hacking group called Handala claims it broke into FBI director Kash Patel's personal Gmail account, posting pictures of him and a link to leaked files.

TechCrunch verified several leaked emails using cryptographic signatures in the message headers, which strongly suggests the messages from both Patel's Gmail and FBI accounts are authentic.

Since the U.S.-Israeli war against Iran started in February, Handala has ramped up its hacks, including a destructive attack on medical tech giant Stryker that wiped tens of thousands of devices.

🤝 Zuckerberg offered to help Musk with DOGE LINK

Mark Zuckerberg texted Elon Musk in February 2025 praising DOGE's progress and offering to help, including having Meta teams remove content doxxing or threatening DOGE employees, according to court documents.

The texts surfaced in Musk's lawsuit against Sam Altman and OpenAI, and they also show Musk asking Zuckerberg if he was open to bidding on OpenAI's IP together.

Musk's lawyers argued the private exchanges with Zuckerberg should be excluded from trial, calling them tangential and saying OpenAI's team is trying to stoke negative sentiments toward Musk.

👓 Meta plans Ray-Ban smart glasses for prescription wearers LINK

Meta is preparing to release new Ray-Ban smart glasses built specifically for people who already wear prescription lenses, according to a report from Bloomberg.

FCC filings for two models called "Scriber" and "Blazer" show they are production units with Wi-Fi 6 UNII-4 band support, which could enable faster data transfers and livestreaming.

The new glasses would be sold through traditional prescription eyewear channels, though it remains unclear how their design will differ from existing Ray-Ban Meta models beyond the prescription focus.

🔍 Yahoo launches AI answer engine Scout LINK

Yahoo has launched Scout, an AI-powered answer engine now available to its 250 million U.S. users, aiming to simplify online search and deliver more personal results tied to each person's interests.

CEO Jim Lanzone, who took over after Apollo Global Management bought Yahoo for $5 billion in 2021, has been cutting dysfunctional parts and overhauling services like email and fantasy sports.

Scout runs on AI technology licensed from Anthropic and will compete against Google's Gemini, OpenAI's ChatGPT, Anthropic's Claude, and the answer engine Perplexity in a crowded market.

Other news you might like

• Why SoftBank’s new $40B loan points to a 2026 OpenAI IPOLINK

• Apple says no one using Lockdown Mode has been hacked with spywareLINK

• Meta will fund seven new natural gas plants to power its biggest data center yetLINK

• OpenAI upgrades Codex to automate your workflows - and compete better with Claude CodeLINK

• Waymo’s skyrocketing ridership in one chartLINK

• Anthropic throttles Claude as demand for ChatGPT rival surgesLINK

• Windows PCs crash three times as often as Macs, report saysLINK

• Meta's hyperagents improve at tasks and improve at improvingLINK

🧰 Trending tools

SlapMac: turns your MacBook's built-in accelerometer into a sound machine: slap it, tap it, shake it, and it reacts with hilarious sound effects. LINK

Twitch Roulette: randomly connects viewers to live Twitch streamers with low viewer counts to help them gain exposure.LINK

Crossnode: a platform that converts AI agent workflows into paid products by handling authentication, billing, and usage limits without requiring backend development.LINK

Aera Browser: a browser that runs AI agents to automate multi-step web tasks in the background using MCP protocol, keeping all data stored locally.LINK

CrabTalk: a lightweight 8 MB daemon that streams real-time agent events including text deltas, tool calls, and thinking steps to your client application.LINK

Apparent for Gmail: a browser extension that reverses Gmail's conversation threading to show newest messages first and removes visual clutter for faster email reading.LINK

📚 Trending papers & reports

AI tools in competitions increase entry rates by 42% but don't improve individual success rates, suggesting AI lowers barriers to participation without enhancing quality or changing who wins.LINK

Repeating prompts multiple times boosts accuracy by up to 4.7% on tasks like translation and summarization in smaller language models that lack chain of thought reasoning capabilities.LINK

LLM code generation tools produce vulnerable code 30% of the time in tests, so researchers propose a framework that filters risky suggestions before showing them to developers.LINK

Want to get the latest news differently? Find us on:

See you tomorrow for a new dose of ☕️ Techpresso!