🇨🇳 China Trains a Giant Model at Home

PLUS: Meta's brain-to-text system, faster Apple security patches, phone-location privacy, and WhatsApp usernames

In partnership with

China's AI stack is getting more self-reliant. The headline is a giant model trained on local chips, but the bigger story is how quickly AI infrastructure is turning into industrial policy.

Today: Meituan's LongCat-2.0 claim, Meta's brain-to-text research, Apple moving faster on security fixes, a Supreme Court ruling on phone-location privacy, and WhatsApp turning usernames into a privacy feature.

In today's menu:

  • 🇨🇳 China trains a giant model on local chips

  • 🧠 Meta keeps pushing brain-to-text AI

  • 🔒 Apple shortens the security patch window

  • ⚖️ Phone-location dragnet searches hit a limit

  • 💬 WhatsApp makes usernames a privacy layer

  • 🧰 5 sharp tools for builders and operators

P.S. Move this email to your primary inbox to make sure you do not miss the useful bits.

First time reading? Sign up here.

Academy

Learn the Workflows Behind the Brief

Use the Academy to turn AI tool discovery into repeatable systems your team can actually run.

One brand shipped 30+ landing pages last week. No developers.

A DTC brand briefed Viktor inside Slack: one landing page per Meta ad group, mapped to a different headline variant. He wrote the code, deployed each page to their subdomain, posted the URLs back in #marketing, and now monitors performance across the set.

Their content team uses him to draft email flows, generate creative variants, and audit Klaviyo segments every Friday. Their growth lead uses him to catch spend anomalies before the day starts.

20,000+ teams now have the same setup: one AI employee across every marketing tool. A teammate who ships work in Slack and Microsoft Teams. SOC 2 certified.

"Not only have we caught up on several months of work, we are automating manual tasks and expanding to things previously not possible at scale." Jesse, Director, Torque King 4x4.

Chips

🇨🇳 China Trains a Giant Model at Home

Meituan has launched LongCat-2.0, a 1.6-trillion-parameter model it says was trained entirely on domestically developed chips.

The important signal is not only model size. If Chinese labs can keep training large systems without relying on restricted Nvidia hardware, export controls become a speed bump rather than a hard stop.

What to watch: whether outside developers can reproduce the benchmark claims, how well the model runs on real workloads, and whether domestic chip clusters become good enough for more teams to leave the Nvidia-centered path.

AI

🧠 Meta Keeps Pushing Brain-to-Text AI

Meta-linked researchers are continuing work on Brain2Qwerty, a non-invasive system that decodes intended typing from brain activity recorded with MEG and EEG devices.

This is still lab infrastructure, not a consumer interface. But the direction matters: AI is making messy biological signals easier to translate into language, which could eventually reshape assistive communication.

The operational question is privacy. If neural data becomes easier to decode, healthcare teams, device makers, and regulators will need much clearer rules for consent, storage, and secondary use.

Security

🔒 Apple Shortens the Security Patch Window

Apple is reportedly releasing some security fixes earlier instead of holding them for larger operating-system updates.

The reason is practical: AI can compress the time between a public vulnerability and a usable exploit. Once attackers can turn diffs, advisories, and code patterns into working attacks faster, slow patch cycles become a bigger liability.

For operators, this is another sign to treat software updates as an incident-response habit, not an annual platform ritual. Faster patches also mean stronger testing and rollout discipline inside teams.

10x the context. Half the time.

Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.

Privacy

⚖️ Phone-Location Dragnet Searches Hit a Limit

The US Supreme Court ruled that constitutional privacy protections apply to cellphone location history in a case involving geofence warrants.

That matters because location data is no longer just a carrier record. It is a detailed behavioral map created by apps, operating systems, cloud accounts, and services that users often keep enabled by default.

The broader lesson for product teams: location, identity, and behavioral data carry legal and trust risk even when users technically opted in. Consent that works for a feature may not survive as consent for surveillance.

Messaging

💬 WhatsApp Makes Usernames a Privacy Layer

WhatsApp is opening username reservations ahead of a broader rollout, letting people connect without exposing their phone numbers to every new contact.

The design choice is interesting: usernames are not a social discovery surface. There is no public directory, and an optional username key gives users another filter before new people can message them.

For creators, businesses, and community operators, this creates a cleaner way to publish a contact route without turning a personal phone number into public infrastructure.

Find better reads faster

🧰 AI Tools of the Week

📞 Aircall
An AI voice agent for calling, qualifying, and following up with leads without leaving coverage gaps.

📱 Cursor
AI-assisted coding workflows that now make more sense for teams managing agents and pull requests away from the desk.

📊 Databox Skills Marketplace
A way to bring in vetted analytics help for dashboards, reporting setup, and data operations work.

🗓️ Akiflow
A unified task and calendar command center for operators who need one place to plan the day.

🎨 v0
AI-assisted UI generation for turning product ideas into editable interface components faster.

Thanks for stopping by!
Have some feedback or want to sponsor this newsletter?

BTW - I keep my inbox organized in Meco, game changer for inbox sanity.

Not a subscriber? Sign up for free below.