In partnership with
Welcome, AI enthusiasts
For most of human history, the only way to pull words out of a silenced brain was to cut into it. Meta just skipped that step. Its researchers read full sentences off the brain using nothing but a scanner the volunteers sat under, no implant and no incision anywhere. The accuracy jumped eightfold over earlier non-invasive attempts, close enough to surgical systems that the gap now looks like a matter of data, not biology. The lab racing Meta into your social feed is now racing into your head. Let's dive in!
In today’s insights:
Meta reads your Brain 8x better without Surgery
A Free Chinese Model Matched Mythos
South Korea Bets $1 Trillion on the AI Chip War
Read time: 4 minutes
LATEST DEVELOPMENTS
Evolving AI: Meta read full sentences straight off the brain, and it needed no implant to do it.
Key Points:
Brain2Qwerty v2 turns raw brain activity into typed sentences in real time, with no surgery.
The system reached 61% word accuracy, up from the 8% older non-invasive methods managed.
Meta is releasing the full training code so other labs can take it further.
Details:
Meta trained Brain2Qwerty v2 by feeding raw signals from a MEG scanner into deep learning models, then fine-tuning large language models to recover the meaning. Nine volunteers wore the scanner for ten hours each while typing, which gave the system about 22,000 sentences to learn from. Accuracy also kept climbing as more data went in, and the best participant reached 78%.
Why It Matters:
Meta is chasing the kind of recovery that today still means brain surgery, a non-invasive way to give people their words back. A stroke or ALS can take someone's speech overnight, and ordering coffee or texting a daughter suddenly goes through a caregiver. For anyone who has watched a parent go quiet, a route that stays out of the operating room is the difference that counts.
You've seen the AI demos. Viktor does it without you watching.
The AI tool you tried last quarter waited for a prompt, hallucinated a number, then asked if you'd like a summary.
Viktor opened a PR at 2am, rebased it against main, ran your test suite, and posted a note in #eng: "Two flaky tests in payments service, both pre-existing. Recommended merging after fixing them." Then drafted the customer reply for the support ticket the bug created.
That's 619K autonomous actions per day across 20,000+ teams. Not chat replies. Real work shipped to GitHub, Stripe, Linear, Notion, and 3,000+ other tools, from inside Slack and Microsoft Teams.
You don't supervise him any more than you supervise a senior engineer.
SOC 2 certified. Your data never trains models.
"It's what you probably originally thought AI was going to be when you first heard of it in sci-fi movies." Tyler, CEO.
CHINESE AI
🔓 A FREE Chinese Model Matched Mythos
Evolving AI: Chinese lab Z.ai's GLM-5.2 matched Anthropic's restricted Mythos at finding security bugs.
Key Points:
Semgrep and Graphistry tested GLM-5.2 against the top closed models on cybersecurity tasks.
Graphistry's agentic benchmark put it level with the proprietary models and ahead of every open-weight rival.
GLM-5.2 found each flaw for about $0.17, a sixth of what the closed labs charge.
Details:
Anthropic's Mythos leads the frontier at spotting software vulnerabilities, the exact skill the US moved to lock down in June. The Wall Street Journal reported GLM-5.2 keeping pace with it in real security scenarios, and Semgrep watched the open model beat Claude Code on a hard task with just a bare prompt. Both firms stress the wins stay narrow and that GLM-5.2 still trails the closed labs on broader work. Anyone can download the weights free under an MIT license and run them with no regional limits.
Try GLM-5.2 free at chat.z.ai.
Why It Matters:
Washington moved to keep flaw-hunting like Mythos inside a few trusted labs, and a free AI model just put it on any laptop. That reaches into daily life, since the banking apps and inboxes people trust run on code far more strangers can now comb through. The safety of your everyday software comes down to who points the model at it first.
The browser that reads the room before you ask.
Most browsers get you to the page. Norton Neo gets you to the answer. Magic Box understands your intent before you finish typing — no prompting, no switching apps, no copy-pasting. Built-in AI, instantly and for free. Privacy handled by Norton, by default.
Source: Reuters
Evolving AI: South Korea staked more than $1 trillion on the chips inside modern AI.
Key Points:
President Lee Jae-myung unveiled the plan Monday with the Samsung and SK Hynix chiefs beside him.
The strategy rests on semiconductors, physical AI and AI data centers.
The build-out spreads chip and AI work out of Seoul into new regional hubs.
Details:
Samsung and SK Hynix anchor the plan with $518 billion for four fabrication plants in the southwest. SK Group, GS Group and Naver back the data centers, built to clear 18 gigawatts by 2035. Japan and Taiwan are spending hard on the same technology and US cloud giants will lay out roughly $650 billion on AI this year. Behind all of it sits high bandwidth memory, the specialized chips that feed nearly every AI processor.
Why It Matters:
South Korea makes the memory that powers the AI tools and devices you reach for daily. When that supply tightens, the cost of your phone, your laptop and your AI subscriptions climbs with it. This buildout is the slow fix, and until the new plants come online those prices hold.
👀 Click on the image you think is real
QUICK HITS
🎵 TIDAL Cuts Off Money for AI Music as100% AI Tracks Get Tagged and Demonetized.
💸 Big Tech Has Lost $2.7 Trillion in Value This Month as the AI Bill Comes Due.
🔧 Qualcomm Spends $4B on Modular to Make Nvidia's Chips Optional.
🛡️ CrowdStrike Acquires IperLane as Cyber Firms Race to Own the AI-Inference Layer.
📈 Trending AI Tools
📝 Granola - AI notetaker that captures the real insights and turns every conversation into ready-to-share, action-driving notes*.
🛒 Webotee - Amazon brand and seller intelligence with daily data on 1.5M products.
🤝 Cofound Agent - MCP directory where your AI agent finds and screens human co-founders.
🎙️ Modulate - Voice AI for real-time toxicity detection and trust on gaming platforms.
*partner link