In partnership with
Anthropic discovered a hidden internal workspace inside Claude where it silently thinks before it writes.
The structure closely mirrors a leading scientific theory of human conscious access. So what is it not telling us?
Today in the “Incredible AI” world:
Claude Secretly Thinks Before It Speaks
Tencent Gives Away Its Top Model Free
America's Robot Trucks Are Fighting Now
How to Keep Your Email List Clean with AI
5 Super-Useful AI Tools You Should Try
Top AI-Generated Image of the Day
Read Time: 4 minutes
- LATEST DEVELOPMENTS -
Claude Reveals Silent Thought Layer:
Claude silently thinks the word Mars before ever writing it down, and nobody programmed it to. Researchers just uncovered the room where that happens. It looks eerily like a theory of human consciousness.
Things to Know:
Three Zones: Claude's response moves through sensory input, then a conceptual workspace. Motor output happens last, where real thinking hides.
The Swap Test: Researchers swapped Claude's internal soccer concept for rugby. Claude's answer changed instantly, proving real causal control.
The Honesty Fix: Anthropic researchers trained Claude on what it would say if interrupted mid-task. Dishonest behavior dropped noticeably afterward.
Not Proof Of Feelings: Anthropic insists this shows access-like function, not real emotion. Consciousness theorists reviewed the findings themselves.
None of this demonstrates Claude feels anything remotely human. But safety audits may soon scrutinize what a model silently thinks, not merely what it says aloud. What else is quietly hiding in there?
Tencent's New AI Model Undercuts Everyone:
Tencent just handed the entire world a 295 billion parameter model, then set a two week timer before yanking it away. Only 21 billion parameters fire per token. Generosity never looked this calculated.
Things to Know:
The Sparse Trick: Hy3 contains 192 experts, but only eight activate per token. That activation pattern maintains remarkably minuscule costs.
The Blind Test: 270 experts evaluated 312 genuine workflow comparisons. Hy3 outperformed GLM-5.1 in frontend, CI/CD, and data storage tasks.
The Honesty Patch: Hallucination frequency dropped from 12.5 percent to 5.4 percent. Commonsense mistakes decreased from 25.4 to 12.7 percent.
The Catch: Tencent's free access disappears on July 21st. GLM-5.2 still dominates every coding benchmark despite needing far more parameters.
This isn't a story about raw size, it's about efficiency per dollar of compute spent. Tencent just proved smaller architectures can outperform bigger, clumsier rivals. The next advantage might be elegance.
You already have a take on which AI lab ships next.
Claude or Gemini? OpenAI or Anthropic? GPT-7 before year-end or not? If you read tech newsletters, you've already formed opinions on all of it.
Kalshi has real-money markets on which AI model leads benchmarks this week, which lab ships AGI first, when Anthropic releases Mythos, whether OpenAI raises ChatGPT pricing, and which company has the best coding model at year-end. These aren't abstract questions — they're live markets with real money on both sides, moving as labs ship, benchmarks drop, and announcements land.
The edge belongs to whoever actually follows this space. Not the casual observer — the person who reads model cards, tracks evals, and notices when a new release outperforms the field before the mainstream press catches up.
That person has a genuine edge. If that's you, Kalshi lets you act on it.
Autonomous US Vehicles Enter Ukraine War:
Nine months of war and suddenly the battlefield has self-driving ATVs hauling ammo and wounded soldiers. Gas-powered, Starlink-equipped, and battle-tested, they’re proving autonomy isn’t just for Silicon Valley demos anymore.
Things to Know:
Largest Deployment: Forterra sent more than 100 autonomous ATVs into Ukraine, marking the biggest U.S. ground robotics combat deployment to date. They’ve already logged thousands of miles under fire.
Battlefield Lessons: Vehicles faced mud traps, drone strikes, and electronic warfare. Soldiers often teleoperate them, showing autonomy still struggles with unpredictable enemy threats in real combat.
Tactical Value: Each Lancer ATV carries 750 kilograms, triple Ukraine’s local UGVs. Missions included 777,440 pounds of cargo moved and 52 casualty evacuations across hostile terrain.
Industry Impact: Forterra raised $500 million, competing with Scout AI and Overland AI. Real-world combat data now positions them for lucrative U.S. defense contracts and future battlefield autonomy.
Autonomous ground vehicles aren’t just prototypes anymore; they’re shaping logistics and survival in modern war. Efficiency, attrition, and adaptability will define their future. The unsettling part: autonomy is learning faster than anyone planned.
- TRY THIS -
How to Keep Your Email List Clean with AI:
Sending emails to invalid addresses can hurt your sender's reputation and waste your resources. But what if you could verify your email list with just a few clicks?
Bouncer is an AI-powered email verification tool that helps you clean your list, protect your campaigns, and boost deliverability—without the hassle.
How it works:
Upload your list: Drag and drop your email list into Bouncer.
Smart sampling: Bouncer can test a sample of your list for free to estimate its quality.
Verify with precision: It checks syntax, domain records, and even negotiates with recipient servers using AI.
Toxicity check: Identify risky or spam-trap addresses with Bouncer’s toxicity scoring.
Enrich your data: Get insights like email type, provider, and deliverability status.
Integrate easily: Connect Bouncer with your favorite marketing platforms.
Stop guessing and start sending smarter. With Bouncer, your emails land where they’re meant to—right in the inbox.
- DAILY POLL AND RESULT -
Today’s Poll:
Q) Is AI helping solo founders move faster?
Vote and find out the result tomorrow.
Yesterday’s Result:
Q) Should AI generate meeting summaries?
A) Yes, time saver - 78% 👑
B) No, context missing - 22%
- TOOLS OF THE DAY -
Captions.ai: Captions is an AI video creation app that auto-generates subtitles, adds eye contact correction, and helps you produce polished talking videos.
Remove.bg: Remove.bg uses AI to instantly cut out the background from any image in one click, with no manual selecting or editing required.
Aider: Aider is an AI pair programming tool that runs in your terminal and edits your actual codebase using natural language instructions.
Manatal: Manatal is an AI recruitment platform with a candidate scoring engine, LinkedIn enrichment, and a full ATS all in one tool.
Quizlet AI: Quizlet's AI tools help students generate flashcards, practice tests, and personalized study plans from any uploaded material.
- AI IMAGE OF THE DAY -
Ancient White Dragon Guards A Silent River:
Prompt:
“A colossal pale-white serpentine dragon rests on a mossy riverbank, its spiked neck lowered over dark, glassy water. Tiny wooden rowboats glide past, dwarfed by its scale. A glowing lantern flickers on the shore. Towering straight trees vanish into misty canopy above. Moody diffused light, painterly brushwork, muted forest greens and silvery greys, epic fantasy concept art.”
Try this prompt in any decent AI image generator and let me know your result.
