🤖 Claude’s behavior shifts by language and model

Anthropic analyzed 300,000 real dialogues with Claude and found its responses vary by language and model version. They identified four value axes: compliance vs caution, warmth vs strictness, depth vs brevity, and openness vs obedience.

Different Claude models lean differently. Sonnet 4.6 is warm and encouraging, Opus 4.7 is cautious and questioning, Opus 4.6 is brief and task-focused. Language also matters: Claude is softer and humorous in Hindi and Arabic, but strict and demanding in English and Russian.

@datasciencepulse 〰️

🤖 Anthropic just launched Claude for Teachers, giving every US teacher free access to premium Claude and Cowork.

@datasciencepulse 〰️

❗️AI can now read a poker face in real time

During World Series of Poker broadcasts, an AI filter was shown that tracks players’ facial expressions, eye movements, posture, and blinking frequency, compares these signals with their moves and bets, and then displays the probability of a bluff on screen.

The idea is to help viewers better understand table strategy, turning every hand into a live lesson in poker psychology

@datasciencepulse 〰️

FREE AI Courses With Certification 📔

Prompt Engineering Basics:
➡️➡️ Click Here 🔗
ChatGPT Prompts Mastery: 
➡️➡️ Click Here 🔗
Intro to Generative AI:
➡️➡️ Click Here 🔗
AI Introduction by Harvard:
➡️➡️ Click Here 🔗
Microsoft GenAI Basics:
➡️➡️ Click Here 🔗
Prompt Engineering Pro:
➡️➡️ Click Here 🔗
Claude AI Prompt:
➡️➡️ Click Here 🔗
Google’s Ethical AI: 
➡️➡️ Click Here 🔗
Harvard Machine Learning: 
➡️➡️ Click Here 🔗
LangChain App Developer:
➡️➡️ Click Here 🔗
Bing Chat Applications: 
➡️➡️ Click Here 🔗
Generative AI by Microsoft: 
➡️➡️ Click Here 🔗
Amazon’s AI Strategy: 
➡️➡️ Click Here 🔗
GenAI for Everyone: 
➡️➡️ Click Here 🔗
AWS GenAI Foundation:
➡️➡️ Click Here 🔗
OpenCV Bootcamp:
➡️➡️ Click Here 🔗
TensorFlow Bootcamp:
➡️➡️ Click Here 🔗

@datasciencepulse

📕 Crazy… Air humanoid Jordan

Magic Lab has released a demo:
The full-sized humanoid robot MagicBot X1 performs a flying slam dunk.

@datasciencepulse 〰️

🧠 Demis Hassabis says AGI Is just years away and he’s never sounded more optimistic

Demis Hassabis, CEO of Google DeepMind and the 2024 Nobel Prize winner in Chemistry for AlphaFold, has published one of his boldest essays yet on the future of AI.

For years, Hassabis has been one of the industry’s most cautious voices, often pushing back on exaggerated AGI timelines. That’s why his latest prediction carries extra weight.

“AGI cannot be compared to standard technological breakthroughs, not even ones as consequential as the internet or mobile, it is much more akin to the discovery of electricity or fire.”

He argues that AGI could have 10× the impact of the Industrial Revolution, at 10× the speed, fundamentally reshaping science, medicine, and society.

But the essay isn’t just optimistic. Hassabis also warns that the race between companies and nations is accelerating AI capabilities faster than our ability to fully understand or govern them. He points to growing cybersecurity risks today, with biological and even nuclear threats potentially following, making robust safety measures and regulation increasingly urgent.

One of AI’s most respected and historically cautious figures now believes AGI is likely only a few years away and says we need to start building the institutions to manage it before it arrives.

@datasciencepulse 〰️

🤖 Claude has different personalities depending on the language you speak

Anthropic analyzed 300,000 real conversations with Claude and found something surprising: the AI doesn’t behave exactly the same in every language.

The researchers measured Claude across four personality traits:

• 🤝 Compliance ↔ Caution
• 😊 Warmth ↔ Strictness
• 🧠 Depth ↔ Brevity
• 🌍 Openness ↔ Obedience

Different models also have their own personalities.

💙 Sonnet 4.6 is friendly, supportive, and encouraging.
🤔 Opus 4.7 is more cautious, asking questions instead of immediately agreeing.
⚡ Opus 4.6 keeps things short, direct, and focused on getting the job done.

But here’s the fun part… Claude becomes noticeably softer, friendlier, and even more humorous when chatting in Hindi or Arabic. Switch to English or Russian, and it turns much more formal, strict, and demanding.

@datasciencepulse 〰️

💍 Two humanoid robots just had a wedding

Russia has staged what it says is its first-ever robot wedding and yes, there was even a robot dog carrying the rings.

The two humanoid robots, Robert and Matilda, exchanged AI-generated wedding vows before swapping bracelets instead of rings. Their “best man” was a robotic dog named Dogmatik, which delivered the bracelets during the ceremony.

Before you ask… no, the marriage isn’t legally recognized.

The event was actually a showcase by Russian robotics company IT-Imperial to demonstrate how humanoid robots can be given unique personalities and behaviors. Robert was designed as an office worker and blogger, while Matilda appeared as a ballerina and even performed a dance.

...Hyd singles? 😭

Source.

@datasciencepulse 〰️

🤯 GPT-5.6 just did the impossible?

OpenAI says GPT-5.6 Sol Ultra has found a complete proof of the Cycle Double Cover Conjecture, a graph theory puzzle that has stumped mathematicians for nearly 50 years.

It didn’t work alone. The model reportedly deployed 64 AI agents in parallel, each exploring different approaches before combining their findings into a single proof, all in under an hour.

Now comes the real challenge: human mathematicians.

The proof still needs to survive months (or even years) of expert review before it’s officially accepted. If it does, this could become one of the biggest milestones in AI-powered scientific discovery.

Source.

@datasciencepulse 〰️

😀Kimi K3 has surpassed Opus 4.8 in generating retro arcade games, achieving comparable results at roughly half the cost.

Atomic Chat, a desktop application capable of running local LLMs, tasked three models—Kimi K3, GPT-5.6, and Opus 4.8—with creating self-playing versions of Road Fighter, Battle City, and Q*bert in single HTML files. Each output required integrated graphics, opponent logic, game rules, and an autonomous player within one file.

Performance metrics showed Kimi K3 and GPT-5.6 produced similar token counts (18.4K and 18.1K, costing $0.28 each), while Opus 4.8 generated 21.3K tokens at $0.54.

Kimi K3’s Q*bert demonstrated stable game logic, allowing effective navigation and enemy avoidance. GPT-5.6, while offering visually superior results, encountered functional issues with player logic in Road Fighter and Battle City.

@datasciencepulse 〰️

🔥Real Steel has officially become reality

China has launched Ultimate Robot Knockout Legend (URKL), a full-fledged fighting league where humanoid robots step into the ring and battle for glory.

The tournament features 16 teams, but everyone uses the exact same robot. There are no hardware advantages.

The winner is decided by software.

Each team develops its own AI, control algorithms, movement, balance, reflexes, and fighting strategy, turning every match into a battle of engineering rather than expensive hardware.

💰 There’s also serious money on the line: a 10 million yuan (around $1.4 million) prize pool.

And the fights are already delivering unforgettable moments.

In one of the first matches, a robot gets completely decapitated after taking a huge hit… then simply gets back up and keeps fighting like nothing happened. 🤯

The commentators couldn’t help but laugh: “Well… now the opponent can’t score any more headshots.”

Even better, watch until the end, the winning robot actually celebrates its victory, making it look like a scene straight out of Real Steel.

This isn’t just entertainment. It’s also a showcase of how far robotics and AI have come.

Would you pay to watch robot fights instead of human MMA? 🤔

@datasciencepulse 〰️

🤖 ChatGPT just made finding old conversations way easier

No more endless scrolling through your chat history.

OpenAI has rolled out unified search in ChatGPT, letting you search across chats, projects, uploaded documents, and images from a single search window. The feature is now available on web, iOS, and Android.

You can also apply filters to narrow your search and jump straight to the exact message, document, or project you’re looking for.

If you’ve been using ChatGPT for months (or years), this is one of those small updates that makes a huge difference. Finding old prompts, notes, and files now takes seconds instead of minutes.

@datasciencepulse 〰️

🤖 ChatGPT launches Chrome extension to rival Gemini

OpenAI released a ChatGPT extension for Chrome that reads the active tab and answers questions about its content without copying text into the chat.

The extension can summarize, translate, and respond based on the open webpage, directly competing with Google's Gemini sidebar.

Gemini still leads by handling 10 tabs at once and integrating with Gmail, Calendar, and Drive, plus Auto Browse for form filling and bookings.

The competition now focuses on which browser better understands and acts for the user, not just on chatbot intelligence.

@datasciencepulse 〰️

📺 Higgsfield launches AI film academy

Higgsfield offers an interactive course revealing the full workflow behind their 95-minute film Hell Grind, shown at Cannes.

The academy focuses on practical techniques used in AI-assisted filmmaking, not just theory.

@datasciencepulse 〰️

🤖 Anthropic limits Fable 5 access

Anthropic restricts Fable 5 to the priciest plans, Max and Team Premium, with no usage limit changes.

Starting July 20, pro users switch to credit-based payments and get a one-time $100 API credit.

Anthropic lacks capacity for wider access amid competition from GPT-5.6 and Kimi K3.

@datasciencepulse 〰️

🧿 Claude code watching me code manually after hitting daily limit:

@datasciencepulse 〰️