Howdy again. It’s Barsee, and welcome back to AI Valley.

Another day, another AI adventure.

Today’s climb through the Valley reveals:

• Anthropic new tool enables Claude to stop and think before answering

• OpenAI and MIT studies ChatGPT's emotional impact

• Ant Group claims AI breakthrough on China-built chips 

• Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

PEAK OF THE DAY

Anthropic new tool enables Claude to stop and think before answering

Anthropic has introduced a "think" tool for its AI assistant Claude, designed to enhance problem-solving by giving Claude a dedicated space to pause, reason, and strategize during multi-step challenges.

Here’s what you need to know:

• The "think" tool acts as a dedicated scratchpad where Claude can jot down its reasoning process before arriving at a final answer. This helps it break down complex problems into smaller, manageable steps, improving accuracy and logical consistency.

• Unlike Claude's "Extended Thinking" feature, which helps it plan before responding, the "think" tool allows Claude to “stop and think” mid-conversation, especially when processing new information from other tools.

• The tool is especially useful for analyzing external tool outputs, policy intensive decision-making, and handling step-by-step reasoning, where each action depends on the previous one and errors can be costly.

• The addition of the "think" tool significantly improves performance in complex reasoning tasks. Claude 3.7 Sonnet saw a 54% increase in accuracy in airline customer service simulations and a 1.6% improvement in software engineering tasks.

• The tool integrates seamlessly with existing Claude systems and only activates when needed, ensuring no unnecessary performance overhead. While most tests used Claude 3.7 Sonnet, Anthropic confirms similar improvements with Claude 3.5 Sonnet (New).

• Developers can implement the tool by identifying use cases where Claude struggles with policy compliance or complex reasoning, defining a customized "think" tool tailored to their needs, and refining its outputs through prompt adjustments.

Why it matters: 

Allowing Claude to pause and think during conversations boosts its ability to solve complex, multi-step problems accurately and efficiently. This makes it particularly valuable for tasks like policy compliance, customer service, and software engineering, where every decision counts.

3x faster than typing using AI

Meet your new productivity hack: Work smarter, not harder with an app that transforms jumbled thoughts into clear communication. Flow captures your words in real time and formats them so you don't have to. Flow automatically edits out filler words and clutter. Want a different tone or a quick AI-driven rewrite? Just ask Flow.

• 3x Faster Than Typing – Get more done in less time.

• Natural Voice – Flow sounds like you, not a stilted bot.

• Auto-Edits & Formatting – No more typos or punctuation errors.

• 100+ Languages – Communicate globally without missing a beat.

Because typing is so last decade. Stop typing, start speaking—Flow will handle the rest.

Download Wispr Flow for free today.

_{*This is sponsored}

VALLEY VIEW

🔮 Tencent's new reasoning AI matches DeepSeek's R1 in performance and pricing

Tencent has unveiled Hunyuan T1, a self-developed AI model with enhanced reasoning capabilities. It excels in long-text processing and delivers responses twice as fast, thanks to its hybrid Transformer-Mamba architecture. T1 matches or surpasses competitors like DeepSeek R1, OpenAI’s o1, and GPT-4.5 across key benchmarks, particularly in math and coding. Pricing is also competitive with DeepSeek, charging $0.14 per million input tokens and $0.55 per million output tokens.

👀 OpenAI and MIT studies ChatGPT's emotional impact

OpenAI, in collaboration with MIT Media Lab, has explored how interacting with ChatGPT affects users emotionally. Analyzing data from 40 million interactions and surveying over 4,000 users, they found only a small subset engages emotionally with the chatbot. Interestingly, those who "bonded" more with ChatGPT reported higher loneliness and emotional dependence. OpenAI sees this as a first step in understanding the long-term impact of AI on users, guiding safer and healthier digital interactions.

🧠 Jack Ma-backed Ant Group claims AI breakthrough on China-built chips 

Ant Group, supported by Jack Ma, has announced a breakthrough in AI training by leveraging Chinese-made chips from Alibaba and Huawei. The company claims its new technique reduces AI model training costs by 20% compared to global competitors like OpenAI and Google, all while delivering performance on par with NVIDIA’s H800 chips.

🍎 Apple wants to turn its watches into wearable AI

Apple is exploring AI-powered enhancements for future Apple Watch and AirPods models, with a target release in 2027. These devices could feature Visual Intelligence, enabling them to analyze surroundings and provide real-time contextual insights. The standard Apple Watch might embed the camera in its display, while the Ultra model could house it near the side button, bringing advanced situational awareness to everyday wearables.

🤖 1X will test humanoid robots in ‘a few hundred' homes in 2025

1X is preparing to launch early trials of its Neo Gamma humanoid robot in hundreds of homes by late 2025. During these tests, teleoperators will assist users, allowing them to choose when a 1X employee can access the robot's camera feed. This data will help refine AI models and enhance performance. A recent demo showcased Neo Gamma vacuuming a carpet, hinting at its potential to handle everyday household tasks.

💃 Perplexity proposes open-source version of TikTok algorithm 

Perplexity AI has proposed acquiring TikTok with a bold vision: rebuilding its recommendation algorithm with transparency at its core. The company plans to open-source the 'For You' feed, offering a glimpse into the mechanics behind content curation. To address privacy concerns, Perplexity also aims to store TikTok’s data in U.S.-based centers, ensuring compliance with local standards while reshaping the platform for greater accountability.

TRENDING TOOLS

• Flora > Every creative AI tool, thoughtfully connected.

• SynCity > Training-free generation of 3d worlds.

• HuggingFace Daily Papers > Find, read, and discuss the latest AI research papers, all in one place.

• Vimcal > Schedule meetings faster with AI that finds times, manages calendars, and creates events easily.

• Currents > Analyze social media in real-time to uncover trends, sentiment, and competitor insights for better decisions.

• Zapier MCP > Let your AI assistant interact with thousands of apps without complex API integrations.

THINK PIECES / BRAIN BOOST

• Humanoid Robotics: A rare 0-to-trillion opportunity.

• From screenshot to working Airbnb clone (backend, ui, db, etc) entirely through cursors new agent.

• The bitter lesson: Rethinking how we build AI systems.

• New jailbreak technique uses fictional world to manipulate AI.

• Why AIs sometimes make things up.

• Understanding AI vulnerabilities.

VALLEY GEMS

1/ A guy rebuilt what i worked on for 5 years in 8 days vibing in Cursor AI.

2/ Japanese scientists have created a hydrogel that reverts cancer cells back to cancer stem cells in 24 hours.

SUNSET IN THE VALLEY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.