Good evening! 🫡 🧠

We have just witnessed another CRAZY week in AI. From Chinese models running locally to Claude getting a literal "Think" tool.

The AI Train NEVER stops.

Let's dive into today's menu ↓

Today we're diving into🤿

🖥️ DeepSeek-V3: The 641GB Model Running on Your Mac
🎨 Grok 3 Introduces Image Editing by Text Prompts
🧠 Anthropic's Revolutionary "Think" Tool
⚡ Tencent's Hunyuan-T1: The Speed Demon of Reasoning
📰 AI News Roundup - Everything You Need to Know
🛠️ Tools of the Week

So, lean back and let’s go! ↓

_{PRESENTED BY SALESFORGE:}

🔮 Samurai Spotlight - Salesforge

Maximize Sales Pipeline Coverage with AI Agent Frank

Lower costs by letting Agent Frank take care of prospecting, crafting messages and booking meetings while your team can focus on closing deals!

Agent Frank can operate 24/7 and is fully customizable to best suit your company’s needs. By adding AI Agents to your team you can replace dozens of tools and scale your outreach without scaling your human team. Book a demo now to get a complete walkthrough:

Book a demo

🎨 Grok 3 Introduces Image Editing by Text Prompts

Elon Musk's GROK has rolled out an impressive new feature for Grok 3 that allows users to edit any uploaded image using simple text prompts. This capability, powered by the Aurora model, enables seamless image modifications without requiring specialized editing skills.

The feature was showcased by Musk himself on X (formerly Twitter), demonstrating how users can transform images with natural language instructions like "add a black hat to my picture" - with the AI quickly implementing the requested changes while maintaining visual consistency.

How It Works:

1. Upload your image to Grok 3

2. Select the "Edit Image" option

3. Describe your desired modifications using text prompts

4. Download your transformed image with a single click

According to user feedback, the tool is really good at preserving character consistency while making significant edits. The feature leverages Grok 3's multimodal capabilities and the Aurora model's image generation technology to create realistic visual effects.

This update positions Grok 3 competitively against other AI platforms like Google's Gemini, which has demonstrated similar capabilities but hasn't fully released them to the public yet. The image editing feature is currently available to X Premium subscribers through both the X platform and the standalone Grok app.

🖥️ DeepSeek-V3: The 641GB Model That Could Challenge OpenAI's Dominance

"The new DeepSeek-V3-0324 in 4-bit runs at > 20 tokens/second on a 512GB M3 Ultra with mlx-lm!" - AI researcher Awni Hannun

Chinese AI startup DeepSeek has quietly released a MASSIVE 641-gigabyte model that's sending ripples through the artificial intelligence industry. The model appeared on Hugging Face with virtually no announcement, continuing the company's pattern of low-key but HUGELY impactful releases.

The implications are ENORMOUS - this could seriously challenge OpenAI's business model, which relies on expensive data centers and cloud infrastructure.

Here's what makes DeepSeek-V3-0324 a potential GAME-CHANGER:

🔓 MIT license makes it freely available for commercial use
💻 Runs at 20 tokens/second on high-end workstations
🧠 685-billion-parameter model that rivals cloud-only competitors
🌐 Significantly reduced infrastructure requirements
💰 Lower operating costs for businesses and researchers

This 685-billion-parameter model demonstrates impressive performance on high-end hardware like Apple's Mac Studio with M3 Ultra. While the $9,499 price tag places it firmly in the professional category, it represents a DRAMATIC shift from the data center requirements typically needed for models of this scale.

DeepSeek-V3 is already showing IMPRESSIVE benchmark results:

• MMLU: 88.5% (compared to GPT-4o's 88.7%)

• HumanEval: 82.6% (compared to GPT-4o's 90.2%)

• MATH: 61.6% (compared to GPT-4o's 75.9%)

According to AI researcher Xeophon: "Tested the new DeepSeek V3 on my internal bench and it has a huge jump in all metrics on all tests. It is now the best non-reasoning model, dethroning Sonnet 3.5."

What do you think - will models like DeepSeek-V3 eventually reduce reliance on cloud-based AI services? Or will companies like OpenAI find new ways to stay on top?

🧠 Anthropic's "Think" Tool: Claude Gets a Brain Upgrade

"When Claude needs to carefully process the output of previous tool calls before acting and might need to backtrack in its approach" - Anthropic on when to use the "think" tool

Anthropic just unveiled a GAME-CHANGING new capability for Claude that fundamentally transforms how AI assistants work behind the scenes.

The new "think" tool extends Claude's agentic workflows through various applications (including the Claude Desktop App, Cursor, Windsurf, and more), essentially giving it a private space to reason through complex problems before responding.

Think of it like giving Claude its own mental scratch pad - a place to work through problems step-by-step without cluttering your conversation.

Details about the "think" tool :

🧠 Creates a private reasoning space for Claude to work through complex problems
📝 Records thoughts for later retrieval with "get_thoughts"
🧮 Processes statistics and analyzes data more thoroughly
🧹 Clears thoughts with "clear_thoughts" when starting fresh
🔍 Acts as a quality check to ensure reasoning aligns with desired outputs

When to use the "think" tool

Based on these evaluation results, we've identified specific scenarios where Claude benefits most from the "think" tool:

1. Tool output analysis. When Claude needs to carefully process the output of previous tool calls before acting and might need to backtrack in its approach;

2. Policy-heavy environments. When Claude needs to follow detailed guidelines and verify compliance; and

3. Sequential decision making. When each action builds on previous ones and mistakes are costly (often found in multi-step domains).

This represents a significant evolution in how AI assistants handle complex reasoning tasks - moving from simple input/output exchanges to more human-like thought processes.

You can see more here: https://www.anthropic.com/engineering/claude-think-tool

Does this make you trust AI outputs more?

⚡ Tencent's Hunyuan-T1: The Speed Demon of Reasoning

"Strong Logic & Concise Writing – Precise following of complex instructions" - Tencent on Hunyuan-T1's capabilities

While Western AI companies grab most headlines, Chinese tech giant Tencent just quietly released a model that could change the game for AI reasoning.

Meet Hunyuan-T1, a breakthrough in AI reasoning powered by Hunyuan TurboS and built for speed, accuracy, and efficiency. What makes this model particularly interesting is its unique architecture and BLAZING fast performance.

Here's what makes Hunyuan-T1 special:

🔄 First-of-its-kind hybrid Mamba-Transformer MoE architecture
⚡ Blazing fast - first character in 1 second, 60-80 tokens/sec generation
📚 Excellent long-text processing capabilities
🧠 Strong logic and precise instruction following
💬 Low hallucination rate in summaries

The hybrid architecture is particularly noteworthy - combining the sequential processing strengths of Mamba with the parallel processing power of Transformers, all wrapped in a Mixture of Experts approach that activates only the most relevant parts of the model for each task.

This represents a significant architectural innovation that could influence the next generation of AI models globally.

You can try Hunyuan-T1 right now through their experience site or Hugging Face demo.

📰 Latest AI News

🧰 Tools of the Week

1. Freepik

Access millions of premium design resources.

Create stunning visuals with ready-to-use templates and assets.

Perfect for marketers, designers, and content creators.

Try Freepik (Free plan available)

2. Gamma AI

Create beautiful presentations instantly using AI.

Transform ideas into professional slides with minimal effort.

Perfect for professionals, educators, and students of all skill levels.

Try Gamma AI ($8/mo)

3. Mistral AI

Access powerful open-source language models.

Build custom AI applications with state-of-the-art performance.

Saves me $20 each month compared to other enterprise LLMs.

Try Mistral AI (Free & paid plans)

Thank you for reading!

What's on your mind?

Feel free to share your ideas with me at: _{[email protected]} 

If your idea is good, I’ll make sure to bring it to life. Just send them to the above e-mail and I’ll get to work on making them a reality.