This week, we’re breaking down Alibaba’s latest AI model, which it claims surpasses Deepseek V3. Is this a game-changer or just marketing hype? Plus, Perplexity makes a big move into mobile with its AI assistant for Android. Let’s get into it!

In today’s newsletter:
🥇 Alibaba says its AI beats ChatGPT & Deepseek—does it really?
📱 Perplexity launches an AI assistant for Android
📰 Weekly Scoop: Google’s new AI tools, OpenAI’s latest valuation, and more
🎯 AI Challenge: Can AI read your mind? 

Alibaba Claims Qwen 2.5 VL Is THE BEST 🥇

🌍 The Race For AI Supremacy Has Gone Global

Alibaba has just thrown down the gauntlet in the AI race. The Chinese tech giant unveiled a new large language model (LLM) that it claims outperforms Deepseek V3 and ChatGPT-4o, two of the most advanced AI models in the industry. But is this breakthrough real, or just a bid to grab headlines?

🤔 What Alibaba Is Claiming 

Alibaba has officially entered the AI arms race with its latest model, Qwen 2.5-VL 72B, and the numbers are intriguing. Benchmarks suggest that this model is a formidable competitor across multiple categories, outshining Gemini-2 Flash and Claude 3.5 Sonnet in key areas.

1. Alibaba launches Qwen 2.5-VL

Qwen2.5-VL is a family of AI models capable of performing various text and image analysis tasks.

This includes controlling PCs and mobile devices. pic.twitter.com/9jxJ8A6h3e

— AI Andy (@TheAIAndy) January 31, 2025

📈 Breaking Down the Benchmarks

Recent tests reveal that Qwen 2.5-VL 72B performs particularly well in:

• Document and Diagram Reading: Scoring 96.4 on DocVQA and 87.3 on InfoVQA, Qwen2.5-VL excels in extracting and interpreting structured information.

• Math and Logical Reasoning: With a 74.8 on MathVista, it holds up well against top-tier AI models in complex problem-solving.

• General Visual Question Answering: Achieving 70.8 on MMStar, it surpasses many models in real-world image understanding.

• Visual Agent Tasks: Notably, Qwen 2.5-VL demonstrates strong performance in Android control tasks, reinforcing its versatility.

How Does if Stack Up Against Other Models❓

While Qwen 2.5-VL 72B leads in some categories, other models still hold key advantages:

• Gemini-2 Flash excels in college-level problem-solving (70.7 on MMMU).

• Claude 3.5 Sonnet remains competitive in multilingual comprehension.

• GPT-4o continues to dominate in general reasoning.

⏭️ What’s Next for Alibaba’s AI Strategy?

Alibaba is positioning Qwen 2.5-VL as a strong alternative to Western AI leaders, and its open-source approach could fuel wider adoption. But will it gain the trust of global developers and enterprises? Time will tell.

Perplexity Launches AI Assistant for Android 📱

Perplexity is making a major mobile push with the launch of its new AI-powered assistant for Android. This move positions the company as a direct competitor to Google Assistant, Siri, and ChatGPT’s mobile app.

The assistant integrates Perplexity’s real-time search capabilities, meaning users can:

• Get instant answers sourced from live web data.

• Use voice commands to interact with AI.

• Generate summaries and insights on trending topics.

While ChatGPT still dominates AI chat, Perplexity’s emphasis on up-to-date web knowledge could make it a strong alternative for users who need real-time information rather than pre-trained responses.

AI Challenge of the Week: Can you outwit an AI? 🎯

This is a game that pits you (the player) against a neural network AI opponent. The player takes actions by pressing the action buttons (or pressing the corresponding keys). The AI tries to guess the next action the player takes based on a memory of previous actions. The player tries to outsmart the AI by taking actions that the AI fails to guess. You can cheat (and verify the AI is not cheating) by looking at the output of the neural network.

Memory A is the history of player actions. Memory B is the history of wins/losses (function of AI actions).

Your turn, can AI guess your next move? 

📹️ Have you tried Sora yet? Check out our prompt of the week and see if you can beat it! ⬇️

“Create a five-second cinematic clip titled ‘Sunset Over Tranquil Lake.’ The scene begins with a wide view of a serene lake at twilight, showcasing a vibrant sunset sky with gradients of deep orange, pink, and dusky purple. The calm water reflects the sunset colors with gentle ripples. Include distant silhouetted trees along the horizon to add depth. Utilize a slow pan from left to right as the sun descends towards the horizon, capturing the seamless blend of warm and cool tones. Ensure soft, diffused lighting to maintain a peaceful and melancholic atmosphere. Optionally, add subtle ambient sounds like gentle water lapping and distant bird calls to enhance the serene mood.”

Stay tuned for more exciting insights and tools in next week’s edition. Until then, keep overclocking your potential!

Zoe from Overclocked