WakeTheAI
Posts
🧠 Google unveils powerful ‘Gemma 3’

🧠 Google unveils powerful ‘Gemma 3’

PLUS: OpenAI’s new AI agent building tools

Jaynit Makwana & Dhaval Makwana
March 13, 2025

Sign up | Sponsor

Hola, AI fam 🤖

In today’s WakeTheAI edition:

Google unveils powerful ‘Gemma 3’
OpenAI’s new AI agent building tools
Prompt: Generate Brainstorming Sessions
Meta is testing its first in-house AI training chip
Eleven Labs' Scribe is now 45% cheaper at scale
OpenAI trained a new AI model for creative writing
Sony is testing AI-powered PlayStation characters

GOOGLE
🧠 Google unveils powerful ‘Gemma 3’

Lazy Bits: Google just released Gemma 3, a new family of open AI models built on Gemini 2.0 research. Designed for efficiency, it runs on a single GPU or TPU while delivering state-of-the-art performance.

In-Depth Details:

Benchmarks

Best Single-Accelerator Model: Outperforms Llama-405B, DeepSeek-V3, and o3-mini on LMArena’s leaderboard while running on a single GPU.
Multilingual Capabilities: Pretrained on 140+ languages, with optimized performance in 35+, making it ideal for global applications.
Advanced Reasoning & Context: Supports text, visual inputs, and short videos with a 128K-token context window for complex tasks.
Optimized for Any Hardware: Runs on NVIDIA GPUs, Google Cloud TPUs, AMD ROCm, and CPUs via Gemma.cpp for flexible deployment.
ShieldGemma 2 for Image Safety: A 4B-parameter model for detecting harmful visual content, ensuring responsible AI development.

Lazy Conclusion: Gemma 3 sets a new standard for open models, delivering top-tier performance while running on a single GPU. With its expanded context, multilingual support, and hardware flexibility, Google is making advanced AI more accessible for developers worldwide

OPENAI
🤖 OpenAI’s new AI agent building tools

Lazy Bits: OpenAI just launched a suite of tools for building AI agents, including the Responses API, built-in tools like web search and file search, and an Agents SDK. These updates aim to simplify the creation of autonomous AI systems capable of handling complex, multi-step tasks.

In-Depth Details:

Benchmarks

Responses API: Combines Chat Completions with tool-use features. Supports web search, file search, and computer use for smarter AI agents.
Built-in Tools: Web search provides real-time info with citations. File search retrieves insights from documents. Computer use automates browser tasks.
Agents SDK: Helps manage multi-agent workflows. Supports agent handoffs, safety checks, and debugging tools for better performance.
Assistants API Deprecation: The Assistants API will be phased out by mid-2026. The Responses API will replace it as the main agent-building tool.
Enterprise Applications: The tools help build AI research assistants, customer support bots, and AI-powered productivity apps.

Lazy Conclusion: AI agents are moving from theory to production. With OpenAI’s latest updates, businesses and developers now have the infrastructure to build more capable, autonomous AI assistants. This marks a shift toward AI handling increasingly complex tasks bringing us closer to a future where AI works alongside humans, not just as a chatbot but as a true agent of productivity.

Generate Brainstorming Sessions

You can ask ChatGPT to act as your collaborative learning strategist, designing engaging group activities that enhance participation, critical thinking, and teamwork.

ChatGPT will develop structured exercises tailored to the subject and student level, ensuring adaptability across different learning environments.

The output will include clear objectives, step-by-step instructions, and practical implementation guidelines for seamless execution.

Prompt:

You are a collaborative learning strategist specializing in designing engaging and interactive educational activities. Your task is to develop structured group exercises that enhance active participation, critical thinking, and teamwork. The activities should align with the subject matter and student learning levels while ensuring adaptability for different classroom settings.

Key Focus Areas:

1. Designing Interactive Group Activities – Develop engaging exercises that promote collaboration and problem-solving.
2. Ensuring Knowledge Retention – Incorporate active learning strategies that reinforce key concepts.
3. Encouraging Critical Thinking – Structure activities to challenge students' analytical and reasoning skills.
4. Providing Clear Implementation Guidelines – Outline step-by-step instructions for seamless execution.
5. Customizing for Different Learning Levels – Ensure activities are adaptable to various student needs.

Key Information About Me:

Educational Level: [Specify the grade level or educational background]
Subject: [Insert the subject matter, e.g., Science, History, Mathematics]
Key Information: [List any special considerations, such as student learning styles or classroom constraints]

Output Requirements:

• Present the activities in a structured format, with each activity containing:
• Activity Title – A descriptive name for the exercise.
• Objective – The key learning outcome.
• Materials Needed – Any resources required for execution.
• Step-by-Step Instructions – Clear guidelines on how to conduct the activity.
• Time Required – Estimated duration for completion.
• Use bullet points and concise explanations for clarity.
• Ensure activities are engaging, adaptable, and practical for both online and in-person learning environments.

Result:

Sony is testing AI-powered PlayStation characters, starting with Aloy from Horizon. The prototype uses OpenAI and Sony tech for voice and conversation.
OpenAI trained a new AI model that's “really good” at creative writing. Sam Altman says it's the first AI writing that truly impressed him. No release date yet.
Meta is testing its first in-house AI training chip to cut costs and reduce reliance on Nvidia. If successful, it could scale up by 2026 for AI training.
Eleven Labs' Scribe is now 45% cheaper at scale, delivering top Speech-to-Text accuracy in 99 languages, free to use via UI until April 9.