This Week in AI with The Neural

TECHIES, ASSEMBLE!

AI Industry

Google Brings AI to the Physical World with Gemini Robotics

DeepMind is integrating its Gemini 2.0 models into robots, giving them advanced reasoning and real-world awareness. Partnering with Apptronik, Google is bringing AI-powered robots closer to real-world deployment.

Discover the future of AI-powered robotics.

Anthropic’s Claude 3.7 Sonnet—Now with ‘Think Harder’ Mode

Anthropic has introduced a game-changing “extended thinking” mode for Claude 3.7 Sonnet, allowing deeper reasoning for complex problems. Users can now control the AI’s thinking time and budget, making decisions more transparent.

See what this means for AI alignment.

Google Unveils Gemma 3 Models

Google’s Gemma 3 is here—an open-source AI model built on Gemini 2.0 research. With strong multilingual support, efficient performance on single GPUs, and enhanced text and visual reasoning, it’s a powerful tool for developers. Plus, Google introduced ShieldGemma 2 for better image safety.

Find out how developers can use it.

OpenAI’s New Agent Tools—Game-Changer for Developers?

OpenAI has launched the Responses API and Agents SDK, making it easier to build autonomous AI systems. With built-in web search, file access, and safety features, these tools simplify multi-agent workflows.

Learn how businesses can use OpenAI's latest tools.

Meet Manus—The AI Agent That Thinks and Acts

Manus has unveiled a general AI agent designed to bridge thought and action. Whether it’s financial analysis, travel planning, or research, this AI performs at an impressive level.

See how Manus stacks up against other AI agents.

Baidu Takes on DeepSeek with New AI Models

Baidu just dropped Ernie 4.5 and Ernie X1—two AI powerhouses. Ernie 4.5 boasts top-tier language skills, logic, and emotional intelligence, while Ernie X1 is a budget-friendly reasoning model built to rival DeepSeek R1. Both come with multimodal capabilities and will be open-sourced by June 30th.

Read more about Baidu’s AI Models.

Mistral Small 3.1—The Open-Source Challenger

Mistral AI just released Mistral Small 3.1, a high-performance, multilingual, multimodal model with an extended context window and lightning-fast inference speeds. It’s open-source under Apache 2.0 and set to challenge proprietary AI leaders.

Explore Mistral 3.1's potential.

AI in the world around us

Grok’s Multilingual Wit Stuns Indian Users

Elon Musk’s AI chatbot, Grok, left Indian users amazed when it seamlessly responded in Hindi and Telugu on X (formerly Twitter). With slang-filled, witty comebacks, Grok is redefining cultural adaptability in AI chatbots.

Check out Grok’s unexpected linguistic skills.

Can AI Judge AI? Meet Judge-Image

Patronus AI has launched Judge-Image, a multimodal model that evaluates AI-generated image captions for accuracy. Used by Etsy to verify product descriptions, it helps detect hallucinations and biases. Built on Google’s Gemini, it outperforms OpenAI’s GPT-4V in bias reduction.

See why AI self-evaluation is critical.

OpenAI & Google Urge the U.S. to Step Up in AI 

OpenAI and Google are warning that policy inertia could cost the U.S. its AI leadership. They’re calling for stronger investment in AI research and ethical deployment as global competition intensifies.

Find out why they’re pushing for action.

Chinese AI Firms Ditch NVIDIA—And Still Win

A Chinese AI team has cracked video generation without NVIDIA’s industry-dominant GPUs, proving AI innovation doesn’t have to rely on a single hardware giant. By leveraging industrial chips, they delivered high-quality results and even won a global award.

See how this impacts the AI hardware market..

AI in Schools—The New Cheating Epidemic?

Students are turning to AI tools like ChatGPT and Gemini to cheat, leaving educators scrambling to detect AI-generated work. While AI can enhance learning, its misuse is raising serious concerns about academic integrity. Schools are testing AI-integrated learning and detection tools, but effectiveness remains uncertain.

See how schools are tackling AI cheating.

Teaching Robots with a Simple Nudge

MIT researchers have developed a way for humans to guide robots in real time—just by pointing, tracing paths, or nudging them. This eliminates the need for retraining, making robots more adaptable to human environments.

Discover how this improves human-robot collaboration.

‘How to’ with AI

Tree Search Supercharges AI Agents

Carnegie Mellon researchers have developed a breakthrough tree search method that makes AI agents smarter at navigating the web. By treating interactions like a branching tree, these agents can backtrack, learn from mistakes, and find the most efficient paths for tasks like online shopping or forum interactions.

This could redefine how AI interacts with the internet. Learn more here.

AI - Point of View

AI Transparency vs. ‘Open-Washing’—Who’s Really Open?

Endor Labs is calling out companies that claim to be “open” while keeping their AI under tight restrictions. They’re pushing for real transparency—clear communication about AI capabilities, limitations, and data usage to build trust.

See why this matters.

Honors in AI

Markus Buehler Wins the 2025 Washington Award

MIT professor Markus J. Buehler has been honored with the 2025 Washington Award for his groundbreaking work in computational modeling of biological materials. His research bridges engineering and nature, driving innovation in material science.

Read about his contributions.

From The Neural - AI IN Her Voice

AI Trailblazer: Shilpa Rao’s 2025 Blueprint

Tune in as TCS' Shilpa Rao reveals her inspiring AI journey, breaking barriers and sharing the one must-have skill for 2025. Don’t miss this game-changing convo!