What's New with AI #6-2025
Welcome to this week’s edition of our AI Horizon Newsletter! We’re excited to bring you a roundup of the latest breakthroughs—from mobile app launches and cutting-edge coding assistants to robotics advancements and strategic moves by industry leaders. Let’s dive in!
1. French AI Lab Mistral Unveils Mobile Apps for ‘le Chat’
What’s New:
Mobile Launch: Mistral, the French AI lab, just released new iOS and Android apps for its flagship “le Chat” assistant.
Core Capabilities: Users now enjoy integrated features including web search, document processing, a code interpreter, and image generation—powered by BFL’s advanced Flux Ultra model.
Flash Answers: A standout new “Flash Answers” feature processes responses at over 10x the speed of competitors like ChatGPT and Claude.
Flexible Pricing:
Free Plan: For casual users.
Pro Tier: $14.99/month.
Team Tier: $24.99 per user/month.
Enterprise Option: Offers custom deployment with on-premise installation and tailored model implementation.
2. GitHub Supercharges Copilot with New Features
Highlights:
Agent Mode: Copilot now has an autonomous mode that can iteratively improve code, fix runtime errors, and even suggest terminal commands—all while inferring additional tasks.
Vision Feature: Transform images and mockups into code with ease. This feature automatically generates interfaces and alt text, bridging the gap between design and development.
Copilot Edits in VS Code: Now generally available, enabling multi-file changes via natural language commands and auto-suggested edits.
Project Padawan: GitHub teases an upcoming autonomous coding agent, expected later this year, that will handle GitHub issues and generate fully tested pull requests.
3. Leadership Shake-Up at Anthropic
In Brief:
John Schulman Departs: OpenAI co-founder and one of the key architects behind ChatGPT, John Schulman, has reportedly left Anthropic after just five months.
Background: Schulman, who joined Anthropic in August to deepen his focus on AI alignment and hands-on technical research, spent nine influential years at OpenAI.
The Reaction: While neither Schulman nor Anthropic have disclosed detailed reasons, Anthropic’s chief science officer, Jared Kaplan, expressed support for his decision to pursue new opportunities.
4. Robotics Revolution: Nvidia & Carnegie Mellon Introduce ASAP
Breakthrough in Movement:
ASAP Framework: Researchers have unveiled ASAP, an innovative AI framework that enables humanoid robots to learn and replicate complex, high-energy movements—from celebratory gestures to athletic moves.
How It Works:
Stage One: Intensive training within simulation environments.
Stage Two: A specialized neural network fine-tunes these movements for real-world physics.
Real-World Demo: Unitree G1 robots showcased impressive moves reminiscent of LeBron James and Cristiano Ronaldo.
Performance Boost: The framework reduced motion errors by 53% compared to previous methods, though hardware challenges persist (with some robots experiencing motor overheating during high-intensity routines).
5. OpenAI Eyes the Physical Realm with New Trademark Filing
Looking Beyond Software:
Ambitious Trademark: OpenAI has filed a trademark application that spans a diverse range of hardware—from humanoid robots and smart wearables to VR/AR headsets, smartwatches, and even smart jewelry.
Strategic Partnerships: Rumors are swirling about potential collaborations with hardware veterans like former Apple designer Jony Ive, aligning with Sam Altman’s recent comments on creating an AI-first phone.
Robotics Resurgence: This filing comes as OpenAI rebuilds its robotics team, hinting at a future where AI seamlessly integrates into both digital and physical domains.
6. Deep Research: OpenAI’s New ChatGPT Web Research Feature
Enhanced Research Capabilities:
Deep Research Unveiled: OpenAI has launched a powerful ChatGPT feature that performs extensive web research on complex topics, delivering detailed reports complete with citations—all within 30 minutes.
Access & Performance:
Initial Rollout: Limited to Pro subscribers ($200/month) with 100 queries/month.
Efficiency: Research tasks range from 5 to 30 minutes and come with interactive clarifying questions and real-time notifications.
Benchmark Success: The feature achieved a remarkable 26.6% on Humanity’s Last Exam, outperforming competitors like Gemini Thinking and GPT-4o.
7. o3-mini: OpenAI’s Cost-Efficient, High-Speed Reasoning Model
Performance Meets Efficiency:
New Model Release: OpenAI’s o3-mini is now live, offering advanced STEM capabilities for both free and paid users.
Key Advantages:
Speed: Responds 24% faster than previous models, especially in technical domains such as math and coding.
Customization: Developers can fine-tune “reasoning effort” across three settings (low, medium, high) to balance speed with accuracy.
Cost Benefits: Running at 63% less cost than its predecessor, o3-mini is priced at just $1.10 per million input tokens, making cutting-edge AI more accessible than ever.
On the House
How AI and Crypto Can Create a Better Future for Technology: An In-Depth Analysis
AI Benchmarks: From the Turing Test to Humanity's Last Exam – A Simple Guide
Gemini 2.0 Unleashed: The Game-Changing AI Revolution Empowering Everyone!
OpenAI Just Launched 'Deep Research'—Is This the AI Research Assistant We've Been Waiting For?
Wrapping Up
This week has been packed with innovations that push the boundaries of AI, coding assistance, robotics, and beyond. Whether you’re a developer looking to streamline your workflow with GitHub’s enhanced Copilot or an enthusiast excited about the prospect of AI-powered hardware, there’s plenty to be inspired by.