Tech • IA • Crypto
Microsoft AI Code 1 Flash, OpenAI math shock, RTX Spark PCs
Briefing
Microsoft unveils full-stack AI suite
Microsoft signaled a निर्णायक shift toward independence with a full portfolio of in-house models, reducing reliance on OpenAI and Anthropic. Key launches include AI Code 1 Flash (5B parameters) optimized for lightweight coding and AI Image 2.5, ranked highly in user benchmarks. The company also introduced Transcribe 1.5 with support for 43 languages, alongside upgraded voice synthesis tools. The move positions Microsoft as a direct end-to-end competitor to Google and Apple in AI infrastructure and applications.
OpenAI model cracks Erdős conjecture
An advanced OpenAI reasoning model produced a credible disproof of the unit distance conjecture, a long-standing problem posed by Paul Erdős. The result challenges decades of assumptions about optimal geometric arrangements in combinatorial math. Notably, the system was a general-purpose model rather than a specialized math engine. The breakthrough highlights how rapidly AI has progressed from weak math performance in 2023 to surpassing elite human-level reasoning.
Nvidia RTX Spark powers AI PCs
Nvidia introduced RTX Spark, a unified CPU-GPU-memory architecture designed for AI-native computing in laptops and compact systems. Built in close partnership with Microsoft, Windows has been adapted to fully leverage this integrated design. The first device, the Surface Laptop Ultra, showcases high-performance local AI with reduced latency. This marks a broader shift toward on-device AI computing, echoing the model popularized by Apple Silicon.
AI leaders push DNA screening rules
Executives from OpenAI, DeepMind, Anthropic, and Microsoft are calling for mandatory safeguards in nucleic acid synthesis. The proposal includes screening DNA orders, verifying customers, and maintaining audit records to prevent misuse. Historical precedents such as the 2002 polio reconstruction and 2005 Spanish flu synthesis underscore the risks. Leaders warn that AI could accelerate the design of harmful biological agents if left unchecked.
Gemini 3.5 Flash reshapes AI economics
Google’s Gemini 3.5 Flash reflects a shift toward agent-based AI systems that act autonomously rather than respond passively. At the same time, pricing models are tightening, with tiers like €99/month Ultra and €219 premium plans introducing strict usage limits. Users must now manage AI as a scarce resource, balancing cost against productivity. The trend signals a broader transition from unlimited access to metered, enterprise-style consumption.
OpenAI Codex adds Sites platform
OpenAI launched Sites in Codex, enabling teams to build and deploy internal applications directly from prompts. The system integrates hosting, authentication, databases, and tool connections without manual setup. Organizations are already replacing static documents with interactive tools like dashboards and planning systems. This marks a step toward AI-generated software as a default workflow inside enterprises.
China unveils Pi humanoid robot
Jaka Robotics introduced Pi, a compact 4-ft humanoid with 27 degrees of freedom and a dual “fusion brain” architecture. The system separates high-level reasoning from low-level motor control, enabling more precise embodied AI performance. Designed for research and real-world tasks, it reflects China’s push beyond industrial robotics into humanoids. The launch intensifies global competition alongside efforts from Nvidia and other players.
Microsoft Majorana 2 advances quantum timeline
Microsoft claims a breakthrough with its Majorana 2 chip, potentially bringing practical quantum computing to 2029. A redesign using lead-based superconductors has improved qubit stability from microseconds to around 20 seconds. This enables progress toward logical qubits, a key milestone for reliable computation. If validated, the development could significantly accelerate timelines previously estimated closer to 2033.
Videos covered
Microsoft lâche OpenAI avec 7 IA
- •Microsoft shifts to full-stack AI independence
- •Launch of lightweight coding model
- •Advances in image generation
How a reasoning model cracked an 80-year-old math problem — the OpenAI Podcast Ep. 20
- •Breakthrough on an Erdős Conjecture
- •A Major Open Problem with Real Stakes
- •Unexpected Speed of Progress
RTX Spark : le PC devient IA
- •RTX Spark: a unified computing architecture
- •Designed for laptops and compact systems
- •Deep integration with Windows
Build and share interactive prototypes in Codex
- •End-to-End Design Workflow
- •AI-Guided Ideation
- •Automated Prototype Generation
Palantir CEO Alex Karp on Tokenmaxxing & Taste
- •Shift from hype to practical deployment
- •“Token-maxing” and shallow usage concerns
- •AI enhances but does not replace core systems
Palantir's AIPCon 10, Ramp Hits $44B, 60 Minutes Eyes Rogan | Diet TBPN
- •Industry coalition warns of biosecurity risks
- •Historical precedents highlight the threat
- •Reconstruction of deadly viruses already proven
🔴 Alex Karp LIVE from AIPCon 10
- •Industry unites over AI-driven biosecurity risks
- •From open science to potential threat vector
- •AI accelerates biological capabilities
China Just Built A Two Brain AI Robot: One Body, Two Minds
- •China’s compact humanoid “Pi”
- •Dual-brain architecture for embodied AI
- •Jaka’s shift beyond industrial cobots
Gemini 3 5 flash: Comment mieux utiliser l'IA agentique ?
- •Shift from Unlimited AI to Cost-Constrained Usage
- •Rise of Agentic AI Systems
- •New Subscription Economics
Microsoft : ordi quantique en 2029
- •A revised timeline for quantum computing
- •Majorana 2 chip redesign
- •Dramatic improvement in qubit stability
La lutte pour notre indépendance face à l'IA s'organise
- •OpenAI pushes toward proactive AI
- •Autonomous agents at the core of usage
- •Simplification… at the cost of control
Gemma 4 : l’IA multimodale locale
- •Open and locally deployable
- •Efficient performance on modest hardware
- •Competitive capability for most tasks
Introducing Sites in Codex
- •Rapid app creation inside Codex
- •Built-in infrastructure and security
- •Integration with existing tools
Anthropic Just Dropped Their Claude Skills Secrets (steal these)
- •Skills Defined as Functional Systems
- •Hundreds of Skills in Active Use
- •Nine Skill Categories Identified
Claude Code + GoHighLevel MCP = New Sales Meta
- •AI-CRM Integration Gains Traction
- •Operational Time Savings and Revenue Impact
- •High-Frequency Automation in Practice