LIM Center, Aleje Jerozolimskie 65/79, 00-697 Warsaw, Poland
+48 (22) 364 58 00
ts@ts2.pl

Category: Access

Europe’s New AI Code of Conduct: Inside the Plan to Tame Big Tech’s Models by 2025

The final General-Purpose AI Code of Practice was delivered to the European Commission on July 10, 2025. The EU AI Act will apply on August 2, 2025, with a one-year grace period for new GPAI models and two years for existing models before penalties. Adhering to the GPAI Code creates a rebuttable presumption of conformity…
Read more

AlphaEvolve: DeepMind’s Gemini-Powered AI That Invents Algorithms and Breaks a 56-Year Record

In mid-2025, Google DeepMind unveiled AlphaEvolve, an AI coding agent that uses Google’s Gemini LLMs, an evolutionary search process, and automated testing to autonomously design and improve algorithms. AlphaEvolve generates, tests, and evolves programs in an autonomous pipeline, refining entire codebases rather than a single function. It discovered a 4×4 matrix multiplication method over the…
Read more

32B AI Model Trained by a Swarm of Volunteer GPUs – Inside INTELLECT-2’s Decentralized Revolution

In May 2025, Prime Intellect unveiled INTELLECT-2, a 32B-parameter LLM trained on a globally distributed swarm of volunteer GPUs. INTELLECT-2 is the first model of its scale trained via fully asynchronous reinforcement learning across hundreds of heterogeneous, permissionless machines on chakra.dev. PRIME-RL coordinates the training loop by separating experience generation on inference workers from policy…
Read more

The “HTML of the AI Era”? Microsoft’s NLWeb Aims to Put ChatGPT on Every Website

NLWeb is an open protocol unveiled by Microsoft at Build 2025 to put ChatGPT-like conversational interfaces on websites. NLWeb works by indexing a site’s content using HTML, RSS feeds, and Schema.org metadata and exposing it to AI models via a standard JSON over REST API with a core ask method. The system is model-agnostic and…
Read more

Mixture-of-Memories (MoM): The “Linear Attention” Breakthrough That Doesn’t Forget Long Contexts

MoM was released in May 2025 as a linear-attention sequence model that preserves long-term context without forgetting. MoM breaks the single-memory bottleneck by maintaining multiple independent memory states and using a token router to assign information to memory slots. A trained router assigns each token to the top-k most relevant memory modules, with typical settings…
Read more

Nothing Phone 3: Big Leaks Confirmed, 50MP Quad Cameras, and How It Compares to 2025’s Best

The Nothing Phone 3 was officially unveiled on July 1, 2025, with pre-orders opening July 4 and a full market release on July 15, 2025. Prices start at $799 for 12GB RAM/256GB storage and $899 for 16GB RAM/512GB, with India pricing ₹79,999 and ₹89,999. The 6.67-inch flat AMOLED display offers 1260×2800 resolution (460 ppi), 20:9…
Read more

Pixel 10’s AI Reveal, Foldable Phone Showdowns, and iOS 26’s “Liquid Glass” Revolution – Mobile Tech News Roundup

Google scheduled its Made by Google event for August 20, 2025 in New York City, where four Pixel models—Pixel 10, 10 Pro, 10 Pro XL, and Pixel 10 Pro Fold—will debut alongside the Pixel Watch 4. The Pixel 10 family will run Android 16 with a redesigned UI and be powered by Google’s Tensor G5…
Read more

Apple’s First Foldable iPhone: Stunning $1,800 Price Leak, Ultra-Thin Design & Samsung’s Secret Tech

The iPhone Fold is tipped to debut in the second half of 2026, with mass production of key components planned to begin by late 2025. UBS analysts estimate a launch price around $1,800, with a bill of materials about $759 and gross margins of 53–58%. The device is described as a book-style foldable with a…
Read more

AWS Kiro AI: Amazon’s Bold New Agentic IDE Turning ‘Vibe Coding’ into Viable Software

AWS unveiled Kiro AI in mid-July 2025 as a specification-driven agentic IDE, available in a free public preview on kiro.dev, and powered by Anthropic Claude Sonnet 4 (with Claude 3.7 as a fallback) while built on a VS Code (Code-OSS) core. Kiro converts plain English prompts into full project specs—requirements.md, design.md, and tasks.md—before writing code,…
Read more

Google’s Gemini AI: The Multimodal Supermodel Aiming to Outshine GPT-4 and Beyond

Gemini debuted in December 2023 as Google DeepMind’s multimodal AI with Ultra, Pro, and Nano variants. Gemini is trained as a native multimodal model that can process text, images, audio, code, and video. Gemini Ultra achieved 90.0% on the MMLU test, leading on 30 of 32 academic benchmarks and beating GPT-4’s ~86.4%. Gemini 1.0 released…
Read more