Summary of "Що нового сталось у світі AI? GPT-5 глючить, Маск розводить ср*ч, а Google готується в школу"

Model Changes: GPT-5 replaced previous selectable models (e.g., 4.0, 4.5, 3.0 Pro) with a single "router mode" where GPT-5 dynamically routes user queries to different internal models and tools.
User Experience: This router mode is buggy and unpredictable. Responses vary widely on repeated queries, sometimes providing great answers, other times poor or illogical ones.
Focus: GPT-5 prioritizes accuracy and knowledge over emotional or creative responses, which some users dislike.
Interface Update Proposal: Elon Musk suggests adding user-selectable thinking modes (quick, slow, pro) and reintroducing legacy models as options.
Use Cases: GPT-5’s complex multi-model approach works better for complex tasks like legal reviews or financial consolidations but struggles with straightforward creative or linear tasks.

Controversy: Elon Musk publicly criticized Microsoft, Apple, and OpenAI’s app rankings, threatening legal action against Apple over App Store ratings.
Community Mobilization: Musk urged his user base to boost his Grok AI app ratings on the Apple Store.
Grok AI Features: Recently released Grok 4 model is available free for limited time on Twitter and via the X AI app, including image and video generation capabilities.
User Base Comparison: OpenAI’s ChatGPT has ~700 million monthly active users, while Grok has about 40 million.

New Feature: Google Gemini introduces a "Guided Learning Mode" designed for educational use, helping users learn step-by-step with explanations, diagrams, videos, and interactive content.
Target Audience: Students preparing for the school year, aiming to make AI a learning assistant rather than a solution provider.
Previous Chat Access: Gemini is starting to add the ability to access and continue previous conversations, currently for pro users.

Context Window: Claude Sonet 4 now supports a context window of 1 million tokens (~75,000 lines of code), five times larger than before.
Learning Mode: Claude introduces a learning mode for programmers, guiding users through code creation interactively, similar to having a senior engineer mentor.
Previous Chat Access: Claude now allows users (max subscribers) to access and resume previous chats, improving long-term conversation continuity.

Excel Automation: Manos Agent adds multi-agent functionality to generate complex Excel spreadsheets with formulas and bookmarks.
Performance: Tested on a complex Apollo 13 mission task, Manos generated a spreadsheet but with poor quality and incomplete features.
Comparison: GPT-5 in thinking mode produced a better Google Sheets solution with code and plugin integration, though some visualization issues remain.

Cursor Updates:
- Version 1.3 added terminal mode, token usage tracking, and improved quick actions.
- Version 1.4 introduced asynchronous task execution and GitHub integration with commit pushes.
- Added a CLI (command-line interface) similar to Claude Code and Gemini CLI for terminal-based coding.
V0 platform:
- Launched a full UI with an AI agent capable of planning, designing, debugging, and coding.
- Offers free daily credits for testing.
- Uses proprietary composite models claimed to outperform top commercial AI models (OpenAI, Claude, Gemini).
- Interface praised for quality and usability, resembling popular platforms like Lavable.

Multi-model Routing vs. Legacy Models: GPT-5’s dynamic routing is innovative but unstable; users want fallback to older models.
Learning Modes: Google Gemini, Claude, and Claude Code emphasize guided learning modes, especially for education and programming.
Previous Chat Continuity: All major AI platforms are adding features to recall and continue past conversations.
AI in Code Development: Cursor and V0 are enhancing AI-assisted coding with CLI support, task management, and GitHub integration.
Elon Musk’s AI Ecosystem: Musk pushes Grok and X AI aggressively but faces challenges competing with OpenAI’s dominance.

Main Speakers / Sources