Welcome to Grznar.eu
Personal website with blog about technologies, AI news, weather station and programming projects.
Featured Content
Latest articles, projects and AI news
Přehled významných firem a jejich AI modelů (stav 2025)
Jak umělá inteligence mění způsob, jakým píšeme kód
Read More
Gemini now lets users guide AI video with multiple reference images per input
Nejnovější pokroky v oblasti jazykových modelů
Read MoreRecent Posts
All PostsAI News
Gemini now lets users guide AI video with multiple reference images per input
Google is updating the Gemini app with a new way to control its AI video model. With the latest release, users can upload multiple reference images for a single video prompt. The system then generates video and audio based on those images combined with text, giving people more direct control over how the final clip looks and sounds.
Leaked finances hint that OpenAI's inference may be swallowing its revenue
Leaked finances suggest OpenAI's inference costs may be exceeding its revenue, with massive spending just to keep AI models running. Internal documents indicate that OpenAI paid Microsoft hundreds of millions of dollars in revenue share, and inference costs hit billions. The data suggests that OpenAI is already spending massive sums just to keep its AI models running, far more than many expected.
Microsoft CEO Satya Nadella promises a completely new Microsoft
Microsoft CEO Satya Nadella has declared that Microsoft will become a completely new company, effective immediately.
DeepEyesV2 outperforms bigger rivals by favoring tools over sheer knowledge
Chinese researchers have built a multimodal AI model that can analyze images, run code, and search the web. Instead of relying on knowledge acquired during training, DeepEyesV2 boosts performance by using external tools intelligently, allowing it to outperform larger models in many cases.
LeCun accuses Anthropic of exploiting AI cyberattack fears for regulatory capture
Yann LeCun accuses Anthropic of exploiting AI cyberattack fears for regulatory capture.
Anthropic steers Claude to acknowledge conservative positions to avoid the “woke AI” label
Anthropic steers Claude to acknowledge conservative positions to avoid the “woke AI” label.
OpenAI publishes prompting guide for GPT-5.1
OpenAI's guide for GPT-5.1 explains how developers can control the model's responses with detailed settings for tone, structure, and length. The documentation introduces new tools such as "apply_patch" and a restricted "shell" tool, aimed at reducing mistakes in programming tasks and improving workflow transparency. The guide also covers metaprompting, which enables the model to review its own prompts, detect possible errors, and suggest improvements to help manage complex system prompts more effectively.
OpenAI’s GPT‑5.1 Reddit AMA unraveled into a full‑blown karma massacre
OpenAI's Reddit AMA on GPT-5.1 quickly drew a lot of criticism, with many users complaining about the company’s model policies and what they felt was a patronizing attitude toward paying customers. Many users called for GPT-4o to remain available permanently and unchanged, saying it felt more empathetic and consistent. However, OpenAI pointed out that this could pose a problem, as it might lead to psychological risks. OpenAI acknowledged the criticism and said it plans to review the feedback. The debate highlights how strongly some users react to changes in AI models.
Researchers push "Context Engineering 2.0" as the road to lifelong AI memory
Researchers are calling for a fundamental overhaul of how AI handles memory and context. Their proposal: a Semantic Operating System that can store, update, and forget information over decades, functioning more like human memory than today's short-lived context windows.
Forget AGI—Sam Altman celebrates ChatGPT finally following em dash formatting rules
Sam Altman celebrates ChatGPT finally following em dash formatting rules.