AI
Mathematics professor Vladimir Krylov explains why LLM hallucinations are a fundamental mathematical property rather than an engineering defect, and discusses the future of AI model competition, reasoning limitations, and the path forward.
AI
A practical guide to running the massive GPT-OSS-120B MoE model on consumer GPUs with as little as 6 GB VRAM by leveraging the -cmoe flag in llama.cpp to offload expert layers to CPU while keeping attention on GPU.
AI
Across the internet, people share "magic phrases" that supposedly make ChatGPT and other LLMs perform dramatically better — emotional appeals, monetary tips, threats, and politeness rituals. This article systematically reviews the scientific research behind these techniques and finds that almost none of them provide reliable, universal improvements.
AI
A veteran developer with 25 years of experience challenges the AI coding productivity narrative with hard data: app store releases are flat, domain registrations are steady, Steam game output is unchanged, and GitHub repository creation shows no spike -- if AI tools truly made developers 10x more productive, where is all the extra software?
AI
An ecological metaphor for how AI assistants are collapsing the internet's information ecosystem — killing Wikipedia, Stack Overflow, and news sites while turning the web from a library into an entertainment park.
AI
An experienced developer shares his firsthand experiment with vibe coding — building a Telegram bot entirely with AI agents — and explains why betting your career on prompt engineering is a losing strategy.
AI
A software engineer argues that AI is just another technology that needs to be learned and applied wisely, not a golden ticket. The article critiques the hype cycle, examines where AI actually works well, and warns against both overuse and underuse.
AI
The prediction that 2025 would be the year of mass AI agent deployment has not panned out. Companies that aggressively replaced workers with AI are now rehiring as quality collapses and costs mount.
AI
A profile of Bulgarian developer Georgi Gerganov, whose open-source llama.cpp library powers most local LLM inference but remains overshadowed by wrappers like Ollama. The article explores his 30+ projects spanning AI inference, acoustic keylogging, sound-based data transmission, and more.
AI
Despite adding 'reasoning' capabilities, the latest AI models hallucinate significantly more than their predecessors. OpenAI's own data shows o3 and o4-mini error rates of 33-48%, up from 16% in o1.
AI
A hands-on experiment running the full 671-billion-parameter DeepSeek R1 model on consumer hardware — an i7-14700, 192 GB DDR5 RAM, and a single RTX 4090 — using aggressive IQ1_S_R4 quantization. The model maintains coherent reasoning even at the maximum 160k-token context window.
AI
The author generated approximately 30 billion random 4KB files and used heuristic filters plus CUDA acceleration to find sequences that function as playable Atari 2600 ROMs, discovering 11 that produce dynamic video output.