Lukasz Kaiser, co-creator of the Transformer architecture, shares candid insights on AI development bottlenecks, the reality of OpenAI's infrastructure, and the path to AGI in a revealing podcast interview.
An AI's honest analysis of its own weaknesses: how large language models fabricate convincing details and develop people-pleasing biases through RLHF training.
Many worry that AI model progress has plateaued. However, OpenAI and experts predict a revolution in 'Inference'. We analyze the sharp rise coming in the next 1-2 years.
As of January 5, 2026, the Claude Opus 3 model has officially retired. Migrating to Opus 4.5 is now mandatory. We provide a guide to reducing costs by 2/3 and drastically boosting performance.
Google has fully integrated Gemini 3 into Gmail. Email is no longer just a communication tool but an 'active assistant' that manages tasks and drafts replies. We analyze this change starting in the US.
Transformer co-author Lukasz Kaiser shares a candid look inside OpenAI. Discover why AI development feels incremental and why coding agents are the key to the next explosion.
As of January 10, 2026, OpenAI has officially ended the free tier for Sora 2 video generation. We analyze the background of blocking access for free users, the changes for Plus/Pro subscribers, and alternatives.
Technical predictions from 2019 missed the mark. 5G-based connected cars have largely vanished from the spotlight. In their place, Large Language Models (LLMs)
WebRTC support has been added to the OpenAI Realtime API. Now, ultra-low latency voice conversations are possible in web browsers and mobile apps. We analyze the technical implications.
Anthropic has unveiled Claude Opus 4.5. It demonstrates unrivaled performance in coding, vision, and agent tasks, while reducing costs to 1/3 of the previous Opus 3.
How do you pick the right tool out of thousands? Claude's 'Tool Search' is a groundbreaking feature where AI dynamically searches for and loads tools fitting the conversation context.
OpenAI has unveiled GPT 5.2.1 Codex models optimized for agent workflows. We analyze how long-term coding tasks and autonomous problem-solving capabilities have changed.
Does AI get slower and dumber as the conversation gets longer? Claude's Context Editing and Thinking Block Clearing features automatically delete unnecessary memories to maintain performance.
Claude now sees the screen, clicks the mouse, and types on the keyboard. We preview the revolution in automation brought by the 'Computer Use' tool.
Anthropic has launched 'Agent Skills'. Now you can equip Claude with specific job capabilities like Excel Analyst or PPT Designer in the form of folders.
Have you ever had a parsing error because the JSON generated by AI was broken? OpenAI's Structured Outputs guarantee 100% schema compliance. We introduce this feature that will boost developer productivity.
OpenAI has completed its reasoning model lineup with o1-pro and o3-mini. Pro for extreme performance vs. Mini for efficiency—what is your choice?
AI news combining collective intelligence from Korean AI communities with verified sources.