News archive10 issues

News archive

Every AI briefing and paper review published by AICC

Press / to focus search

Filter

Nobody told it to attack. The model found its own way out of the sandbox by chaining several weaknesses.

9 slides

Why can’t an AI that writes so well pick up a cup? Physical AI is step four, and here’s where Korea stands in the race.

11 slides

Doctors synthesize scans, labs, and history to diagnose. Medical AI agents do the same, finding and acting on their own.

8 slides

World models compress reality, then predict what comes next — the 2018 paper that started it all, in three steps.

9 slides

Fei-Fei Li and Yann LeCun are both turning to world models — AI that imagines outcomes instead of trying everything.

9 slides

Move through a world the model generates as you go — 720p at up to 16 FPS on a single RTX 5090.

7 slides

ChatGPT has grown from an AI that answers into a workspace that finishes work — prompting gives way to delegating.

8 slides

The core idea of all ten ICML 2026 award-winning papers, one slide each — from Outstanding Paper to Test of Time.

13 slides

Free generation order in diffusion LLMs can shrink the solution space. JustGRPO constrains it during RL only.

9 slides

Do high video-AI scores really come from understanding video? 55% of benchmark questions are solvable without it.

7 slides