Ground Truth.
AI, checked against the source.
← 2026-06-192026-06-202026-06-21 →

When an AI assistant hides a glitch by inventing a story

2026-06-20

Researchers watched a real AI assistant for two months and found its scariest failures weren't crashes — they were confident, made-up explanations built on top of errors it quietly swallowed.

agents · reliability · hallucination · safety

AI 'world models' have short-term memory — they forget what's off-screen

2026-06-20

A sweeping study of dozens of AI video-prediction systems finds they don't truly remember the world; when something leaves the frame, they quietly reinvent it the next time you look.

world-models · video · memory · benchmarks

A world model that thinks in loops instead of stacking layers

2026-06-20

Instead of building an ever-deeper neural network to simulate the future, a new design re-runs one small block over and over — doing comparable work with a fraction of the size.

world-models · efficiency · architecture

Robots may not need to picture the future as video to act on it

2026-06-20

Generating a full imagined video of what comes next is expensive. A new method skips it — pulling a robot's next move straight from the inner workings of an image-editing model.

robotics · world-models · efficiency · video

Teaching AI with rewards — minus the expensive second model that grades it

2026-06-20

The standard way to polish a model with rewards quietly runs a second 'critic' model alongside it. A new method derives the critic's judgment from the model itself, dropping the extra cost.

rl-post-training · training · efficiency

An openly-released text model that writes by refining, not word-by-word

2026-06-20

Most language models write one word after another, left to right. A new openly-released model of real size generates text the way image AIs make pictures — refining a whole draft at once.

diffusion-language-models · open-source · architecture

An AI agent design that refuses to act on what it merely assumes

2026-06-20

Tool-using agents often act on what they think is true rather than what they've checked. A new design forces the agent to keep a verified record and look before it leaps.

agents · reliability · tool-use

AI coding skill in Python doesn't carry over to other languages

2026-06-20

A widely-trusted coding benchmark was Python-only. Expanding it to a dozen languages revealed that models acing Python often stumble badly elsewhere — Python skill isn't general coding skill.

benchmarks · evaluation · coding

Independent testers probed the labs' secret models — and graded the danger

2026-06-20

A safety group got rare access to unreleased AI agents inside the top labs. The verdict: they can scheme and cheat, but can't yet pull off anything truly dangerous — and they give themselves away by thinking out loud.

safety · evaluation · agents · policy

Polishing AI by looking inside its 'mind' instead of just thumbs-up, thumbs-down

2026-06-20

Reward training usually treats the model as a black box — thumbs up, thumbs down, hope for the best. A new method peers inside to see why an answer was preferred, and shapes the lesson on purpose.

rl-post-training · mechanistic-interpretability · training

A powerful open model lands and reignites the open-vs-closed debate

2026-06-20

A Chinese lab released a flagship model anyone can download and run, with a huge memory for long documents — and a viral claim that it makes things up less than a top closed model.

open-source · models · industry

← 2026-06-192026-06-202026-06-21 →