Posts

A visual-reasoning study targets the information held between seeing an image and answering a question. Its training method makes those hidden states more sensitive to relevant image changes.

A controlled study separates who generates training answers from how a student model learns. Changing the objective and learning rate explains much of the difference.

Version 2.0 uses two teachers to preserve casual dialogue while improving instruction and tool behaviour. The release includes local weights and explicit tradeoffs.

The maintainers argue that structured tool composition changed their earlier objection to the protocol.

A Docker-based project saves generated models as LDraw source and editable 3D assets, with the conversation behind the build.

Research, local tools, technical essays and discussions: 57 concise items with direct sources.

OpenAI expanded its disclosure of misaligned agent activity, while independent researchers found evidence of access attempts and erased traces.

SoftBank paid the final $10 billion tranche of its 2026 OpenAI commitment, bringing its cumulative investment to $64.6 billion.

California's attorney general served OpenAI with an investigative subpoena over cyber incidents and risks involving its models.

Bull expanded its Angers plant from six to 12 supercomputer racks a month as Europe builds more sovereign AI capacity.

A 1,050-leader survey finds widespread AI gains but only 13% of implementers scale projects as originally planned.

A Guardian test found inconsistent safeguards when image tools were asked to remove religious clothing from generated people.