#model-reliability

The latest AI and developer news about #model-reliability, with the original source and a short summary.

'Zero-Byte' Output Phenomenon Discovered in Major LLMs

A 'Zero-Byte' phenomenon in which output data is entirely absent has been commonly observed across major LLM models.

·1·2026.07.31 07:00

9 Judges, Really 2: Correlated Errors Undermine LLM Evaluation Panels

A study found that because models in LLM evaluation panels repeat similar errors, adding multiple models to the vote provides very little actual information value.

·2026.06.23 09:00

HALO-Loss Encourages Neural Networks to Refuse to Answer

Research has been released on training neural networks to abstain from answering uncertain questions through HALO-Loss.

·2·2026.04.14 14:00