Where Does Alignment Emerge and What Does It Mean? A Closer Look at the Platonic Representation Hypothesis
More capable language models still show greater alignment with vision models—provided the captions are sufficiently clean and informative.
The researchers examine where that similarity appears across model layers and representation dimensions. They contextualize the results with baselines, since model depth and width can confound similarity measurements. Separately trained vision and language models can show greater representational alignment than explicitly aligned vision-language models. But their alignment remains substantially below that of models working in the same modality.
HF Daily Papers' note
The researchers examine where that similarity appears across model layers and representation dimensions. They contextualize the results with baselines, since model depth and width can confound similarity measurements. Separately trained vision and language models can show greater representational alignment than explicitly aligned vision-language models. But their alignment remains substantially below that of models working in the same modality.
HF Daily Papers' note
score 4