The Flow has not summarised this story yet — read it at arXiv Computation and Language.
Sofroniew et al. (2026) showed that emotion concepts in Claude Sonnet 4.5 are encoded as vectors whose geometry mirrors human affect psychology. This study replicates that finding using the base pretrained model google/gemma-2-27b, generating 205,200 Claude Sonnet 4.5 stories, extracting 171 emotion vectors, and recovering a similar affective circumplex with principal components explaining comparable variance. The analysis further identifies a sharp geometric seam at layers 22‑26, demonstrates that much of the geometry already exists in static token embeddings, and shows that the geometry predicts token‑level co‑activation with high correlation.
arXiv:2606. 26987v1 Announce Type: cross Abstract: Recent work identified emotion vectors in Claude Sonnet 4.
The paper demonstrates that a single internal direction in modern language models—called the valence axis (V-axis)—captures how positive or negative a sentence feels. By using only nine emotion category names and 50 short narrative paragraphs per emotion, the authors identify this axis via principal component analysis of frozen encoder embeddings, achieving 93% of supervised performance on SST‑2 and strong correlations with human valence ratings across images, audio, and brain recordings. The method transfers across modalities without target‑modality labels, but works only for continuous attributes and is specific to certain model families.
arXiv:2610.27603v1 Announce Type: new Abstract: Empathetic interaction involves two distinguishable capacities: representing another person's emotional state (cognitive empathy) and generating an app...
arXiv:2508.05880v3 Announce Type: replace-cross Abstract: Understanding human emotions is central to user-facing AI applications, safety alignment, and the simulation of human behavior. As emotional...
The paper uncovers a systematic mismatch in modern text embeddings: while sentiment can be extracted with high accuracy from frozen encoders, the embeddings’ cosine similarity metric barely reflects polarity differences. Across eleven models from 2021–2026, a linear probe achieves up to 91.3% accuracy, yet cosine similarity margins are only 0.003–0.064, leading to poor retrieval performance (e.g., Qwen3-Embedding-4B’s 64.3 MAP). Introducing a lightweight supervised‑contrastive head (Sentiment‑Injected Sentence Embeddings) dramatically improves probe accuracy, retrieval MAP, and polarity margins, but at the cost of downstream semantic similarity. "whyItMatters":"The findings reveal that widely used retrieval and clustering systems may ignore valuable sentiment information encoded in embeddings, impacting their effectiveness."