I’ve seen some promising work from Hume AI, Replika, and others in the applied space — mostly product-focused. But I’m curious if any open-source labs or communities are treating emotional intelligence as a missing pillar of alignment or general intelligence, alongside logic, memory, math, etc.
I’m personally interested in building an emotionally intelligent model (early idea: “Llapine ”) not as a product clone of a chatbot, but as a foundational capability experiment that could contribute to open alignment science.
Are there any public benchmarks, papers, or research collectives working in this space?
Happy to collaborate if anyone else is exploring similar things.
That’s an interesting take! But I’m curious — what makes you think emotional understanding requires image-based (or multimodal) memory?
Text-based interactions have been used in therapy for years (e.g., chat therapy, journaling), and classical literature has moved readers emotionally for centuries — all without needing visuals.
Of course, multimodality could enrich emotional perception, but it’s a different claim to say it’s absolutely necessary for modeling emotional intelligence. Would love to hear more about where that perspective comes from.
You’re confusing downstream effects with root causes.
Emotional understanding isn’t about whether text can evoke emotion in people it’s about whether a machine can store, contextualize and recall emotional state across interactions. Text based memory is lossy, linear, and stateless beyond sessions. It can’t persist emotional continuity without hallucinating structure. Image based memory allows for parallel context retention, nonlinear recall and dense state encoding all requirements for modeling emotional intelligence as a system, not a script.
You don’t simulate empathy by predicting the next token. You simulate it by storing and referencing internalized representations of prior states which requires more than text. Until that’s accepted, everyone’s just animating puppets with sentiment dressing.
TLDR, humans think and operate via symbolic live compression. It’s how our brains work and how we control every part of our body otherwise our nervous systems would fry. If you want emotions, you need to think the same way we do. With images.
”) not as a product clone of a chatbot, but as a foundational capability experiment that could contribute to open alignment science.
