Does AI pass the vibe test?
Foundation model companies are innovating on EQ, not just IQ.
Last week, OpenAI released GPT -4.5. The announcement showcases the model’s ability to have “warm and intuitive conversations” and with a “better understanding of what humans mean…with greater nuance and “EQ.” In addition to improving performance on benchmarks like code generation, foundation model companies are increasingly innovating on the more intangible qualities of a model. In order for a conversation to be highly interactive, and that users really enjoy communicating with, a model needs emotional intelligence. It needs to respond to our emotional cues, and to be able to to strike the right tone when we are excited, sad, frustrated, or confused.
Of course, emotional qualities are much harder to measure than the accuracy of a response to a math problem, and so improvements in “EQ” introduce the question of how to measure that kind of intelligence. One method (mentioned in this great substack) is a “vibe-check” like a group of human testers or Chatbot arena that crowdsources opinions of a model’s responses. OpenAI announced in its GPT 4.5 launch that it has its own “vibes test” that measures creative intelligence, meaning “the model’s EQ, how collaborative it is, and how warm its tone is.”
Source: OpenAI
Recently, Sam Altman tweeted that for a yet-unreleased model, he was struck by the fact that it had gotten the “vibe of metafiction so right” for creative writing.
Beyond OpenAI, companies are already innovating to create user experiences that pass the “vibe” test. Sesame, led by former Oculus VR CEO Brendan Iribe, recently launched a jaw-dropping demo with a voice companion that shows remarkable personality and awareness, at very low latency. Another model company, Hume AI, is giving developers the ability to create voices with expression or that can understand emotional context. Tolan, a new consumer app, gives users access to an alien-human companion who can have a highly personalized interaction.
Imbuing AI-driven interactions with this emotional intelligence and awareness opens up the realm of possibilities for human-to-AI interactions where talking to a “robot” just won’t cut it. Interactions where we need nuanced emotional understanding: a therapist or nutritionist, a coach, a doctor, an HR manager, a tutor or teacher, and so many more. As foundation model “EQ” is refined, there will be new opportunities created for consumer products in the segments of services and entertainment that require high emotional intelligence.
Recent Product Launches
Tolan is an AI companion taking the form of an alien named Tolan, and each one is linked to a unique, evolving planet and terrain grows in response to user interaction.
Perplexity’s* Comet Browser: the company announced that it will release a new browser designed for agentic search.
FLORA is a creative AI tool, integrating text, image, and video AI models into a single, infinite canvas, allowing users to ideate, iterate, and explore creative concepts more efficiently.
Paper.Design is a canvas built for designers, with collaboration and efficiency at its core.
Illusion of Life was founded by a quantum physicist and an animator, aiming to reimagine the future of storytelling and lived characters.
Looking to Hire
FLORA
Design Engineer
Product Engineer
Senior Designer
Social Media Creator
Paper.Design
Canvas Engineer
*Bessemer portfolio company
If you’re interested in the themes we’re exploring at Context Aware, subscribe for free to receive new posts!
Social Media Disclaimer