One post tagged with "semantic-router"

Open Embeddings adds zero value if agents are trapped in their own "spaces"​

Pre-Reading:

• The Platonic Representation Hypothesis
• Harnessing the Universal Geometry of Embeddings
• Matryoshka Representation Learning

What to do about this​

Others have already showed a GAN can be trained to generate a shared latent space between arbitrary embedding models. What has not been shown:

• Do the findings hold at scale with modern / proprietary models?
• What's the smallest possible transform between spaces that doesn't lose (that much) semantic shape / geometry?
• Can we leverage Matryoshka Representations to further reduce the friction to move between worlds?

Yes, we can (probably)!​

Albeit with some trusty AI helpers, due to staffing shortages (we're looking for unpaid interns), we think the answer is that we can publish adapters that allow textual content to be "cheaply" discovered. As an exercise to the community, we can let indivual transforms get revised by popularity / need.