I've been trying to enable a world with personalized content generation since the summer of 2015. I graduated with a Mathematics degree and then built the first end-to-end Generative AI-based model for Alexa's voice (which was a dramatic improvement over the previous stuff that went into Hawking's voice). I then spent a couple of years researching data efficiency for Generative AI models, unsupervised learning, and disentanglement (learning interpretable factors in speech generation - like speaker identity/accent - and being able to control them in human-interpretable ways). Some of this work has been published and patented. Toward the end of my time at Amazon, I spent some time working with Sparse Gaussian Processes in the Inventory Optimisation team. I left quickly after realizing my heart was still more interested in what I had to set out to do in the summer of 2015 i.e. personalized content generation, and started a company with a friend to enable that world instead.
Over the past 4.5 years of my professional career, I've worked extensively with VAEs, Normalising Flows, various Transformer models, & GANs. Even though I trained my first waveform level "language model" (convolutional, not transformer :)) in 2018 (it took more than 2 months for a single model to converge back in 2017), I continue to be shocked daily by their power and capabilities. Discovering what they can do continues to feel like magic.
If this world interests you, please reach out for a coffee, I'd love to meet you.