AI Voice Imitations, ChatGPT Data Controls and Startup Funding

In this April 2023 roundup, I discuss synthetic music, a ChatGPT data-control update and funding for Replit and Pinecone. These are reports and expectations from that week, not current product or financing information.

The viral song “Heart on My Sleeve” used an AI-generated imitation of Drake's voice without his participation, and also featured an imitation of The Weeknd. I find the result striking and compare the music industry's response to its earlier struggles with digital music. My view is that the industry should explore ways to adapt and compensate artists.

I also mention a remix of “Munch” using an imitation of Drake's voice and his negative reaction. I speculate that authorizing such uses might dilute the brand of an already prominent artist, while other artists might see an opportunity. Artists' interests and permissions differ; a synthetic voice does not itself imply consent.

Grimes publicly proposed sharing royalties equally on successful AI-generated songs using her voice and said she was independent of a label. I find that approach interesting, while noting that other artists may face different contractual constraints. I also discuss her proposed registration and royalty mechanism and my interest in smart contracts as a possible way to administer agreements.

Other artists have also been imitated. I expect synthetic songs to pose a significant challenge for the music industry and raise the concern that cheap, abundant releases could saturate the market. I do not have a clear answer to how the industry should handle quality and rights at scale.

OpenAI announced an option in ChatGPT to turn off chat history and training for new conversations. I show the setting as it appeared in 2023 and describe it as a way to reduce use of those chats for training. The feature should not be treated as a blanket promise of anonymity or zero retention; current settings and terms may differ.

OpenAI also mentioned a forthcoming business offering and a way to export chat history. I speculate that business users might receive more administrative controls; those details were not established by the brief announcement I discuss. Exporting data can make it easier to move or analyze conversations, although I note potential risks and terms around training other models.

Replit, known for a browser-based coding environment, announced a $97 million Series B extension at a reported $1.16 billion valuation. I say the funding will support cloud services and AI-assisted software creation, and compare its planned coding model with GitHub Copilot. The comparison is an expectation, not a head-to-head evaluation.

I also anticipate GitHub Copilot X, which was presented as bringing conversational assistance into development workflows. My own experience using ChatGPT for coding has made me more productive, including when working in other languages; that is a personal assessment.

Pinecone is a vector database service. I explain the basic retrieval pattern: represent text as vectors, store them, retrieve relevant material for a query and provide that material as context to a language model. This can help select useful information for a limited prompt, but does not itself give the model permanent human-like memory.

Pinecone announced a $100 million Series B at a reported $750 million valuation, with Andreessen Horowitz and other investors participating. I connect investor interest to the surge in attention around ChatGPT and language-model applications, and note that I use Pinecone myself. The reported valuation and market context are from 2023.