Prediction from a16z @Jennifer Hli and @stuffyokodraws: This year artificial intelligence infrastructure will be split into two markets: LLM and diffusion models.
The developer tools that power the creative market will see tremendous growth.
Here’s why 🧵 and what developers should know 👇
1/While we’ve been asking “how to put artificial intelligence into production”, the reality is that artificial intelligence is already in production at an alarming rate in the creative marketplace: helping artists paint, stitching together new ideas, creating design assets (proven output by @ideogram_ai and @midjourney) The list goes on
2/As a developer, I clearly see how “different” the workload characteristics of the diffusion model are compared to LLMs. Historically, different workloads have introduced different tools to the market: transactional vs. analytical databases, batch vs. stream processing
3/So what will happen this time?
4/First, how to use the LLM with the diffusion model introduces a new layer of abstraction: for LLMs, the abstraction is @LangChainAI@llama _index
However, the diffusion model abstraction exists in use cases with very different latency and computational requirements.
5/Live Drawing from @fal_ai_data @krea_ai @SimianLuo @dabit3
X Post: https://x.com/stuffyokodraws/status/1733216372765950260?s=20
Video: