#Day12/30 #30DaysofGenerativeAI
🖐🖐 “Will open-source LLMs with decentralized training be competitive with closed-source and centralized LLMs?”
Why is this question from Naval Ravikant an important one to answer ❓ ❓
It determines where the generative AI market will move – towards large LLM providers such as OpenAI or is there a market for smaller, decentralised LLMs customised to specific areas that could be economically viable?
Or is there an hybrid model play?
Naval Ravikant discussed this along with other experts on Airchat.
Here are a key takeaways –
1️⃣ Is it possible to train models in a decentralised manner?
Compute and Data are the most important factors and can this be attempted. And training doesn’t come cheap – Microsoft invested a third round of 10 Bn USD in Jan 2023. Today we are yet to figure out how to train AI in a decentralised way
2️⃣ Why Open Source
– Open source works better than user interface doesn’t matter like in this scenario where it is all natural language input and output
– Direct access to weights provides better customisation – thus everyone will want their own LLMs
– Eventually training LLMs for a specific use case can work out better than general outcomes
Does all of the mean that could end up in a hybrid scenario – an Oligopoly if you like?
Listen to more here
**************************
#reviewswithranjani
#Technology | #Books | #BeingBetter