Users praise in-context learning and RAG over fine-tuning custom models for enterprise data because inference-time flexibility enables quick updates, easy revocation, and stronger security without baking decisions into weights.
Based on 14 visible X reactions from 33 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@levie Well said. As someone who values disciplined decision-making, I’d say security and trust should always advance together with innovation. The strongest systems adapt without compromising what matters most
@jerryjliu0 the inference-time flexibility really is the killer feature makes the whole stack way more practical for most teams
@levie exactly. owning the model does not solve keeping sensitive knowledge current.
@_AllieHarris @levie 💯
The biggest challenge right now with the topic of every enterprise having their own model is that your most valuable information and insights are not only always changing, but they’re often your most sensitive information. Your most sensitive information can’t be packed into a model usually because it contains data that not everyone gets to have access to, and you can’t keep your security layer inside the model or an agent. I think there are going to be 100X more use cases for custom trained models, especially inside of domain-focused products, but training a model per enterprise is going to be a lot harder than it looks.
LlamaIndex's Jerry Liu agrees, favoring external data orchestration at inference.
@levie Well said. As someone who values disciplined decision-making, I’d say security and trust should always advance together with innovation. The strongest systems adapt without compromising what matters most
@jerryjliu0 the inference-time flexibility really is the killer feature makes the whole stack way more practical for most teams
The biggest challenge right now with the topic of every enterprise having their own model is that your most valuable information and insights are not only always changing, but they’re often your most sensitive information. Your most sensitive information can’t be packed into a model usually because it contains data that not everyone gets to have access to, and you can’t keep your security layer inside the model or an agent. I think there are going to be 100X more use cases for custom trained models, especially inside of domain-focused products, but training a model per enterprise is going to be a lot harder than it looks.
Agreed with this. people underestimate the importance of good abstractions and maintainability. part of the reason LLMs/AI are so popular in the first place is because of the ease of using the models during inference-time (through prompting, orchestration) instead of training https://twitter.com/thejessezhang/status/2076643554810540496
Users praise in-context learning and RAG over fine-tuning custom models for enterprise data because inference-time flexibility enables quick updates, easy revocation, and stronger security without baking decisions into weights.
Based on 14 visible X reactions from 33 accounts; directional sample.
Ask a question below.
Published answers will appear here.
Agreed with this. people underestimate the importance of good abstractions and maintainability. part of the reason LLMs/AI are so popular in the first place is because of the ease of using the models during inference-time (through prompting, orchestration) instead of training https://twitter.com/thejessezhang/status/2076643554810540496