4 ms·

LLMs will become another building block and we will once again stand on the shoulders of giants.

LeifCarrotson 4y ago
With even less understanding of how that foundation was built. It's definitely convenient to be able to throw together a Python script that imports the world and "just works", or to point my secretary at a no-code tool so she can automate her billing processes and email handling without needing to deal with the engineers. But in that handful of really performance-critical, highly demanding problems, it's important to be able to have someone or a few someones who have some comprehension of the full stack, all the from sand to scripting languages. Most of my day is just translating business logic from natural language to precise machine languages, but the best parts are the performance-critical, complex, demanding problems where you need to understand gate delays and bandwidth limits and RTOS priorities and concurrent programming and API minutiae. That's a tall pyramid of knowledge to have to build on, and the area of our T-shaped expertise grows large, but I worry about the future that involves LLMs within APIs. It's one thing to be aware that multithreading locks or and database availability requirements have to contend with non-determinism, so you have to assume the worst and design around that, but when a tall part of the stack is a non-deterministic LLM taking in natural language and doing unpredictable things based on a neural network with billions of nodes, how can you make any guarantees about the outcome?
- potatoman22 4y ago
  Most software that uses statistical algorithms can't make guarantees about certain outcomes. For example, Netflix can't be sure that you'll actually like its recommendations. However, they do have aggregate measures that give a holistic picture on their algorithm's behavior. We can do the same for LLMs: have big-picture summaries that speak to their behavior in aggregate. We can also restrict their action-space by programming deterministic rules or filters on top of their output. While these algorithms will never be as deterministic as a simple circuit, there are measure you can take to restrict and understand their behavior.

With even less understanding of how that foundation was built. It's definitely convenient to be able to throw together a Python script that imports the world and "just works", or to point my secretary at a no-code tool so she can automate her billing processes and email handling without needing to deal with the engineers. But in that handful of really performance-critical, highly demanding problems, it's important to be able to have someone or a few someones who have some comprehension of the full stack, all the from sand to scripting languages. Most of my day is just translating business logic from natural language to precise machine languages, but the best parts are the performance-critical, complex, demanding problems where you need to understand gate delays and bandwidth limits and RTOS priorities and concurrent programming and API minutiae. That's a tall pyramid of knowledge to have to build on, and the area of our T-shaped expertise grows large, but I worry about the future that involves LLMs within APIs. It's one thing to be aware that multithreading locks or and database availability requirements have to contend with non-determinism, so you have to assume the worst and design around that, but when a tall part of the stack is a non-deterministic LLM taking in natural language and doing unpredictable things based on a neural network with billions of nodes, how can you make any guarantees about the outcome?

potatoman22 4y ago
Most software that uses statistical algorithms can't make guarantees about certain outcomes. For example, Netflix can't be sure that you'll actually like its recommendations. However, they do have aggregate measures that give a holistic picture on their algorithm's behavior. We can do the same for LLMs: have big-picture summaries that speak to their behavior in aggregate. We can also restrict their action-space by programming deterministic rules or filters on top of their output. While these algorithms will never be as deterministic as a simple circuit, there are measure you can take to restrict and understand their behavior.