Should companies architect for model optionality rather than coupling tightly to a single AI model?

Alex Lebrun · Jun 19, 2023

Companies will switch between LLMs very frequently, and treating your model choice as static will leave your product dumber, slower or more expensive than competitors'

Incredible things come out of research every two weeks

13:50 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla

Douwe Kiela · Jun 30, 2023 · hedged

Being language-model agnostic is a competitive advantage right now, though it carries the risk of depending on other people's models

The field is moving so quickly that many new models will come out in the next year

Scope: true at this particular point in time; offset by dependency risk

10:51 20VC: Why Data Size Matters More Than Model Size, Why The Google Employee Was Wrong; OpenAI and Google Have the Advantage & Why Open Source is Not Going to Win with Douwe Kiela, Co-Founder @ Contextual AI

Christian Kleinerman · Sep 22, 2023

Companies should build model optionality rather than coupling tightly to a single model, and those that do will be better positioned

There is so much innovation in the model landscape that building tightly coupled to one model gives up future optionality, even though it's easier to learn one model's ins and outs

24:02 20VC: Are Foundation Models Becoming Commoditised? Do OpenAI and Anthropic of the World Have a Sustaining Moat? Why Smaller Models May Work Better? Why Incumbents with Data Power Win the AI War with Christian Kleinerman, SVP Product @ Snowflake

Harry Stebbings · Nov 15, 2023

The companies that win will be those able to transition between LLMs most seamlessly.

17:16 20VC: How to Survive and Thrive in a World of OpenAI, Are LLMs Being Commoditised, Where Does the Value Lie; Infrastructure or Application Layer, How Apple Could Win in a World of AI, How Amazon Could Threaten OpenAI and Why Google Struggle with Des Trayn

Christian Kleinerman · Sep 22, 2023

Real optionality requires a model abstraction layer, analogous to hardware or cloud abstraction layers, that translates application requests to each model's specific characteristics

Swapping the model is not enough because different models respond differently to the same type of prompt

24:37 True optionality requires a model abstraction layer not just swappability

Des Traynor · Nov 15, 2023

Being able to switch LLMs quickly is an important attribute but not what determines winning.

If someone unlocks new power you want to use it immediately, but winning involves much more than being LLM-agnostic.

17:27 Llm switching ability matters but is not the decisive winning factor