🐙 The octopus offers a useful mental model for understanding AI systems and agents. Each tentacle of the octopus contains its own neural network, functioning as a task-based brain system. When the octopus perceives something like a crab, it signals a tentacle to take action, invoking a specific program.
The evolving interaction between local Small Language Models (SLMs) and cloud-based Large Language Models (LLMs) highlights an interesting dynamic in AI development. When envisioning an enterprise agent today, a mix of specialized fine-tuned SLMs and LLMs, guided by a central orchestrator similar to what we see in the octopus is an approach worth exploring.
The orchestrator determines the most suitable model (tentacle) for the task at hand and the appropriate program to execute, leaving the heavy specialized lifting to run autonomously.
Rather than relying on a single big and costly LLM for all functions, leverage cost-effective SLMs for most tasks and turn to LLMs only when their more powerful capabilities are required.
This approach has a name: Federated Language Models. I expect you'll hear a lot about them in the future.
Be like the octopus.
--
Follow Morten Rand-Hendriksen for practical insights and critical analysis of all things AI.
#ai #llm #slm #federatedlanguagemodels #octopus
VP of Business Intelligence & Data Science | Professor of Practice | Analytical Alchemist: Transforming Data into Business Gold
2moWell of course the cult members believe the dogma of the cult leader!