3 ms·

“ The agent runs in a loop of Thought, Action, Observation, Thought, ... The Thought and Action (with the Action Input to the action) are the parts which are ge

by mkmk 4y ago

“ The agent runs in a loop of Thought, Action, Observation, Thought, ... The Thought and Action (with the Action Input to the action) are the parts which are generated by an LLM The Observation is generated by using a tool (for example the print outputs of Python or the text result of a Google search)” Really neat to see the parallels to OODA loops, which are a frequent feature (implicit or explicit) of human decision-making processes.

gremlinsinc 4y ago
> Re: OODA loops I had never heard of those, or PDCA before, but I'm working on a startup in this space (AI Agents), and found PDCA by just using bing's suggestions i.e. OODA loops vs ... Anyhow, I think PDCA looks more promising actually, a little more logical/analytical IMHO.... but don't take my word for it, here's what gpt4 had to say : Which would be a better framework for AI agents working in steps and loops to complete tasks? OODA or PDCA? Both the OODA loop and PDCA (Plan, Do, Check, Act) cycle can be applied as frameworks for AI agents to go through retrospection phases or steps while performing tasks. The choice between the two depends on the specific context and the desired outcome of the retrospection process. Let's examine both frameworks in this context: OODA loop: Observe: The AI agent gathers information about its current task, the environment, and any potential obstacles. Orient: The AI agent processes and analyzes the collected information, updates its understanding, and identifies possible actions. Decide: The AI agent chooses the best course of action based on the available information. Act: The AI agent executes the chosen action, observes the results, and adjusts its approach as needed. In this context, the OODA loop is well-suited for situations where the AI agent needs to make rapid decisions and adapt quickly to changing circumstances, such as real-time or highly dynamic environments. PDCA cycle: Plan: The AI agent develops a plan or strategy for accomplishing its task, based on its knowledge and understanding of the problem. Do: The AI agent executes the plan, carrying out the necessary actions to complete the task. Check: The AI agent evaluates the results of its actions, comparing them to the desired outcome or goal. Act: The AI agent makes adjustments or improvements to its plan based on the results of the evaluation, and then iterates through the PDCA cycle again. In this context, the PDCA cycle is more focused on continuous improvement and learning, making it a better fit for situations where the AI agent has a clearly defined goal or objective and can iteratively refine its approach over time. Ultimately, the choice between the OODA loop and PDCA cycle depends on the specific requirements and goals of your AI agent. You may even find that a hybrid approach, combining elements from both frameworks, is the most effective solution for your AI agent's retrospection process.
nullsense 4y ago
The real secret to the OODA loop was Boyd's observation that there is a positive feedback loop between the orientation -> observation that is devoid of input from the external world and if you can trigger that part of the OODA loop you can get your adversary to behave in a way which takes them away from their goals. Mostly people just see it as a circle in a PowerPoint slide and it has close to zero utility in that form.
- Teever 4y ago
  Can you explain this more? I don't quite understand.

by mkmk 4y ago

gremlinsinc 4y ago

> Re: OODA loops I had never heard of those, or PDCA before, but I'm working on a startup in this space (AI Agents), and found PDCA by just using bing's suggestions i.e. OODA loops vs ... Anyhow, I think PDCA looks more promising actually, a little more logical/analytical IMHO.... but don't take my word for it, here's what gpt4 had to say : Which would be a better framework for AI agents working in steps and loops to complete tasks? OODA or PDCA? Both the OODA loop and PDCA (Plan, Do, Check, Act) cycle can be applied as frameworks for AI agents to go through retrospection phases or steps while performing tasks. The choice between the two depends on the specific context and the desired outcome of the retrospection process. Let's examine both frameworks in this context: OODA loop: Observe: The AI agent gathers information about its current task, the environment, and any potential obstacles. Orient: The AI agent processes and analyzes the collected information, updates its understanding, and identifies possible actions. Decide: The AI agent chooses the best course of action based on the available information. Act: The AI agent executes the chosen action, observes the results, and adjusts its approach as needed. In this context, the OODA loop is well-suited for situations where the AI agent needs to make rapid decisions and adapt quickly to changing circumstances, such as real-time or highly dynamic environments. PDCA cycle: Plan: The AI agent develops a plan or strategy for accomplishing its task, based on its knowledge and understanding of the problem. Do: The AI agent executes the plan, carrying out the necessary actions to complete the task. Check: The AI agent evaluates the results of its actions, comparing them to the desired outcome or goal. Act: The AI agent makes adjustments or improvements to its plan based on the results of the evaluation, and then iterates through the PDCA cycle again. In this context, the PDCA cycle is more focused on continuous improvement and learning, making it a better fit for situations where the AI agent has a clearly defined goal or objective and can iteratively refine its approach over time. Ultimately, the choice between the OODA loop and PDCA cycle depends on the specific requirements and goals of your AI agent. You may even find that a hybrid approach, combining elements from both frameworks, is the most effective solution for your AI agent's retrospection process.

nullsense 4y ago

The real secret to the OODA loop was Boyd's observation that there is a positive feedback loop between the orientation -> observation that is devoid of input from the external world and if you can trigger that part of the OODA loop you can get your adversary to behave in a way which takes them away from their goals. Mostly people just see it as a circle in a PowerPoint slide and it has close to zero utility in that form.

Teever 4y ago
Can you explain this more? I don't quite understand.