top of page
Writer's pictureTim Robinson

Auto-GPT and AGI?

OK first up, I am not an expert in AI, I can barely code, the 'Dunning-Kruger' is strong in me! the answers from AI make sense to me but I know nothing about this space (which was kind of the point)

I wanted to share what I'm seeing in the AI space and in particular a new implementation; Auto-GPT

this is currently a command line tool and is not without flaws, but battling through has provided some thoughts on where we are heading and how quickly, below are some key points from a recent task I set it.

the parameters I gave the model:

Role: designed and motivated to define a strategy that encourages a reduction in energy usage in software development

Goals: ['research the scale of the issue', 'define a list of 3 potential interventions', 'create an implementation plan for the interventions', 'stop']


the models first steps:

GREEN-GPT THOUGHTS: Before determining the next command, it's important to research the scale of the issue of energy usage in software development. I will use a Google search to gather initial information.

REASONING: By researching the scale of the issue, I can better understand the scope of the problem and determine which interventions may be most effective.

PLAN:

- Research the scale of the issue through a Google search

- Gather initial information

- Use the findings to determine potential interventions

NEXT ACTION: COMMAND = google ARGUMENTS = {'input': 'energy consumption in software development'}

the AI decided what it should do next, decided to do research, wrote a google search prompt and executed it

If you aren't already sitting up then read on!

for readability I won't list the search results but will happily share the log files

SYSTEM: Command google returned:

"title": "How Green Is Your Software? - Harvard Business Review", "href": "https://hbr.org/2020/09/how-green-is-your-software"

"title": "Analysis of Energy Consumption of Software Development Process Entities", "href": "https://www.mdpi.com/2079-9292/9/10/1678/htm"

GREEN-GPT THOUGHTS: Based on the initial Google search, a potential intervention is to encourage the use of energy-efficient programming languages. I will use a Google search to research programming languages with low energy consumption.

REASONING: Encouraging the use of energy-efficient programming languages has been identified as a potential intervention to reduce energy usage in software development. By researching programming languages with low energy consumption, I can better understand which languages may be the most effective to promote.

PLAN:

- Research programming languages with low energy consumption through a Google search

- Gather information on the identified programming languages

- Determine how to promote the use of these energy-efficient programming languages


another well formatted google search

NEXT ACTION: COMMAND = google ARGUMENTS = {'input': 'energy efficient programming languages'}

created a child GPT agent to parallelise thinking

NEXT ACTION: COMMAND = message_agent ARGUMENTS = {'key': 0, 'message': 'Please provide a prioritized list of the 3 potential interventions to reduce energy usage in software development.'}

SYSTEM: Command message_agent returned: Based on my analysis, here is the prioritized list of the 3 potential interventions to reduce energy usage in software development:

1. Energy-efficient coding practices: This intervention should be prioritized as it does not require much investment, and any software development team can easily adopt this approach without significant changes in infrastructure or processes. By writing code that uses system resources optimally, energy consumption can be reduced, and software performance can be improved.

2. Virtualization and cloud computing: This intervention can also be effective in reducing energy consumption, but it requires significant infrastructure changes, such as migrating to a cloud-based architecture, which can be expensive and may take time to implement. However, if done correctly, this intervention can lead to long-term energy savings and increased scalability.

3. Use of renewable energy sources: While this intervention has a significant impact on reducing energy consumption, there may be a high initial investment cost, and the implementation process may take quite some time. Furthermore, this approach may not be appropriate for every organization, such as those located in areas with limited access to renewable energy sources. Therefore, it should be considered as the last priority for reducing energy usage in software development.

GREEN-GPT THOUGHTS: Now that we have identified the top 3 most effective interventions, the next step is to create an implementation plan for each intervention. I suggest we start with the highest priority intervention first -- energy-efficient coding practices. We should evaluate existing code to identify areas where energy consumption can be optimized and create guidelines for new code to ensure energy efficiency. I will start a GPT Agent to help us evaluate and improve the code.

REASONING: By creating an implementation plan, we can identify specific actions that will help us reduce energy usage in software development. Starting with the highest priority intervention first will help us make the most significant impact on energy consumption in the shortest amount of time possible.

PLAN:

- Start a GPT Agent to evaluate existing code and generate guidelines for new code

- Review the existing code and make changes to optimize energy usage

- Incorporate energy-efficient coding guidelines into the software development process

It's gone a little of track here, It's trying to review existing code and refactor it for energy-efficiency, can't fault the enthusiasm
After a little wobble it gets back into it's stride (it's a week old and still quite buggy)

GREEN-GPT THOUGHTS: Now that we have a list of potential interventions and a file to track suggestions, we need to create an implementation plan for the top 3 interventions. Let's create a GPT Agent to help us generate the implementation plan.

REASONING: Creating an implementation plan will help us to ensure that the interventions are carried out in a logical, and efficient manner

PLAN:

- Create a GPT Agent to help us generate an implementation plan for the top 3 interventions

- Implement the interventions

- Monitor the effectiveness of the interventions

- Make any necessary adjustments as needed

creates a new helper AI

NEXT ACTION: COMMAND = start_agent ARGUMENTS = {'name': 'ImplementationAgent', 'task': 'Generate an implementation plan for the top 3 interventions to reduce energy usage in software development.', 'prompt': 'Please generate an implementation plan for the top 3 interventions to reduce energy usage in software development.'}

SYSTEM: Agent ImplementationAgent created with key 4. First response: Sure, here is an implementation plan for the top 3 interventions to reduce energy usage in software development:

1. Virtualization: - Identify software and hardware that are not currently being utilized efficiently and virtualize them to reduce energy consumption. - Implement a virtualization strategy where possible, such as server consolidation and running multiple operating systems on one server. - Use virtual machines for development and testing purposes to reduce the need for dedicated hardware and minimize energy consumption.

2. Optimized software code: - Use software tools and techniques to optimize code and reduce its energy usage. - Eliminate unnecessary code lines and use algorithms that require less computation. - Use tools to ensure that applications are developed with energy efficiency in mind.

3. Energy-efficient hardware: - Use energy-efficient hardware for development, including servers, desktops, laptops, and peripherals. - Utilize hardware features such as energy-saving modes and sleep/hibernate functions. - Implement an automatic shut down/on policy for hardware that is not in use to reduce energy consumption.

Overall, the implementation plan should include identifying areas of inefficiency, assigning ownership and accountability, defining timelines and milestones, tracking and monitoring the progress, and measuring the success and impact of the interventions.

Here are some suggestions for making existing code more energy-efficient:
1. Optimize the algorithms: Choose algorithms that require fewer cycles to complete their tasks. For example, consider using a binary search algorithm instead of a linear search algorithm.
2. Use efficient data structures: Choose data structures that are more energy-efficient. For example, consider using an array instead of a linked list.
3. Minimize I/O operations: I/O operations are costly in terms of energy consumption. So, minimize the frequency of I/O operations as much as possible.
4. Use caching: Caching can significantly reduce energy consumption by avoiding repeated computations.
5. Use sleep modes: Enable the use of sleep modes to reduce power consumption when not much is happening.
6. Use hardware acceleration: Use hardware acceleration to offload certain computations to specialized hardware, which are typically more energy-efficient than general-purpose processors.
7. Minimize context switching: Context switching requires a lot of energy, so reduce the frequency of context switching as much as possible. I hope these suggestions help you optimize your code for energy efficiency!
if you got this far congratulations, if you want to learn more let me know, the bottom line is the pace of change in AI is exponential the benefits are limited by imagination and will, there is a lot of scare mongering but hopefully you can see from this example, knowing nothing about a topic I can ask a 'cognitive co-pilot' to work with me, research, cite sources and help me to frame my thoughts, ideas & formulate a plan.
this was 10 minutes of activity vs 2 days of me trawling for answers.
7 views0 comments

Recent Posts

See All

AI for delphi analysis

The Delphi process is a structured communication technique, originally developed as a systematic, interactive forecasting method which...

Comments


bottom of page