Rise of Autonomous Information Seeking Companions#
We continue observing a rapid movement towards creation of agents that can reason about the objectives they are given, use tools to carry out related tasks using formal languages, and use natural language skills to communicate with their operators. This trend might (and most probably will) lead into thinking companions that can interact with us, seek information and insight on our behalf, and synthesize artifacts towards objectives we have. The ease of interaction with these companions could (and most probably will) get to a point where we can learn / plan / build on our own terms as we commute / run / cook.
SUMMARY
KnowledgeOps is the next stage where automation is created using information seeking agents.
Knowledge jobs involve managing an organization’s knowledge assets and processes to increase efficiency and effectiveness of information intensive tasks.
Information intensive tasks include learning about LLMs and other topics that require consulting different resources and finding information.
An agent like Sherpa could assist with learning by finding and presenting relevant information to the user.
In the future, it may be possible to interact with such agents using voice commands while engaging in physical activities like jogging.
An agent called Sherpa could assist with finding and presenting relevant information to the user.
Knowledge assets include data, code, tools, and documentation, while processes include communication, collaboration, and project management.
Agents like Sherpa could help with collaboration, communication, and project management.
MLOps is a new concept that automates experimentation, data, and model handling.
MLOps is a step forward from DevOps in terms of automating manual processes.
KnowledgeOps is a new concept that automates exploration, planning, and coding of ideas.
Continuous exploration (CE) is a key component of KnowledgeOps, where ideas are explored, built, integrated, delivered, measured, and repeated.
Industry trends suggest a shift towards merging of specialized semantic search, recommendation, and other tools with generative co-pilots.
In order to have an effective agent for information-intensive tasks, three types of skills are necessary: reasoning, content understanding, and execution.
Large language models can provide these skills for an agent.
The AI agent can prioritize tasks, execute them using different tools, keep track of them in memory, and reflect on its observations and achievements.
Similar ideas are being developed by other companies and projects, such as AutoGPT, BabyAGI, and Microsoft Jarvis.
Reflection is a step in the process of using large language models to prompt the system for information, reasoning, and criticism to suggest a resolution.
These agents can call smaller or large language models or other ml models for regression or mathematical calculations.
An engaged community and tools like this can be powerful in allowing people to try out their ideas quickly.
MLOps and DevOps allowed us to move fast and break things, and now it is KnowledgeOps turn to empower trying out ideas cheaply.
KnowledgeOpd can involve creating systems, processes, and collaboration tools that make it easy for anyone to try out ideas.
The process of using large language models can generate implementation-ready code, but it is ultimately up to the operator to ensure its accuracy and feasibility.
The power of this approach is in democratizing entrepreneurship and making it accessible to more people, regardless of their financial or social status.
Ethical questions need to be asked and discussed regarding the control and sharing of data and intellectual property.
For example large language models can be used to generate implementation plans and steps for a project, taking into account available information and time estimates.
Auto-generated documents can be sent to clients or users, providing a starting point for further discussion and exploration of ideas.
TThese agents dynamically construct prompts that control the behavior of the AI agent, allowing for more efficient and effective use of the system.
The system can access both internal and external knowledge resources, synthesizing information from multiple sources to complete tasks.
Prompt engineering may become obsolete as more tools allow for dynamic prompt construction.
Evaluating and reflecting on past actions is important for refining and improving the approach to future tasks.
Access to specialized tools and resources, such as the Notion API, can be integrated into the system to further streamline processes.
The AI agent can be used for a variety of applications, such as teaching assistants or project management tools.
The process of brainstorming with Sherpa and generating an action plan is iterative and can be refined based on new information or feedback.
Collaboration and communication with the AI agent is important to ensure the action plan is tailored to specific needs and constraints.
The AI agent can scrape information from external sources and incorporate it into the action plan, improving its accuracy and effectiveness.
Amir Feizpour (CEO @ Aggregate Intellect)
Amir is the co-founder of Aggregate Intellect (https://ai.science/), a Smart Knowledge Navigator platform for R&D teams in emerging technologies like AI. Prior to this, Amir was an NLP Product Lead at Royal Bank of Canada, and held a postdoctoral position at University of Oxford conducting research on experimental quantum computing. Amir holds a PhD in Physics from University of Toronto.
