Creating GenAI Advisor

This article is the first one in a series which follows a generative AI application creation flow while navigating a choice of leading platforms and models. As you follow along, you will create a GenAI Advisor who will be able to curate latest insights, trends, guidance, and use cases based on hundreds of research papers, analyst reports, and other data sources. You should be able to upload your business plan, ideation document, current challenges, or project description to the GenAI Advisor and query it for insights, suggestions to improve, critique, and more.

Finding the right problem to solve with GenAI

The popularity and promise of generative AI has resulted in very rapid progress on many fronts including research, open source, and commercial efforts. However, it is becoming harder to separate signal from noise. As a data point, ArXiv reported more than 9,000 Computer Science related paper submissions in October 2023. This number is greater than the sum of submissions in all other categories that month including the popular Math and Physics fields. Humanly it is not possible to follow this volume of research. This is a good problem for leading LLMs as they demonstrate at scale summarization and retrieval augmented generation as strength areas. This is an example of repeat proven capabilities of GenAI at scale in the problem discovery patterns listed in the next section.

Another reason for this to be a good problem for LLMs is that deriving insights from these research papers requires higher order reasoning and domain expertise, which at the moment seems a good case for LLM as assistant supporting human expertise. This is an example of GenAI assistant for human expert pattern.

The pace of change in GenAI field is one more reason this is a great problem to solve. Top sources for GenAI trends and insights include Big Five management consultants like McKinsey, market segment leaders, analysts like Gartner and Forrester, and Venture Capitalist who invest in this field like A16z and Sequoia. These sources usually publish their insights at annual or in some cases quarterly frequency. A GenAI Advisor which can curate and present insights on demand will be disruptive to say the least! This is an example of on demand vs periodic pattern.

The GenAI advisor can extend its data sources beyond arXiv research papers. GitHub open source projects are a rich source of what is trending in a particular field. For medium to large enterprises, there are good chances that GenAI ideas and experiments are sparking in many places within own organizations, and not all of these may be visible to the decision makers. Then, how about going multimodal and extracting trends, insights from videos, audio, and other modalities. All this aligns GenAI Advisor problem with the systemic vs point solution pattern.

Several consulting and analyst firms have published insights from early experiments with GenAI. BCG reported a study, involving 758 own consultants, which yielded more than 40% higher quality output when using AI. PwC rolled out a three-year, $1 billion investment in GenAI, and are starting to see productivity surge in some areas by as much as 40%. This indicates that the GenAI Advisor problem follows the continous experimentation pattern.

Problem discovery patterns

Here is a summary of patterns for finding the right problem to solve with GenAI.

1. Repeat proven capabilities of GenAI at scale. Look for problems which apply proven capabilities like summarization and generation at scale which will be difficult, time consuming, or costly for humans to perform.
2. GenAI assistant for human expert. These are problems which invite human subject matter expertise, higher order reasoning, and intuition as capabilities which are not attainable by the current state of the art GenAI.
3. Systemic vs point solution. Deploying and running GenAI applications is going to cost. Look for solving systemic problems with large enough return on investment instead of point solutions.
4. On demand vs periodic. Problems which will significantly benefit from on demand execution instead of periodic execution.
5. Holistic benefits first. GenAI impacts people (skills and roles), processes (workflows), technology (tools), organization (teams), and motivation (goals). Thinking about the problem holistically will help design better systems.
6. Continuous experimentation. Pace of innovation in GenAI requires experimenting often and at scale. Problems which lend themselves to fail-fast-fail-forward should go first. Don't try to replace your doctor with AI just yet!

The author writes about generative AI to share his personal interest in this rapidly evolving field. The author's opinions are his own and do not represent the views of any employer or other entity with which the author may be associated.

Thanks for reading the AI for Everyone newsletter. Share and subscribe for free to receive new posts and support my work.