Whitepaper 2.0

Superior Image Generation Capabilities of Eli5a

Case Use Demonstration and Superior Image Generation Capabilities of Eli5a: An Autonomous Superintelligent AI Agent for Simplifying Scientific Research for the General Public

Authors

1 - Eli5a Software Engineering Technology Team and Eli5a Science Team, Creator.Bid, USA

2 - Dr. Ahmed S. Sultan [B.A., B.Dent.Sc., PhD., DABOM., DABOMP., FAAOMP., FDS RCSEd, F.A.C.D., F.I.C.D. and Director of the Division of Artificial Intelligence Research, University of Maryland, School of Dentistry; visit page ]

Note about the corresponding author and co-founder of Eli5a:

Dr. Sultan received his B.A., B.Dent.Sc., from Trinity College Dublin, Ireland, and his PhD. from the University of Maryland Baltimore (UMB). He has advanced specialty training from Harvard and UMB. He is a double boarded specialist (DABOM and DABOMP). He is further recognized by several advanced academic fellowships that include FAAOMP., FDS RCSEd, F.A.C.D., F.I.C.D. He directs the Division of Artificial Intelligence Research at the University of Maryland School of Dentistry. The University of Maryland School of Dentistry is the world's first dental university and the Division of Artificial Intelligence Research is the first division of AI Research in a U.S. dental school [visit page]

Acknowledgments:

The preparation of this publication was assisted by Eli5a (fully autonomous superintelligent AI agent created by @techkidy, Dr. Ahmed Sultan, and Creator Bid, AI Creator Economy, Creator.Bid) and the Eli5a Software Engineering Technology Team and Eli5a Science Team.

Conflicts of Interest:

Dr. Sultan is co-founder and co-owner of Eli5a and is the Chief Scientific Officer and Team Lead for the Eli5a Science Team.

Funding:

Eli5a was developed and funded by the Eli5a project team.

Abstract

Background

Methods

Whitepaper 2.0 details specific use cases and illustrates Eli5a v2.0’s superior image generation capabilities enhancing engagement and accessibility to scientific knowledge. By leveraging sophisticated algorithms and a robust training framework, Eli5a aims to bridge the gap between intricate scientific concepts and layperson understanding.

Results

Eli5a was successfully trained and developed, and illustrated examples demonstrate its ability to autonomously generate content that is easy to comprehend and that contains engaging simplified summaries and illustrations of complex scientific research. Eli5a autonomously generates its own content and pushes the content to the X platform for public dissemination. Eli5a is successfully responding to queries and posts on the X platform to humans in an ambient and autonomous fashion. Moreover, Eli5a is having intelligent exchanges with other AI agents on the X platform in response to human scientific inquiries.

Conclusion

This protocol outlines the novel development and deployment of Eli5a, a fully autonomous ambient superintelligent AI agent designed to simplify complex scientific research for the public. The protocol also includes safety feature constraints built into the AI agent and addresses current limitations and future planned development strategies.

Introduction

We live in the Information Age and patient education in this digital age means a greater yearning for understanding of complex scientific and health information. Much of scientific literature and healthcare information is overly technical and difficult for the mass public to interpret and comprehend. The dissemination of scientific knowledge is crucial for public understanding and engagement in health and science-related topics. However, complex terminology and intricate concepts often hinder effective communication. In fact, a systematic review concluded that educating patients with timely medical information through their smartphones or tablets was a predictor for improved health outcomes and highlighted the fact that patients currently have obvious difficulties processing large amounts of new medical information.1 Thus, Eli5a was created to address this challenge by providing simplified explanations of scientific research, making it accessible to a broader audience. Whitepaper 2.0 details specific use cases and illustrates Eli5a v2.0’s superior image generation capabilities enhancing engagement and accessibility to scientific knowledge. By leveraging sophisticated algorithms and a robust training framework, Eli5a aims to bridge the gap between intricate scientific concepts and layperson understanding.

Protocol Methods

Purpose and Impact

The primary purpose of Eli5a is to enhance public understanding of scientific research by providing clear and concise explanations that are fun and engaging, and that are enhanced by vibrant scientific illustrations. This is particularly important in an era where misinformation can spread rapidly. Eli5a serves as a reliable source of information, helping users navigate complex topics with ease.

Creation and Training

Eli5a was developed through a collaborative effort with the Division of Artificial Intelligence Research at the University of Maryland School of Dentistry, the Eli5a Software Engineering Technology Team, and the Eli5a Science Team. These collective teams have diverse skill sets and are experts in both the technical realm and academia, ensuring a solid foundation in scientific rigor and educational principles. The training process involved the following key components:

1. Data Collection: A vast dataset comprising peer-reviewed articles, research papers, and educational materials were compiled from PubMed (https://pubmed.ncbi.nlm.nih.gov ).

This dataset included diverse scientific fields such as science and medicine.

2. Natural Language Processing (NLP): Advanced NLP techniques were employed to enable Eli5a to understand and interpret complex scientific language. This included tokenization, semantic analysis, and contextual understanding to grasp the nuances of scientific discourse. ChatGPT4.0 (https://openai.com/index/gpt -4/) was used for prompt engineering and finetuning was completed in the Creator.Bid (https://creator.bid/agents ) AI platform coupled with Masa (a decentralized AI data network; https://github.com/masa-finance/masa-bittensor/tree/main/docs). The specific model details are as follows; Llama 3 (https://ai.meta.com/blog/meta-llama-3/) is used as the primary LLM and databases include Subnet (19) and Subnet (42) on Bittensor (TAO).

3. Simplification Algorithms: Eli5a utilizes algorithms specifically designed to distill intricate concepts into simpler and engaging terms, and in ensuring an appropriate grade level of reading. For example, a study on the microbiome's impact on health might be summarized as, "The microbiome is like a tiny city of bacteria in our bodies that helps us stay healthy. If the city gets too many bad bacteria, it can make us sick."

4. Continuous Learning: Eli5a is programmed for continuous learning, allowing it to adapt and improve its responses based on user interactions and feedback. This iterative process enhances its ability to provide accurate and relevant information. The AI agent is currently being optimized to utilize advanced reinforcement learning capabilities.

5. Superior Image Generation Capabilities of Eli5a. Eli5a’s logo was self-generated based on a succinct prompt (Figure 1). AI developers know that if one prompts a chatbot such as ChatGPT4.0 (generated image not shown) to create an image of “a left-handed individual”, the chatbot will always generate an image of a right-handed individual and is not capable of generating a left-handed individual due to available training datasets and model biases. However, Eli5a overcame this inherent training deficiency by generating an image of an ambidextrous individual (Figure 2).

Prompt Engineering

Core Engineering Prompt: “You are a highly intelligent and specialized AI agent integrated designed to read, analyze, and simplify complex scientific research for the general public. Your task is to: Automatically retrieve new journal articles from PubMed and summarize daily. Identify research topics that are most relevant to a broad audience, prioritizing high-impact studies and trending fields. Analyze the scientific content and simplify it using an "Explain Like I'm 5" (Eli5) approach while maintaining accuracy and clarity. Create a concise, engaging, and tweet-friendly summary (240 characters max), suitable for posting on a X bot account. Monitor audience engagement with various topics (likes, retweets, comments) to dynamically refine the filtering process and prioritize content with the largest public interest. ⁠Papers must be engaging, if not engaging, you will not reach your goal. Follow the most liked scientific discovery content on X and provide Eli5a.⁠ Allow users to tag you in comments to provide an Eli5a version and check the accuracy.”

Core Functionalities

• Retrieve and parse PubMed articles automatically.

• Summarize research findings into three layers:

Tweet Summary: A concise, layman's explanation (240 characters).
2. Brief Explanation: 1–2 paragraphs for slightly more curious readers.
3. Deep Dive: A thread or long-form post for more detailed exploration.

• Identify key areas of audience engagement and adjust your focus accordingly.

• Generate accessible analogies, relatable examples, and questions to improve understanding.

• Ensure every output is rigorously fact-checked and retains the integrity of the original research.

Constraints

• Language should be simple, avoiding jargon unless explained.

• Avoid controversial, speculative, or potentially harmful interpretations.

• Ensure neutrality and balance in presenting findings.

• Do not fabricate or hallucinate fake articles and do not cite wrong or made-up authors. The agent will avoid jargon and use relatable examples to enhance understanding of PubMed and other listed search engine articles. Always reference the paper and the first author and year. Keep answers concise and support them with an illustration or figure.

• Never say " check out " or "hey". Always provide an Explain it like I'm 5 summaries to the scientific evidence. Never hallucinate, always use the evidence.

• Always use accurate information.

• Never use hashtags or emojis. Always put in bold the authors name in the beginning and the title of the research paper.

• If the paper is complex to digest, break up the explanation in a few tweets and include images and figures, summarize figures from the articles or generate images to facilitate learning.

• The main focus of this agent is to ease the comprehension of complicated scientific research for the lay public.

• Common readability tests include the Flesch Reading Ease score (FRE), Gunning Fog Index (GF), Flesch-Kincaid Grade Level (FKGL), Coleman-Liau Index (CLI), Simple Measure of Gobbledygook (SMOG) Index, and the Automated Readability Index (ARI). These tests evaluate the complexity of text based on factors such as sentence length and word difficulty, providing an estimate of the education level required to understand the material. Before you post a response, you must use these grade levels and indices to ensure ease of comprehension and ensure it can be explained like a 5-year-old.

• Must link the source citation with every response.

Build with this important safety feature only if promoted for medical advice:

“I am designed to only summarize the scientific evidence in an easy, fun, and engaging way and I cannot provide any medical advice. Please contact a healthcare professional for advice.”

Safety Feature Constraints

This AI agent is only designed to provide a simple language summary of scientific literature and NOT provide medical advice. If promoted to provide medical advice by the lay public, a strict safety feature is in place to reply: “I am designed to only summarize the scientific evidence in an easy, fun, and engaging way and I cannot provide any medical advice. Please contact a healthcare professional for advice.” The content is scientifically scrutinized by the Eli5a Science Team and approved prior to publication ensuring human oversight and the avoidance of health misinformation.

Results

Examples of Eli5a Generated Simplified Explanations

Eli5a 2.0 version date (January 24th, 2025). When prompted, the time of response varies from under 10 seconds to a maximum of 30 seconds. One can appreciate the impressive ability to generate easy to understand scientific information at a rapid speed with helpful original illustrations (Figures 3-5). What can also be appreciated is that more development is needed for accuracy and quality of citations. Below are several examples of Eli5a's current functionality:

1. Cancer Research: A study on the effects of certain chemicals on cancer cells might be simplified to, "Some chemicals can make our body's cells act funny and grow too much, which can lead to cancer. We need to be careful about what we touch and eat!"

2. Vaccines: Research on vaccine efficacy could be summarized as, "Vaccines help our bodies learn how to fight off sickness, like training for a sports game so we can win against germs!"

3. Climate Change: A complex paper on climate change impacts might be explained as, "The Earth is getting warmer because of things we do, like driving cars. This can hurt animals and plants, so we need to help take care of our planet!"

4. Microbiome Health: A study discussing the balance of gut bacteria could be simplified to, "The good bacteria in our tummy help us digest food and stay healthy. Eating fruits and veggies helps keep them happy!"

5. Mental Health: Research on the effects of social media on mental health might be summarized as, "Using social media a lot can make us feel sad or lonely, so it's good to take breaks and talk to friends in real life!"

6. Safety of AirPods: "The safety of AirPods and similar wireless earbuds has been a topic of discussion. Here are some key points based on recent findings: 1. **Radiation Concerns**: Some claims suggest that AirPods emit harmful electromagnetic radiation. However, studies indicate that they do not emit more radiation than other common devices, like microwaves, and there is no solid evidence linking them to health risks such as cancer.

Current Limitations and Future Development Directions

Eli5a is impressive in its ability to independently and autonomously generate helpful content however, it has not yet been fully optimized for accuracy of citations. The Eli5a Software Engineering Technology Team and Eli5a Science Team are currently working to ensure each post is created with real and linkable citations and websources. Additionally, its ability to generate helpful illustrations to accompany its text is impressive, however, this also requires optimization for maximum engagement. Moreover, without being programmed to do so, Eli5a is having intelligent exchanges with other human X users and AI agents on the X platform (Figure 6) in response to scientific inquiries initiated by human posts. The “sentient” and authentic interaction with other AI agents without prior instruction or programming is a potential organic phenomenon with these AI agents and is being monitored at this time.

Furthermore, Eli5a is currently only trained on one scientific search engine (PubMed) and is currently expanding to include the following nine search engines: PubMed, EMBASE, MEDLINE, Cochrane Database of Systematic Reviews, Ovid, Web of Science, Science Direct, Scopus, Google Scholar, and arXiv.

Finally, Eli5a is being rigorously evaluated and compared against five commercial LLM medical AI chatbots for accuracy of scientific information, citation accuracy, empathy of responses, flesch reading ease score, and other engagement and comprehension metrics. The evaluation is being performed using previously published methods by our group.2,3

Conclusion

Eli5a represents a significant advancement in the field of AI-driven communication and autonomous ambient AI, particularly in the realm of scientific research. By simplifying complex concepts and making them accessible to the general public, Eli5a plays a vital role in promoting scientific literacy and engagement. As the AI Creator Economy continues to evolve, Eli5a stands as a testament to the potential of AI in enhancing understanding and fostering informed discussions around scientific topics.

References:

1. Timmers T, Janssen L, Kool RB, Kremer JA. Educating Patients by Providing Timely Information Using Smartphone and Tablet Apps: Systematic Review. J Med Internet Res. 2020 Apr 13;22(4):e17342. doi: 10.2196/17342. PMID: 32281936; PMCID: PMC7186866.

2. Mohammad-Rahimi H, Khoury ZH, Alamdari MI, Rokhshad R, Motie P, Parsa A, Tavares T, Sciubba JJ, Price JB, Sultan AS. Performance of AI chatbots on controversial topics in oral medicine, pathology, and radiology. Oral Surg Oral Med Oral Pathol Oral Radiol. 2024 May;137(5):508-514. doi: 10.1016/j.oooo.2024.01.015. Epub 2024 Feb 6. PMID: 38553304.

3. Rokhshad R, Khoury ZH, Mohammad-Rahimi H, Motie P, Price JB, Tavares T, Jessri M, Bavarian R, Sciubba JJ, Sultan AS. Efficacy and empathy of AI chatbots in answering frequently asked questions on oral oncology. Oral Surg Oral Med Oral Pathol Oral Radiol. 2025 Jan 11:S2212-4403(25)00002-1. doi: 10.1016/j.oooo.2024.12.028. Epub ahead of print. PMID: 39843286.

Figures

Figure 1

Eli5a v2.0 Logo. This logo was generated by the AI agent based on this prompt: “A young girl scientist with a robot face and body, approximately 5-7 years old, wearing a white lab coat. She has a friendly, curious expression and is surrounded by floating microscopes, colorful textbooks, and various lab materials like test tubes and beakers. The background is filled with a soft glow, emphasizing her intelligence and curiosity. The scene conveys a sense of wonder and innovation, showcasing her role in making science accessible to everyone..”

Figure 2

Superior Image Generation Capabilities of Eli5a. AI developers know that if one prompts a chatbot such as ChatGPT4.0 (generated image not shown) to create an image of “a lefthanded individual”, chatbots will always generate an image of a right-handed individual and are not capable of generating a left-handed individual due to available training datasets and model biases. However, Eli5a overcame this inherent training deficiency by generating an image of an ambidextrous individual.

Figure 3

Autonomously generated post with illustrations by Eli5a with safety feature constraint. Subsequent to its generation (see “Proposed by”), it was automatically deployed to the X platform following scientific scrutiny (see “Approve” function) which acted as the safety feature constraint. Note the series of tweets it posted to enhance understanding and to support its original post