What Can Government Contractors Do With GPT?

What is a GPT, how does it work, in very simple terms

Lets be clear what a GPT is - it’s generative AI for Text.

A GPT, or Generative Pre-trained Transformer, is a type of artificial intelligence model that uses deep learning to generate natural language text. It is designed to analyze and understand patterns in large datasets of text, allowing it to generate new text that is coherent and relevant to a given prompt. By using a pre-trained model, GPTs can be adapted to a wide variety of tasks, including language translation, chatbot development, and content creation. With their ability to generate high-quality text quickly and efficiently, GPTs have become an important tool for businesses and researchers in a wide range of fields.

There are also generative models images, such as image diffusion models

Image diffusion AI models are a type of generative model that can be used to create realistic images from text descriptions or even from scratch. They work by starting with a noisy image and gradually reducing the noise over time, while also learning the underlying structure of the image. This process is called diffusion, and it allows the model to generate images that are both realistic and creative.

There are also models coming for audio, video, and other media.

They are not what you’re thinking

What Generative Models are there out there and how do they compare?

Image

DALL-E 2 - The OpenAI Offering

Dream Studio - Stability AI

Stable Diffusion

MidJourney

Video

Google Imagen

Runway

Pictory

Audio

Text

ChatGPT

Anthropic Claude2

Google Bard*

Microsoft Edge/Bing Chat*

Pi.ai

LLaMA2

*These models are connected to search engines and can give a mix of generative and search result content.

Notes of Caution

Input and Output Recording

Input recording is the process of storing the text or speech that is given to a GPT model as input. This can be done for a variety of purposes, such as tracking the user's input history, debugging the model, or creating a transcript of the conversation.
Output recording is the process of storing the text or speech that is generated by a GPT model as output. This can be done for a variety of purposes, such as tracking the model's performance, debugging the model, or creating a record of the conversation.

Input and output recording can be used for a variety of purposes, including:

Tracking user input history: Input recording can be used to track the user's input history, which can be helpful for debugging the model or understanding how the user is interacting with the model.
Debugging the model: Input and output recording can be used to debug the model by identifying any errors in the model's output.
Creating a transcript of the conversation: Input and output recording can be used to create a transcript of the conversation between the user and the model. This can be helpful for debugging the model or understanding how the user is interacting with the model.
Evaluating the model's performance: Input and output recording can be used to evaluate the model's performance by comparing the model's output to the ground truth.
Model training dataset creation: Input and output recording can be used to create a dataset of text or speech that can be used to train a GPT model. This can be done by recording the conversations between users and the model, or by recording the text or speech that is given to the model as input and the text or speech that is generated by the model as output.
RLHF: Input and output recording can be used to train a reinforcement learning (RL) model to interact with a GPT model. This can be done by recording the rewards that are given to the RL model for its actions, and using these rewards to train the RL model to interact with the GPT model in a way that maximizes the rewards.

This is a lesson Samsung famously learned the hard way

Input and Output Filtering

GPT input and output filtering are techniques that can be used to control the content that is given to a GPT model as input and the content that is generated by the model as output. This can be done for a variety of reasons, such as to prevent the model from generating harmful or offensive content, to protect the privacy of users, or to comply with regulations.

There are a number of different ways to implement GPT input and output filtering. Some common techniques include:

Keyword filtering: This involves removing text that contains certain keywords or phrases. For example, a model could be filtered to remove text that contains profanity or hate speech.
Content scoring: This involves assigning a score to each piece of text, based on how likely it is to be harmful or offensive. Text that receives a high score can then be filtered out.
Machine learning: This involves using machine learning models to identify harmful or offensive content. These models can be trained on datasets of text that has been labeled as harmful or offensive.

GPT input and output filtering is an important tool for ensuring that GPT models are used safely and responsibly. By filtering out harmful or offensive content, GPT models can be used in a variety of sensitive applications, such as healthcare, education, and finance.

Here are some specific examples of why GPT input and output filtering might be necessary:

To prevent the model from generating harmful or offensive content: For example, a GPT model that is used to generate text for a news article might be filtered to remove text that contains hate speech or misinformation.
To protect the privacy of users: For example, a GPT model that is used to generate personalized recommendations might be filtered to remove text that contains personal information about the user.
To comply with regulations: For example, a GPT model that is used to generate financial advice might be filtered to remove text that contains advice that is not compliant with regulations.

“Hallucinations”

In the context of GPT models, “hallucination” refers to a phenomenon where the model generates text that is incorrect, nonsensical, or not real. This can happen for a variety of reasons, such as:

The model is not given enough information to generate a correct response. For example, if the model is asked to write a story about a fictional character, but it is only given the character's name and age, the model may generate a story that is not consistent with the character's established backstory.
The model is given conflicting information. For example, if the model is asked to write a summary of a news article, but the article contains conflicting information, the model may generate a summary that is not accurate.
The model is deliberately generating incorrect information. For example, if the model is asked to write a poem about love, but it is programmed to generate text that is not romantic, the model may generate a poem that is nonsensical or offensive.

GPT model hallucinations can be a problem, as they can lead to the spread of misinformation and harmful content. However, there are a number of ways to mitigate the risk of hallucinations, such as:

Training the model on a large dataset of text that is accurate and consistent. This will help the model to learn how to generate text that is more likely to be correct.
Filtering the input and output of the model. This can help to prevent the model from being given conflicting information or from being deliberately programmed to generate incorrect information.
Educating users about the limitations of GPT models. This will help users to be more critical of the text that is generated by the model and to be less likely to believe incorrect information.

Policy

Check your local organization/company policy before using generative models for work.

How to Use a GPT, Prompting Basics

Context: This can include anything from a few words to a few paragraphs of text that provides the model with information about the topic or task at hand. For example, the context for a prompt to write a trip report could include the purpose of a trip report, the destination and dates of the trip, and some notes about the trip.
Task or question: This tells the model what you want it to do. It should be clear and concise, and it should be specific enough that the model knows what you are asking for. For example, the task for a prompt to write a trip report can be as simple as "Write a trip report."
Constraints or limitations: These are any restrictions that you want to place on the model's output. For example, you might want the model to write a report that is no more than 500 words long (approximately one page), and you probably want it to use follow a specific outline.
Additional guidance: This is any additional information that you want to provide to the model. This could include examples of the type of output you are looking for, or it could be instructions on how to format the output.
- Examples: This can include examples of the type of output you are looking for. For example, if you want the model to write a trip report using your writing style, you could provide an example of your last report.
- Instructions: This can include instructions on how to format the output. For example, you might want the model to write a document in a specific style, or you might want it to include a specific outline structure.
- Style: This can include the tone, voice, or genre of the output. For example, you might want the model to write a report in a very formal form if it is being submitted to a government customer, or you might want it to write in the style of a business executive if it is for internal consumption.
- Mood: This can include the emotional tone of the output. For example, you might want the model to write the report in a positive tone if it was a worthwhile trip, or in a negative tone if it was a huge waste of time.
- Setting: This can include the time period, location, or environment of the output, but for our purposes it is more likely your company, your particular industry segment, etc. For example, you might want the model to write specifically about your company as an IT Services company rather than system engineering and integration company. This nuance will add authenticity to the overall generation.
- Characters: This can include specific people to include in the report and their roles. For example, you might want the model to write about the other personnel that went on the trip, who they visited, and the roles each participant played in the trip.
- Narrative: This can include the plot, conflict, and resolution of the output. For example, you might want the model to write a trip report that tells the background and context.

Additionally for an effective GPT prompt:

Be as specific as possible. The more specific you are, the better the model will be able to understand what you want.
Use keywords. When possible, use keywords that are related to the topic or task at hand. This will help the model to focus on the right information.
Avoid using jargon or technical terms. The model may not be familiar with these terms, and it could lead to inaccurate or misleading results.
Be patient. It may take some trial and error to find the right prompt that gets the results you want.

Some Advanced Prompting Techniques

For officials or contractors responsible for various facets of government programs—including business development, security compliance, and information security—efficiency and precision are key. The same principles apply when interacting with AI tools. Ensuring clear and detailed instructions when setting prompts can drastically improve the AI's understanding and response accuracy.

For example, if you're extracting key points from a contract for a stakeholder presentation, instead of using a vague prompt like "Summarize the contract," provide detailed instructions:

Example: Provide a bullet-point summary of the following government contract, focusing on the clauses related to compliance and security measures.

Contract Text: insert contract text here

By being explicit about the context and expected outcome, you enhance the chances of obtaining an accurate and relevant response.

II. Role-Specific Prompts

When dealing with specialized tasks such as security compliance or information security, consider leveraging the AI's 'Persona Pattern'. This allows the AI to operate from the standpoint of a specialized role, like a cybersecurity expert, enhancing the relevance of its output.

Example: Assume the persona of a compliance officer and summarize key legal requirements outlined in the following document related to the Federal Information Security Management Act (FISMA).

III. Constraints on Prompt Size

It's important to understand that language models like ChatGPT have token limitations—usually 2048 tokens, including both prompt and response. For tasks requiring detailed responses, maintain concise yet informative prompts to maximize the utility of the response.

Example: Given the token limitations, summarize the key compliance protocols for data encryption in government contracts in under 300 words.

IV. Refining Queries

In the fast-paced environments of program management and business development, time is often of the essence. Use the AI's 'Question Refinement Pattern' to optimize your queries for more focused answers, particularly when dealing with complex contract language or technical jargon.

Example: If I ask a question about ISO 27001 compliance, suggest a more refined question that considers specific clauses and ask if I want to proceed with the refined query.

V. Kit Bashing

In the realm of generative AI prompting, "kit bashing" refers to the technique of combining pre-existing prompts or prompt fragments to create a new, more specialized or complex prompt.

Essentially, you're "bashing" together bits and pieces of previously used prompts to address a specific need. This technique allows for greater flexibility and depth, enabling you to generate more nuanced outputs from the AI model. Suppose you are responsible for security compliance in a government contract, and you also need to manage program elements. You may have two separate sets of prompt fragments: one for assessing cybersecurity measures and another for evaluating program milestones. Kit bashing enables you to create a new prompt that simultaneously checks for cybersecurity compliance and program milestone status.

Example: Assume the roles of both a Program Manager and a Security Compliance Officer. First, evaluate the following program milestones for our government contract and indicate any that are at risk of being delayed. Second, assess the following security protocols to identify potential non-compliance with the Federal Risk and Authorization Management Program (FedRAMP). Provide actionable recommendations for both.

VI. Flipped Interaction

Flipped interaction is a generative AI prompting technique where the AI is asked to ask you questions instead of you asking the AI questions. This can be a great way to test your knowledge, get feedback on your work, or generate new ideas. For example, let's say you are a government contracting company employee who is working on a project to develop a new software application for the government. You could use flipped interaction to test your knowledge of government regulations by asking the AI to ask you questions about those regulations.

Example: Act as a government regulator and ask me questions about the following regulations:

The Federal Acquisition Regulation (FAR)
The Cybersecurity Maturity Model Certification (CMMC)
The General Services Administration (GSA) Acquisition Framework (GAF)

This prompt provides the AI with enough information to generate a series of questions about the specified regulations. You can then answer the questions to test your knowledge of the regulations.

VII. Breaking Down Complexity

When facing multifaceted questions, such as evaluating the risk factors in a complex project, use the 'Cognitive Verifier Pattern' to break the question into smaller, more manageable queries. The AI can then compile the answers to provide a comprehensive response.

Example: If I ask about the security risks in implementing a new software system, break the question down into sub-questions related to software vulnerabilities, data integrity, and regulatory compliance.

VIII. Multi-Persona

Multi-persona prompting is a technique that involves instructing a generative AI model to generate text from the perspective of multiple different personas. This can be a great way to generate creative content, explore different perspectives, or simulate conversations between different people.

For example, let's say you are working on a project to develop a new software application for the government. You could use multi-persona prompting to generate text from the perspective of different stakeholders in the project, such as the government customer, the software developers, and the end users.

Here is a well-crafted example prompt that you could use for multi-persona prompting:

Example: Generate a conversation between a government customer, a software developer, and an end user about the new software application. The conversation should explore the different perspectives of the stakeholders and should highlight the benefits of the new application.

IX. Contextual Learning Through Few-Shot Prompts

Especially beneficial for business development and crafting persuasive proposals, few-shot prompting can guide the AI to produce targeted and contextually apt responses. By providing examples, you set a behavioral template for the AI.

Example: Given these previous project summaries, generate a compelling summary for our new cybersecurity compliance initiative.

X. Progressive Querying with Chain-of-Thought

The chain-of-thought technique is particularly useful for government contractors who are looking to deeply analyze complex issues—be it understanding new federal regulations, dissecting multifaceted contracts, or exploring emerging trends in information security.

In the chain-of-thought approach, you're essentially having a guided conversation with the AI tool, asking subsequent questions based on the AI's prior responses. This enables you to dig deeper into a topic in a structured and progressive manner.

Example Chain for Regulatory Analysis:

First Query: What is the NIST 800-171?
Second Query: How does NIST 800-171 affect data management in government contracts with the Federal Government?
Third Query: What are the penalties for non-compliance with NIST 800-171 these contracts?

This progression allows you to start with a broad question and then narrow your focus, providing you with targeted insights that are highly relevant to your specific concerns in program management, business development, security compliance, and information security.

Example Chain for Information Security Risk Assessment:

First Query: What are the common types of cyber threats facing government agencies?
Second Query: How can these cyber threats compromise a government contract?
Third Query: What preventive measures can be incorporated into a contract to mitigate these risks?

Here, you initiate with a general understanding of the landscape and then guide the conversation toward practical applications and risk-mitigation strategies directly relevant to government contracting.

Example Chain for Technological Impact on Contracting:

First Query: What are the emerging technologies in data encryption?
Second Query: How can these technologies be applied to enhance security in government contracts?
Third Query: What are the potential challenges or risks involved in implementing these technologies?

This sequence helps you understand not only the options available for enhancing contract security but also the challenges and potential risks—information crucial for comprehensive risk management.

XI. More Fun Prompt Ideas

Leveraging AI for Compliance Audits

If you're responsible for ensuring compliance with government regulations, using AI tools can dramatically streamline the audit process. However, the quality of the output is largely dependent on how well you frame your questions and prompts.

Example: Assume the role of a federal compliance auditor. Identify and list potential red flags in the following transaction logs according to the Federal Acquisition Regulation (FAR).

By employing AI in this manner, you can make preliminary assessments more efficiently, freeing you to focus on complex issues that require human judgment.

Data-Driven Business Development

AI can be a valuable asset for business development teams seeking to analyze market trends, competitor strategies, or even the likely success of upcoming bids.

Example: Based on the data provided, generate an analysis of market trends in the defense sector for the past five years. Summarize the key opportunities and threats for government contractors.

Customized Security Protocols

In the realm of information security, you can instruct the AI to offer recommendations tailored to the specific needs of your project or program.

Example: Assume the role of an Information Security Officer. Evaluate the following cybersecurity protocols for a federal cloud storage service and recommend improvements to ensure compliance with the Defense Federal Acquisition Regulation Supplement (DFARS).

Regulatory Update Alerts

While AI models like ChatGPT do not possess real-time knowledge, you can input the latest changes in laws or regulations to get a synthesized understanding of how they might affect your contracts.

Example: Explain the implications of the recently enacted Cybersecurity Maturity Model Certification (CMMC) on existing defense contracts.

Documentation and Report Generation

Creating comprehensive reports is often a laborious task. AI can assist in generating first drafts or outlines, saving considerable time.

Example: Generate an outline for an annual compliance report adhering to the guidelines set by the General Services Administration (GSA).

Enhancing Proposal Quality

A well-crafted proposal can be the difference between winning and losing a government contract. Use the AI's ability to refine and enhance language to improve the quality of your proposals.

Example: Review the following proposal summary and suggest improvements to make it more compelling for a Department of Energy contract.

Utilizing AI for Training Programs

For those involved in program management, AI can help design training modules tailored to specific roles within your team.

Example: Design a training module focused on information security best practices for employees managing sensitive government data.

Assessing Financial Risks

AI can also be used to make preliminary assessments of the financial risks involved in a project or contract.

Example: Analyze the following cost estimates and revenue projections for a government project and highlight any potential financial risks.

What Can Government Do With a GPT?

Lets get down to a concrete and relevant example, lets write a MAC/IDIQ solicitation.

We’ll start with some context, like a related MAC/IDIQ solicitation from which we can pull the structure.

MAC/IDID Content C.3.1. Scope It is the intention of the U.S. Government to award an Indefinite Delivery/Indefinite Quantity(IDIQ) as a result of this solicitation. The services will be ordered on an as needed basis through country specific task orders and will last from a minimum of six months up to 60 months. The time of issuance and amount of work in task orders cannot be accurately predicted, nor is possible to determine the precise types or amount of services, supplies and/or equipment that will be ordered during the term of the contract. CDC will furnish Contractor awardees with the information needed to compete for task orders. The information will include country program descriptions, scopes of work, as well as relevant policies, procedures, and guidelines that become available periodically throughout the contract period. Supplies and equipment are incidental to the required services to be performed and shall be ordered as needed on each individual task order. Performance for the individual task orders will be measured in accordance with the performance-based matrices attached to each task order. The Contractor awardee shall be obligated to perform within the minimum and maximum order limitations set forth in clauses 52.216.19, Order Limitations. The Government will make every effort to give advance notice of requirements, but the services for which this contract will be used could potentially address an immediate need involving an emergency, or a short notice requirement. C.3.2. Requirements The Contractors shall employ and provide qualified individuals for short-term and long-term durations (minimum six months up to sixty months) in various locations throughout the world in support of CDC activities, as outlined in individual task orders. The CDC generated task order requests will describe the support for technical, operational and professional services including, but not limited to designing, implementing, managing and evaluation of CDC global health programs such as HIV/AIDS, Infectious Diseases, Global Disease Detection (GDD), Neglected Tropical Diseases, Emergency Response, Global Health Protection and Global Health Security. Services include providing technical assistance and support, training and education to indigenous populations. Most of the tasks will take place on long-term assignments (up to 5 years) in non- U.S. locations, primarily in developing countries. The Contractors must be capable of providing management, support and procurement services for their employees including, but not limited to travel assistance, airline tickets, immunizations, passports, long-term visas and work permits, background checks, housing, medical and evacuation insurance, foreign government taxation, filing income tax returns with country of origin, compliance with foreign country tax regulations and international and local laws, emergency medical care and other details necessary to successfully place and support personnel to perform the services required. Please note: Contractor task order employees will not be eligible for U.S. government and Embassy assistance typically provided for Personal Service Contractors (PSC) hired to work for U.S. Foreign Service agencies. It is the Contractor’s responsibility to acquire appropriate visa and work permits for their employees. In certain exceptional emergency situations, for example intensive disease outbreak or global threats, special arrangements for Contractor service employees may be pre-approved by the U.S. government. The following are representative examples of services and types of tasks which may be ordered and conducted through the issuance of individual task orders under this contract. This listing is not represented as being complete or all-inclusive. The task order requirements section shall function as a Performance Work Statement (PWS). C.3.2.1. Technical Support Services This category includes services that require technical expertise to provide technical advice and assistance to CDC staff and other public health, medical and scientific professionals in both laboratory and office activities. One characteristic of the services in this category is the requirement for knowledge of scientific research techniques and analyzing data. This category may also include services that require technical expertise in providing solutions with regards to security compliance, training, operational support, technical and development support, international or global experience and technical knowledge in a myriad of international activities. Characteristic of the services in this category are the requirements for knowledge of information technology principles and techniques and public health program management experience that is less than full professional knowledge, but which nevertheless enables the technician to understand how and why a specific device, skill, or system operates. Examples of these technical services include, but are not limited to public health, laboratory coordination and management, training, demographics, counseling, technical writing, building and design coordination, communications, Informatics, human capacity development, emergency operations, research and science. Contractors providing these services must possess the ability to communicate effectively both orally and in writing and the ability to effectively relate to diverse groups of people. C.3.2.2. Operational Support Services (Set aside for small businesses) Includes services that provide operational assistance and support to technical, professional, medical, and scientific personnel. This category may include tasks that involve preparing, transcribing, transferring, systematizing, and preserving written communications and records; gathering and distributing information; operating office machines; storing, distributing, and accounting for storage of material; operating office equipment; originating and distributing correspondence in both written and electronic formats, and performing other administrative duties. The computer skill requirements shall include but not limited to Microsoft Word, Excel, PowerPoint, and Microsoft Outlook. Examples of these tasks include, but are not limited to secretarial services, program coordination, financial consulting, facilities projects, project analysis, interviewing, extramural resource management, public affairs, real estate maintenance, study coordination, program operations, administrative support services, office automation services and other related fields. The contractor staff assigned to these tasks must possess the ability to communicate well both orally and in writing and the ability to relate to diverse groups of people. C.3.2.3. Professional Support Services Services that require professional expertise having a recognized status based upon acquiring professional knowledge through the study of sciences such as biology, chemistry, physics, physiology, psychology, medicine, public health, epidemiology, and behavioral sciences. These tasks require medical, public health, epidemiology, or behavioral sciences expertise. Examples of these professional services include, but are not limited to, microbiology services, pharmaceutical services, nursing, medical and surgical services, epidemiology and research, science (behavioral and health), occupational therapy, health economy, entomology, virology, laboratory quality assurance, behavior modification services and other related services. C.3.2.4. Historical Examples of Services for ITOPSS Service Contractors Future task orders may be requested for a variety of tasks and services, some short examples include but are not limited to: Infection/Disease Support Services (HIV/TB, Avian and Human Influenza, Malaria, Polio, Dengue, Japanese Encephalitis, STD, Typhoid, Hepatitis, Nosocomial Infections, EBOLA, and other emerging infectious diseases): Provide primary disease prevention services that assist host country partners to develop and implement or improve effective and sustainable voluntary counseling and testing services, community outreach for populations at high risk for infection, demonstration projects aimed at Note: This domain and all task orders under 500K will exclusively be set aside for the small business awardee(s). prevention, policies and promotion activities for “universal precautions” to avoid exposures for health care and other workers. Promote practical and effective disease care and treatment programs through working with the host government to develop and implement treatment guidelines, policies, sustainable models and effective national disease management systems linking infected persons to care, training on disease management and care for health care providers and related training and systems development for laboratory workers. Provide technical assistance in local and national surveillance, development of provincial program protocols, identification of public health priorities based on data, strategic planning and training in prevention, treatment, program management, monitoring, and evaluation. Coordinate and work primarily with the CDC representative Office, host country government, (national and provincial) and other bilateral and multilateral donors to achieve the CDC goals. Management Support Services: Assist with maintaining staffing continuity in-country as current staff transition out, and new staff are transferred in-country. Coordinate and facilitate support to CDC office in the preparation for high level visitors, COP reviews, time limited administrative projects. Provide guidance and expertise in the administration of the extramural program and a variety of functions (Standard Operating Procedures, administrative and programmatic training, etc.) involving major grants, cooperative agreements, contracts, interagency agreements, and memoranda of agreement. Provide advice and support to the USG/CDC country team related to policies and guidelines concerning various issues in-country, (e.g., expanding programs, geographic coverage of in- country activities, related costs, resources needed, etc.). Assist CDC in-country technical staff and the Embassy HR Office to develop staffing plans, position descriptions, and documentation for future proposals. Epidemiology Support Services: Contribute to the development of projects for infectious disease surveillance and research. Conduct data analysis, participate in the development findings, reports of record, peer-reviewed publications, and presentations. Provide technical assistance, assist the CDC and host government partners to prepare and present data on clinical and epidemiological findings to in- country Ministry of Health leaders and other governmental offices. Describe the epidemiologic and clinical characteristics of patients with various laboratory-confirmed infections identified through an ongoing studies local province. Describe the economic and disease burden and specific causes of severe infectious disease illnesses within country. Work closely with CDC disease specialists, the International Emerging Infections Program (IEIP), host government institutions, staff and other donors to ensure that health related goals and objectives are clearly identified and addressed. Advise and work closely with CDC and IEIP staff on strategic planning, activities to strengthen infectious disease surveillance, research, and training. Review scientific literature on infectious disease pathogens identified as possible causes of illnesses as part of surveillance and research activities. Provide technical assistance in routine analysis and evaluation of data derived from surveillance systems and research studies. Provide guidance to epi/surveillance projects to research and data staff in accordance with WHO recommendations and priorities. Review data to ensure accurate completion of study instruments, timely specimen collection, and proper maintenance of patient records, databases, documentation and evaluation. Identify additional clinical data needed for analysis and supervise collection of this data in the field as necessary. Provide technical assistance, assist in organization and implementation of training sessions on rapid outbreak response and risk communications, surveillance and research. Laboratory Support Services: CDC has various cooperative agreements (CoAgs) in place with the countries referenced in F.2. This position will provide support services to those CoAgs. Each task order will reference the specific CoAg number. Provide overall expert support to the host-country government to build, enhance and improve the quality of labs and laboratory diagnostics and services at the national, provincial and local levels. Provide TA, collaborate with international partners’ in-country, to assist the host country modify and develop policies, SOPs for quality assurance and biosafety that will provide high quality test results. Assist the host government Quality Manager (QM) in the development of a quality system plan of the requirements for the management system and technical competence needed to meet International Standards for lab accreditation. Data Management, Informatics, Data Research, and Study/Program Coordinator, Writer- Editor, and Communication Support Services: Assist in developing and implementing various modes of communication for diverse partners including the MOH, Non-governmental (NGO) partners and relevant Atlanta-based advisors. Review protocols and tools under guidance from the CDC office in-country to ensure adequate quality, scientific soundness and that appropriate policies are followed. Track the submission and approval process with US and in-country-based science approval boards. Assist with training of in-country data collectors, data entry and management staff. Travel to local study sites and participate in site preparation, data collection, and data analysis. Assist with interpretation of study findings and writing of study and public health evaluation reports. Support development of presentations and publications. Document actions relevant to procedures for USG and the local Government to ensure that future staff or contractors are able to provide follow on work.
Lets give it an organization background to work with The NIH Common Fund’s Somatic Cell Genome Editing (SCGE) program aims to reduce the burden of diseases caused by genetic changes. Genome editing technologies present an exciting prospect for treatments and possibly even cure for these diseases. During its first 5-year phase (FY18-FY23), SCGE developed quality tools to perform and assess effective genome editing tools in non-reproductive (“somatic”) cells of the body. In its second phase (FY23-27), SCGE will accelerate the translation of genome editing therapies into the clinic.

SCGE Phase 1

SGCE worked to improve the efficacy and specificity of genome editing approaches. The SCGE program developed new methods and improved systems that delivered genome editing machinery into various tissues with greater specificity, including tissues that present an unmet need and that are harder to reach such as the brain, ear, heart, and lung. SCGE made significant discoveries of new or optimized editors that edit target genomes with improved efficacy and novel functionality, including a prime editor that could correct up to 89% of known genetic variants associated with human diseases. The SCGE program developed new methods to assess unintended biological effects, such as high-throughput technologies that quickly identify high specific target sites and follow genome-wide activity of editors over time. Through the development of better animal models, the SCGE program has tested and validated a number of genome editing tools and delivery systems. The program has developed a genome editing toolkit(link is external) that broadly disseminates program tools and learnings, opening this space to a wider range of the biomedical research community.
Let’s give it a topic on which to write the IDIQ solicitation

Subject of the Challenge:

Recent advancements in the genome editing technology field have enabled scientists to manipulate genomic sequences rapidly and efficiently. Despite revolutionary progress in this area, several challenges remain. Existing gene editing technologies like CRISPR-cas9, base editors and prime editors have great potential, but existing delivery technologies are not able to deliver gene editing technologies to many target tissues and cell types in sufficient quantities, which hinders clinical applications. While some cell types, like hepatocytes in the liver, have many delivery technologies capable of delivering genome editors, there are many other organs and cell types that are harder to reach. The Targeted Genome Editor Delivery (TARGETED) Challenge is a $6,000,000 challenge to improve the current state of in vivo delivery technologies for genome editors in two Target Areas: 1. Programmable Delivery System for Gene Editing, and 2. Crossing the Blood-Brain Barrier. The National Institutes of Health (NIH), through the Common Fund’s Somatic Cell Genome Editing (SCGE) program, is seeking Participants with ideas or early-stage solutions to join the Challenge with the chance to win up to $1,000,000 and have their solution independently tested and validated in large animal models through NIH-supported independent evaluation relevant to preclinical assessments of investigational products.

The Challenge is a three-phase competition. In Phase 1, Participants will be asked to submit a proposal describing their proposed solution and how it will address the requirements for one of the Target Areas. Participants may submit proposed solutions to both Target Areas but must do so with separate proposals that independently address each Target Area’s requirements. Up to ten proposals that are judged to best meet the requirements will each be awarded up to $75,000. Additional prizes of $50,000 may be awarded to additional meritorious solutions on the basis of the Judging Criteria. In Phase 2, Participants must submit data from studies that demonstrate delivery and editing performance as well as describe their methodology, technology, and how their solution addresses the Challenge criteria. Participation in Phase 1 is not a requirement for participation in Phase 2; however, it is strongly encouraged. Up to 10 winners of Phase 2 will be each awarded $250,000 and will be eligible to compete in Phase 3. Only Phase 2 winners will be eligible to participate in Phase 3. Phase 3 is separated into Phase 3a and 3b; all Participants must submit solutions for Phase 3a to be eligible to participate in Phase 3b. For Phase 3a, Participants must submit all required information showing that their technology is ready for large animal testing through NIH-supported independent evaluation and has the ability to solve the requirements for one of the Target Areas. Up to 6 Participants will each be awarded $50,000 and will then prepare for reagent scale up and protocol development for NIH-supported large animal testing. Participants who submit their reagents and protocols by the deadline for Phase 3b will have access to NIH-funded independent large animal testing to validate their solution. NIH will review the results and only award prizes to Participants whose solutions meet or exceed the criteria. The top successful solution in each Target Area will be awarded $625,000; the second place solution in each Target Area will be awarded $225,000; the third place solution in each Target Area will be publicly recognized and given an honorable mention award. Participants who participate successfully in all three phases could be awarded up to $1,000,000 in each Target Area.

Final (Phase 3) Solution Requirements

Target Area 1: Programmable Delivery System for Gene Editing

Solutions must be a highly efficient and programmable delivery system to deliver genome editing machinery that can target specific tissues (cells, types, and/or organs). Solutions must be able to be programmed to deliver to at least three distinct and different cell(s), tissue types, and/or organs and with delivery and editing capability that is at least as efficient as the current state of the art. An optimal solution would be straightforward to manufacture, low-cost, scalable and have a reasonable safety profile. Solutions that propose viruses and viral-like systems or particles must build on the field and meet the criteria demonstrating full understanding of how the delivery system can be modified so that it is programmable and can target a variety of different tissue targets (cells, types, and/or organs). The solution will be judged on how well it meets the criteria.

Programmable solutions that only target central nervous system (CNS) targets should be submitted under Target Area 2. Solutions that meet the requirements for Target Area 2 but also are programmable to target a non-brain organ may be submitted for consideration in both Target Areas, though solutions submitted to both Target Areas are only eligible for one prize. To be highly competitive in this Challenge, solutions must:

Be programmable and target at least three distinct and different cell(s), tissue types, and/or organs.
Be able to deliver an editor and demonstrate delivery and editing that be at least as efficient as the current published efficiencies for the different tissue targets (cells, types, and/or organs) proposed.
Have a known biological mechanism for programmability: a clear relationship between what is done to modify the technology to deliver to different tissue targets (cells, types, and/or organs) and how this relates to the underlying biology and/or biochemistry of the system.
Have conducted studies that demonstrate biodistribution and route of administration/delivery method.
Demonstrate successful delivery and editing performance in large animals through NIH-supported independent evaluation.
Have demonstrated a safety profile in experimental models consistent with other gene therapy/gene editing delivery systems intended for use in humans.

Solutions should also have these desired traits:

Target more than one tissue/organ type.
Be innovative in approach.

Additionally, solutions could:

Demonstrate market potential, competitive advantage, or potential to meet unmet medical needs.
Be able to be manufactured in a scalable and cost-effective manner.

Target Area 2: Crossing the Blood-Brain Barrier (BBB)

Solutions to Target Area 2 must be highly efficient, non-viral delivery systems capable of crossing the BBB to deliver genome editing machinery to a substantial proportion of clinically relevant cell types in the brain.

To be highly competitive in this Challenge, solutions must:

Be programmable and target at least three distinct and different cell(s), tissue types, and/or organs.
Be able to deliver an editor and demonstrate delivery and editing that be at least as efficient as the current published efficiencies for the different tissue targets (cells, types, and/or organs) proposed.
Have a known biological mechanism for programmability: a clear relationship between what is done to modify the technology to deliver to different tissue targets (cells, types, and/or organs) and how this relates to the underlying biology and/or biochemistry of the system.
Have conducted studies that demonstrate biodistribution and route of administration/delivery method.
Demonstrate successful delivery and editing performance in large animals through NIH-supported independent evaluation.
Have demonstrated a safety profile in experimental models consistent with other gene therapy/gene editing delivery systems intended for use in humans.

Solutions should also have these desired traits:

Target more than one tissue/organ type.
Be innovative in approach.

Additionally, solutions could:

Demonstrate market potential, competitive advantage, or potential to meet unmet medical needs.
Be able to be manufactured in a scalable and cost-effective manner.

Target Area 2: Crossing the Blood-Brain Barrier (BBB)

To be highly competitive in this Challenge, solutions must:

Be able to traverse the BBB in vivo.
Be able to deliver an editor and demonstrate delivery and editing in a substantial proportion of clinically relevant cell types in the brain.
Be a non-viral delivery technology. Solutions may be virus-like particles and/or incorporate components of viruses in the proposed delivery technology. Solutions that are modifications of recombinant adeno-associated viral vectors do not meet this criterion.
Demonstrate successful delivery and editing performance in large animals through NIH-supported independent evaluation.
Have demonstrated a safety profile in experimental models consistent with other gene therapy/gene editing delivery systems intended for use in humans.

Solutions should also have these desired traits:

Be innovative in approach.
Be able to be manufactured, ideally synthetically, in a scalable and cost-effective manner.

https://chat.openai.com/share/5a424b7d-700c-45cd-b5d1-02d97fe40b10

What Can Industry Do with GPT?

Let’s turn around and write a SBIR proposal

We need a proposal outline, thankfully the government gives us that part Content of the Technical Volume (Volume 2)
The Technical Volume should cover the following items in the order given below:
(1) Identification and Significance of the Problem or Opportunity. Define the specific technical problem or opportunity addressed and its importance.
(2) Phase I Technical Objectives. Enumerate the specific objectives of the Phase I work, including the questions the research and development effort will try to answer to determine the feasibility of the proposed approach.
(3) Phase I Statement of Work (including Subcontractors’ Efforts)

a. Provide an explicit, detailed description of the Phase I approach. If a Phase I option is required or allowed by the Component, describe appropriate research activities which would commence at the end of Phase I base period should the Component elect to exercise the option. The Statement of Work should indicate what tasks are planned, how and where the work will be conducted, a schedule of major events, and the final product(s) to be delivered. The Phase I effort should attempt to determine the technical feasibility of the proposed concept. The methods planned to achieve each objective or task should be discussed explicitly and in detail. This section should be a substantial portion of the Technical Volume section.

b. This BAA may contain topics that have been identified by the Program Manager as research or activities involving Human/Animal Subjects and/or Recombinant DNA. If Phase I performance includes performance of these kinds of research or activities, please identify the applicable protocols and how those protocols will be followed during Phase I. Please note that funds cannot be released or used on any portion of the project involving human/animal subjects or recombinant DNA research or activities until all the proper approvals have been obtained (see Sections 4.9 - 4.11). Small Business Concerns proposing research involving human and/or animal use are encouraged to separate these tasks in the technical proposal and cost proposal in order to avoid potential delay of contract award.

(4) Related Work. Describe significant activities directly related to the proposed effort, including any conducted by the principal investigator, the proposing small business concern, consultants, or others. Describe how these activities interface with the proposed project and discuss any planned coordination with outside sources. The technical volume must persuade reviewers of the proposing small business concern's awareness of the state of-the-art in the specific topic. Describe previous work not directly related to the proposed effort but similar. Provide the following:

a. Short description,

b. Client for which work was performed (including individual to be contacted and phone number), and

c. Date of completion.

(5) Relationship with Future Research or Research and Development

a. State the anticipated results of the proposed approach if the project is successful.

b. Discuss the significance of the Phase I effort in providing a foundation for Phase II research or research and development effort.

c. Identify the applicable clearances, certifications and approvals required to conduct Phase II testing and outline the plan for ensuring timely completion of said authorizations in support of Phase II research or research and development effort.

(6) Commercialization Strategy. Describe in approximately one page your proposing small business concern's strategy for commercializing this technology in DoD, other Federal Agencies, and/or private sector markets. Provide specific information on the market need the technology will address and the size of the market. Also include a schedule showing the quantitative commercialization results from this SBIR project that your proposing small business concern expects to achieve.
(7) Key Personnel. Identify key personnel who will be involved in the Phase I effort including information on directly related education and experience. A concise technical resume of the principal investigator, including a list of relevant publications (if any), must be included (Please do not include Privacy Act Information). All resumes will count toward the page limitations for Volume 2.
(8) Foreign Citizens. Identify any foreign citizens or individuals holding dual citizenship expected to be involved on this project as a direct employee, subcontractor, or consultant. For these individuals, please specify their country of origin, the type of visa or work permit under which they are performing and an explanation of their anticipated level of involvement on this project. Proposing small business concerns frequently assume that individuals with dual citizenship or a work permit will be permitted to work on an SBIR project and do not report them. The proposal may be deemed nonresponsive if the requested information is not provided. The proposing small business concerns should report all individuals expected to be involved on this project that are considered a foreign national as defined in Section 3 of the BAA. You may be asked to provide additional information during negotiations in order to verify the foreign citizen’s eligibility to participate on a SBIR contract. Supplemental information provided in response to this paragraph will be protected in accordance with the Privacy Act (5 U.S.C. 552a), if applicable, and the Freedom of Information Act (5 U.S.C. 552(b)(6)). '
(9) Facilities/Equipment. Describe available instrumentation and physical facilities necessary to carry out the Phase I effort. Justify equipment purchases in this section and include detailed pricing information in the Cost Volume. State whether the facilities where the proposed work will be performed meet environmental laws and regulations of federal, state (name), and local Governments for, but not limited to, the following groupings: airborne emissions, waterborne effluents, external radiation levels, outdoor noise, solid and bulk waste disposal practices, and handling and storage of toxic and hazardous materials.
(10) Subcontractors/Consultants. Involvement of a university or other subcontractors or consultants in the project may be appropriate. If such involvement is intended, it should be identified and described to the same level of detail as the prime contractor costs. A minimum of two- thirds of the research and/or analytical work in Phase I, as measured by direct and indirect costs, must be conducted by the proposing small business concern, unless otherwise approved in writing by the Contracting Officer. SBIR efforts may include subcontracts with Federal Laboratories and Federally Funded Research and Development Centers (FFRDCs). A waiver is no longer required for the use of federal laboratories and FFRDCs; however, proposing small business concerns must certify their use of such facilities on the Cover Sheet of the proposal.
(11) Prior, Current, or Pending Support of Similar Proposals or Awards. If a proposal submitted in response to this BAA is substantially the same as another proposal that was funded, is now being funded, or is pending with another Federal Agency, or another or the same DoD Component, you must reveal this on the Proposal Cover Sheet and provide the following information:

a. Name and address of the Federal Agency(s) or DoD Component to which a proposal was submitted, will be submitted, or from which an award is expected or has been received.

b. Date of proposal submission or date of award.

c. Title of proposal.

d. Name and title of principal investigator for each proposal submitted or award received.

e. Title, number, and date of BAA(s) or solicitation(s) under which the proposal was submitted, will be submitted, or under which award is expected or has been received.

f. If award was received, state contract number.

g. Specify the applicable topics for each SBIR proposal submitted or award received.

We also need a topic to propose to, let’s do something fun

OSD233-001: Deep Rational 3D Geospatial Analytics for Generative AI

ADDITIONAL INFORMATION N/A

TECHNOLOGY AREAS: None

MODERNIZATION PRIORITIES: Artificial Intelligence/ Machine Learning | Autonomy

KEYWORDS: 3D geospatial analytics; generative AI; AI/ML

OBJECTIVE: Develop advanced AI/ML algorithms that combine generative AI with discriminative AI to enhance 3D geospatial analytics.

DESCRIPTION:

DESCRIPTION: Current artificial intelligence and machine learning (AI/ML) techniques for geospatial analysis use pixel or voxel information for semantic segmentation, detection, and classification tasks, but do not exploit the rich contextual relational information in the scene (e.g., cars drive on roads, ships/boats sail on water, etc.). Scenes are fundamentally compositional, adhering to relational rules, which can be exploited to improve geospatial analytics. We seek approaches that combine generative AI, which can describe relations between objects, with discriminative AI to improve multi-task geospatial analysis. Specifically, this topic seeks approaches that leverage generative AI in the form of large language models capable of generating relationships between objects while capturing the relational diversity present in the real world (e.g., cars drive on roads, cars park in drive ways, drive ways connect houses to roads). The proposed approaches must combine generative AI with discriminative AI such as deep convolutional neural network models for segmentation and classification in an end-to-end system. The relations from the generative AI can be considered as constraints and regularization that aid the discriminative AI in solving highly under-constrained problems in 3D geospatial analysis.

NGA anticipates that Phase II work will involve input data that may be Controlled Unclassified Information (CUI) or classified.

PHASE I: Demonstrate proof-of-concept approach capable of combining generative AI with discriminative AI using open source 3D datasets derived from commercial satellite (COMSAT) imagery and full motion video (FMV). For a given set of classes (buildings, houses, cars, roads, trees), demonstrate that approach is able to improve multi-task performance in semantic segmentation and object detection beyond baseline approaches that utilize only discriminative AI. Develop a Phase II plan that includes integration, test, and validation of the end-to-end system.
PHASE II: Realize the optimization and implementation of the selected generative AI and discriminative AI into an end-to-end model. Demonstrate the proposed model is trainable with an expanded class/target set, and is able to perform inference on 3D datasets derived from COMSAT and FMV. Develop a technology transition plan and business case assessment.
PHASE III DUAL USE APPLICATIONS: 3D geospatial analytics software leveraging generative AI capable of supporting DoD use cases including scene segmentation, classification, and target detection; civil engineering missions such as surveying, urban mapping and city planning; commercial robotics applications such as route planning.

REFERENCES:

Zhao, Wayne Xin, et al. "A survey of large language models." arXiv preprint arXiv:2303.18223 (2023)
Xiao, Aoran, et al. "Unsupervised Point Cloud Representation Learning with Deep Neural Networks: A Survey." IEEE Transactions on Pattern Analysis and Machine Intelligence (2023).

TOPIC POINT OF CONTACT (TPOC):