“Red Hat Product Security is responsible for ensuring that the software we release is free of security flaws, and to respond to new vulnerabilities that appear. My team is tasked with automating processes that allow us to ensure the integrity of the software supply chain. As a software company, we have a number of requirements we need to meet, including industry certifications, government requirements, specifications from customers, and industry-best practices. Most of our clients are internal development teams. That said, the information we generate is used by our sales and support teams to respond to customer requests regarding how we are ensuring secure code releases. We have a pretty broad customer base, everything from governments to the financial sector to small mom and pop shops. The guidance I’ve given my team is to identify problems and then see if an AI tool or workflow would be the best solution. Some of the tools we’re either using or testing include Ansible Lightspeed, ChatGPT, and PentestGPT. We’re primarily focused on using AI to increase efficiency. Many of our tasks require analysis and writing documentation, so using an AI tool to reduce the number of hours it takes a human to do the same work means we can be more effective. AI should be used to enhance what we can already do and know. It allows us to be more efficient and accurate, but it doesn’t replace the skills required to actually produce the work. Generative AI is a revolutionary leap in how we do things, but WE still have to DO them.” Sean Waite ’07 Manager, Product Security Team, Red Hat Inc. Computer science and music dual major “In my position as a research nurse specialist, I train the AI platform that we use with natural language processing. What this means is I manually go into our training environment and mark out relevant information from ingested provider notes from our client’s medical records systems. For instance, this may include pulling out a medical diagnosis. An example of this would be ‘diabetes mellitus- type 2.’ I then identify it with a marker which tells our AI it is a medical diagnosis. I then attach the appropriate ICD-10 diagnosis code. By doing this, we train our AI to recognize medical terms (including medications, procedures, lab values, and other specific metrics) that may not be entered into a discreet field in the patient chart. This is especially useful in cases where physicians may document in notes slightly differently from the standard. It also is able to capture variations in terms of the sheer volume of training data we have. The biggest drawback is that AI is only as good as the trainer. However, the trainers are all registered nurses with the exception of a team member who is a Ph.D.-trained scientist. One of the biggest trials with the use of AI is that it can become ‘confused’ when it is identifying certain medical terms. One particular case where this has been especially noticeable is the identification of a specific genetic marker for cancer. The abbreviation for this marker is EGFR, and it represents a specific gene found in some cancers. However, eGFR is a specific lab and an abbreviation for a different test which is a marker of kidney function.” Brittane (Miller) Strahan, MSN, RN, CCRP ’09 Research Nurse Specialist at BEKhealth Biology major “I develop AI applications for government use. The main challenges are model hosting, on-premise deployments, and security. For example, government systems have additional compliance requirements such as FEDRAMP and ITAR that most commercial AI solutions do not. Specifically, I work on a project called LIGER. A goal of LIGER is to bring the user experience of ChatGPT into government applications. For example, you can’t put classified data (or even controlled unclassified data) into ChatGPT without it becoming a massive security breach because that data is going out of controlled networking boundaries. LIGER tries to bring both chat-my-data (i.e. ChatGPT) as well as catalogue/manage my document (similar to Google Drive’s Gemini integration) into enterprises. In addition, we have some bespoke use cases. For example, if a client wants specific features for their documents/data, we build it into the platform. I can’t talk about specifics, but hypothetically, imagine a branch of the government called ‘FooGov’ had a corpus of thousands of policy regulation documents dating back to the 1970s and they wanted to ask questions like, ‘If I change this policy in a new bill, would that violate policies in any of these existing docs?’ This is the type of thing we’re building. There’s never been a class entering the job market that’s more equipped to hit the ground running than the students of today. I think Lycoming students should pursue what they love but be realistic about the way AI is going to affect it and really lean into it.” Adam Hughes ’08 Software Developer, LMI Physics major 21 www.lycoming.edu
RkJQdWJsaXNoZXIy NTA3NDk=