Im Strudel des technologischen Fortschritts stehen derzeit vor allem Chatbots mit künstlicher Intelligenz wie ChatGPT und Claude im Rampenlicht. Diese bahnbrechenden Technologien, die auf großen Bild-Sprache-Modellen (VLMs) basieren, sind mehr als nur intelligente Maschinen. Sie wurden anhand einer riesigen Datenmenge aus Büchern, Websites, Programmcode und Bildern trainiert und können alles bewältigen – vom Verfassen professioneller E-Mails bis hin zur Erstellung komplexer Reisepläne.
What makes these AI chatbots even more impressive is their ability to learn and evolve through extensive human feedback. Developers meticulously pour over each interaction to ensure these digital bots follow instructions accurately and avoid delivering unwanted or damaging output. The result? Chatbots that can deftly generate text or images based on the user’s input, all while opening doors to traditionally specialized fields like computer programming. Yet, let’s remember, they might have their limitations.
A fascinating project, spearheaded by the U.S. Department of the Air Force–MIT AI Accelerator’s Phantom Program, takes AI chatbots to an all-new level. Imagine a complete novice developing a fully functional program using AI chatbots – that’s precisely what U.S. Air Force cadet Joshua Lynch did, under the mentorship of Laura Niss from MIT Lincoln Laboratory.
Lynch’s innovative approach, dubbed “vibe-coding,” involved steering an AI chatbot to write and refine code purely based on verbal prompts. His ultimate goal was to arm the military workforce with the ability to transform their software ideas into reality without the hoops of traditional military software development’s restrictions.
Over three months of arduous work, Lynch successfully developed the Remote Operating Modular Augmentation Device (ROMAD-AI) even while battling typical AI challenges like a lack of hierarchical focus and non-correlated changes in the code sections. As a newbie in the AI world, he worked mainly via the chat functions of three AI chatbots: Anthropic’s Claude, OpenAI’s ChatGPT, and Google’s Gemini.
Navigating through these challenges gave Lynch a new perspective on problem-solving. Right from dissecting larger problems into manageable bits to framing questions with precision, he learned to direct the conversations seamlessly. Recognizing AI’s limitations and crafting smart workarounds became a significant part of his project journey.
While the developed prototype didn’t fully align with Lynch’s original plan, it revealed substantial potential. The concept of leveraging AI chatbots for tasks like analyzing tactical maps or designing mission-planning documents was indeed promising. However, this progress wasn’t without a few hiccups. Lynch discovered a security risk – the application sent input documents to a Gemini AI model for analysis rather than analyzing them on-site.
Yet, as Niss observed, Lynch’s evolving understanding of AI language models paved the way for greater accomplishments. AI chatbots could indeed empower non-technical users to create practical software applications. They could serve as more of a prototyping assistant rather than full-blown production tools dealing with sensitive information.
Niss also stressed the indispensable role of collaboration in such technologically advanced projects. In her words, “No matter how good AI gets, we’ll always need to collaborate to get to the best solutions for the most important problems.”
Dieses ehrgeizige Projekt wurde vom „Artificial Intelligence Accelerator“ des US-Luftwaffenministeriums finanziert. Erfahren Sie mehr über dieses Projekt im Original-Artikel. hier. Und falls Ihr Unternehmen auf der Suche nach KI-Automatisierungslösungen ist, sollten Sie das Potenzial von implementi.ai.
Diese Website verwendet Cookies.