LLMs break down text into “tokens”–which can be words or chunks of words–and use probabilities to predict the next token.
any veterinarians have been reaping the benefits of new artificial intelligence (AI) technologies for over a year now. And thanks to large language models (LLMs), vets can drastically reduce the amount of time they spend each day writing records. These tools quickly transcribe audio recordings from veterinary consultations into comprehensive and accurate medical records.
LLMs are a type of AI that have been trained using vast amounts of text data, enabling them to learn tasks such as predicting the next word in a sentence based on context. Models like OpenAI’s ChatGPT are notable examples, and modern veterinary scribe startups utilize this same underlying technology.
LLMs break down text into “tokens”—which can be words or chunks of words—and use probabilities to predict the next token. For example, if given the tokens “The cat is,” the model might predict “hungry” or “sleeping” as the next word based on learned probabilities. To enhance their human-like qualities, these models incorporate “sampling” to add randomness, preventing repetitive and robotic text generation. Beyond text production, LLMs are also used in modern-day transcribing, converting audio recordings into written form.
LLMs facilitate the creation of medical records by summarizing lengthy consultation transcripts, identifying pertinent information and filtering out irrelevant content. Most existing LLMs were trained on more generalist content, so they may misinterpret or misspell veterinary words, but this is changing with the rise of models that are trained on more specific content.
Of course, many practitioners have been using AI-powered tools for image analysis for a few years now. Scribes and image analysis are only the beginning. Expect many AI tools to emerge over the next short while to help with many elements of practice administration—tasks such as ensuring there are no missed billable items on client invoices, managing insurance claims, coordinating employee rosters and tracking specialist referrals.
AI also has a significant role to play in the critical but often intractable issue of treatment compliance, as well as in improving client communications. Vets will be able to leverage AI’s ability to quickly create tailored client summaries, enhancing the level of personalized care and client buy-in. It also has the ability to “learn” about habits and behaviors that may interfere with client compliance, and counteract them with personalized communication strategies in the home setting.
From a clinical standpoint, AI’s unprecedented pattern recognition abilities will prove useful as a diagnostic aide and assistant in case management. While it may never substitute the intricate and nuanced clinical reasoning of a veterinarian, it does have the potential to augment such decisions. These are exciting times ahead, both from a clinical perspective and also in terms of the impact that reduced workloads will have on veterinary morale and workplace satisfaction.