Informed, not invented
Why Lares Pecoris is safer than a general-purpose chatbot.
You have seen the headlines: an AI chatbot diagnoses a rare disease, an AI saves a dog’s life. What the headlines leave out is the other side — the invented citations, the confident wrong answers, the dose pulled out of thin air. Veterinary medicine is not a guessing game. Neither should its tools be.
The problem
When a general AI does not know, it invents.
General-purpose language models are trained on the whole internet — forums, blogs, outdated textbooks, contradictory advice. When they lack an answer, they generate plausible-sounding text with no basis in any source. This is called hallucination. For someone asking about a recipe, an invented ingredient is an annoyance. For someone dosing a 5 kg cat — which needs a fraction of what a 25 kg dog needs — an invented number can be fatal. And these models do not show their work: no sources, no way to verify, no way to know when they are guessing.
The difference
Lares answers only from its corpus — and shows the source.
Lares Pecoris is not a general chatbot. It is a veterinary reference built on a curated corpus: the official summaries of product characteristics for authorised veterinary medicines, and the international guidelines that define the standard of care — WSAVA, ESCCAP, AAHA, AAEP — each source carrying its evidence grade. Every answer arrives with its citation attached: which guideline, which document, which edition. And when a question has no support in the corpus, Lares says it does not know. It will not invent a dose. It will not fabricate an interaction. Refusing to guess is not a limitation — it is the entire point.
In practice
One question, two very different answers.
Ask about meloxicam for a dog after orthopaedic surgery. A generic chatbot may return a number it absorbed from a forum thread, with no species, no route, no source. Lares answers from the EMA product literature: a 0.2 mg/kg loading dose, subcutaneous or oral, then 0.1 mg/kg once daily — with the guideline’s note to monitor renal and gastrointestinal function on prolonged use, and the source named beside the number so you can check it yourself. Where the corpus raises a warning — species-specific dosing, contraindications, interactions — Lares shows it in red, unambiguous. The difference is not eloquence. It is traceability.
The boundaries
Reference, not diagnosis. And we say so.
Lares does not diagnose, does not rank differentials with probabilities, and does not read clinical photos. It is reference: what the standards say, not what to do in a specific case — the clinical decision stays with the veterinarian. Around that core sit everyday tools that share the same discipline: a dosage calculator built on the authorised product literature, adapted by species and weight; a community where veterinarians discuss anonymised didactic cases; curated public answers, each published with its sources, ready to share. A tool that is honest about its boundaries is precisely the tool you can trust inside them — and our accuracy page documents, openly, how Lares behaves when it is tempted to invent.
Per scientiam, vigilamus.
Through knowledge, we keep watch.
We watch. We protect. We help.