Praesto uses AI
Praesto uses third-party large language models (currently hosted by Microsoft on Azure OpenAI Service and Azure AI Foundry) to generate things like:
- Summaries of your transcripts and notes.
- Flashcards, practice problems, and explanations.
- Topic maps of what you've covered.
- Answers in the chat, research, and "clever questions" features.
- Inline quizzes generated from your study material.
- Group study signals: in group conversations, how the group answered a quiz, how hard the class found a problem, and follow-up problems generated from the group's most common wrong answer.
We call these AI Outputs. They are produced by a probabilistic model that predicts likely text. They are not a search of an authoritative database, and they are not reviewed by a human before you see them.
What AI Outputs can do badly
AI Outputs can be wrong, in any of these ways:
- Fabrication ("hallucination"). The model can produce confident-sounding text that has no basis in your transcript or in reality.
- Omission. The model can leave out the part of your transcript that actually matters.
- Stale knowledge. The model's general knowledge has a cutoff date. Anything more recent than that cutoff is unreliable unless we have grounded it on your transcript or a document you uploaded.
- Citation errors. When the model includes a citation, the citation can point to the wrong source, to a non-existent source, or to a real source that does not actually say what the AI claims it says.
- Math and code errors. The model can produce arithmetic and code that looks right but is not.
- Bias and one-sidedness. The model can present a contested topic as if it were settled, or repeat stereotypes from its training data.
This is not a defect we expect to fix. It is the nature of the technology today. The right way to use Praesto is to treat AI Outputs as a starting point, not a destination.
What you should do
- Verify anything important before you rely on it. Read your own transcript, check the document you uploaded, look up the citation, do the math.
- Don't paste AI Outputs into work you are submitting under your own name without checking what your school, your employer, or your licensing body permits.
- Don't make decisions you'd regret on the AI's say-so. Especially for medical, legal, financial, tax, professional licensing, immigration, or safety decisions.
- Don't use Praesto in an emergency. If you or someone else is in danger, call local emergency services (in the U.S., 911) or a qualified professional. The AI does not know you are in a crisis and is not equipped to help.
Academic integrity
Many institutions restrict or prohibit AI tools in coursework, exams, and licensure. Praesto does not authorize you to use AI Outputs in any way that would violate the rules of your school, employer, or licensing authority. You are responsible for knowing those rules and following them. The fact that Praesto generated the text does not change who is responsible for it: you are.
If you are using Praesto to study, the safer pattern is:
- Use Praesto to explain material, quiz you, or summarize your own notes.
- Do the actual assignment yourself, in your own words.
- If you need to cite a fact, verify it against your transcript, your textbook, or another reliable source, not against the AI.
Not professional advice
Nothing in any AI Output is medical, legal, financial, tax, accounting, mental-health, or other professional advice. Praesto is not your doctor, your lawyer, your accountant, or your therapist. If you need professional advice, talk to a qualified professional.
Sensitive topics
If you ask the AI a question about a sensitive topic (for example, self-harm, suicide, an eating disorder, intimate-partner violence, or a medical emergency) the AI may give you a response that is well intentioned but unsuitable for your situation. Please reach out to a qualified professional or a local crisis resource. If you or someone you know is in immediate danger, call or text 988 (Suicide and Crisis Lifeline) in the United States, or call 911.
How we limit risk on our end
- We do not use Your Content to train AI models, and our AI providers are contractually prohibited from doing so. See the Privacy Policy for details.
- We monitor errors so we can improve prompts and detect systematic failure modes.
- Where supported, the app may indicate which transcript segments or documents the AI relied on, but a citation displayed in the app is not a guarantee that the AI used it correctly.
Bottom line
Treat Praesto's AI like a smart but unreliable assistant. Use it to help you think faster, surface things you missed, and quiz yourself. Don't use it as the last word on anything you would not be comfortable defending in your own words.