AI disclosure
Everything aitubenotes gives you is written by machines, with no person reading it first. That is the product, not a caveat buried in the terms, so here is exactly what runs and exactly what it is not able to do.
Is a person involved in producing my notes?
No. The whole run is automated: nobody reviews your notes, checks them against the video, or sees them at all in the normal course of things. The sibling agency product this company also operates puts humans in front of client deliverables. This one does not, and pricing it at the level it is priced at is only possible because it does not.
A person can see your data when something is being fixed or investigated, and the Privacy Policy says when.
Which models read my video?
- Speech. OpenAI's Whisper model, in its large-v3 size, run on our own hardware. Your audio is not sent to OpenAI: the model is open and we load it ourselves. Where a video already carries a caption track we use that instead and skip this step.
- Text on screen. EasyOCR, an open optical-character-recognition model, run on our own hardware over the deduplicated frames.
- What a frame shows. BLIP, an open image-captioning model, again on our own hardware, for frames whose meaning is not carried by text.
- Writing the notes. Claude, from Anthropic, reached through Anthropic's own interface. This is the one step where material from your video leaves our infrastructure. Under the terms we use, Anthropic does not train its models on it.
We may change any of these as better ones appear. You are buying notes, not a particular model, and we will keep this list accurate rather than keep the list frozen.
How are the notes marked as machine-generated?
The notes file carries a line at the top saying a machine wrote it, which video it came from, and when. It is plain text in the document itself, so it survives being copied, pasted and forwarded, and anything reading the file can see it without being told to look.
This is deliberate and it is not decoration. EU law from August 2026 requires content generated by an AI system to be marked as such in a way a machine can detect. Removing that line before passing the notes to somebody else is a breach of the Acceptable Use Policy.
What is the machine not able to do?
- Know whether the video was right. If a presenter states something false, the notes will record it faithfully as though it were true. The notes are a record of what was said and shown, not a check on it.
- Read what it cannot see. A blurred slide, a low resolution recording, a chart with no labels, handwriting: all of these produce gaps, and sometimes produce a confident guess in place of a gap.
- Hear everything. Accents, crosstalk, background noise and technical vocabulary all cost accuracy in transcription, and an error there propagates into the notes.
- Tell you when it is unsure. Machine-written prose reads with the same confidence whether the underlying material was clear or a mess. This is the failure mode most likely to catch you out, which is why it is in the list rather than in a footnote.
So: check anything that matters against the video, and do not rely on the notes alone for anything medical, legal, financial or safety-critical.
Is my video used to train AI models?
No. Not by us, and not by the provider we send material to. We do not sell your videos or your notes, and we do not build training sets from them. The models we run on our own hardware are used as they were downloaded and are not adjusted using your material.
What do we do if the machine gets something wrong?
Tell us and we will look at it. If a run produced something clearly broken rather than merely imperfect, we will credit the minutes back. Finding out that something is wrong is worth more to us than the minutes are, so please do write.
What we will not do is claim the system is more reliable than we have measured it to be. This site publishes no accuracy figure because we have not produced one honestly. When we do, it will appear with the method next to it.