Video course
Judgement
Provenance, licences, not established, and diagnosing your own failures.
Take the course with you
Read the cards
- Training data and copyright uncertainty
- Open versus closed weights
- Diagnosing generation failures
- Separating a vendor claim from evidence
What does Open weights mean in Open weights versus closed API: renting or owning the model?
A model whose trained parameters can be downloaded and run locally, regardless of whether its licence meets any open-source standard.
Read the sourceWhat does Open Source AI Definition mean in Open weights versus closed API: renting or owning the model?
The Open Source Initiative's 1.0 standard, which asks for information about training data, the training and inference code, and the weights.
Read the sourceWhat does Version pinning mean in Open weights versus closed API: renting or owning the model?
Locking a production to one exact model build so results stay stable across a revision cycle.
Read the sourceWhat does On-premises inference mean in Open weights versus closed API: renting or owning the model?
Running generation on hardware you control so confidential material never leaves your network.
Read the sourceWhat does Community licence mean in Open weights versus closed API: renting or owning the model?
A vendor-written licence that permits broad use with carve-outs, such as territorial exclusions or separate commercial terms, and is not an open-source licence.
Read the sourceWhat does Fair use mean in Training data and copyright: what is not settled?
A US statutory defence that can excuse otherwise infringing copying based on a multi-factor analysis, applied case by case rather than as a blanket rule.
Read the sourceWhat does Fair dealing mean in Training data and copyright: what is not settled?
The narrower UK and Commonwealth equivalent, available only for specified purposes rather than as an open-ended balancing test.
Read the sourceWhat does Provenance record mean in Training data and copyright: what is not settled?
The log of model, model version, prompt, seed, parameters and input references behind a generated shot, kept so a delivery can be explained later.
Read the sourceWhat does Indemnification mean in Training data and copyright: what is not settled?
A vendor's contractual promise to cover defined legal costs arising from use of its outputs, usually subject to caps, conditions and correct use of the product.
Read the sourceWhat does Multidistrict litigation mean in Training data and copyright: what is not settled?
A US procedure that consolidates related federal cases before one judge for pretrial purposes, which is why several separate complaints can appear under a single docket.
Read the sourceWhat does Object permanence mean in Why models fail: hands, text, physics and permanence?
Whether things persist correctly through occlusion and after leaving frame, which current models hold often but not reliably.
Read the sourceWhat does Object state change mean in Why models fail: hands, text, physics and permanence?
The correct transformation of a thing under an action, such as food being eaten, which the Sora report lists as unreliable.
Read the sourceWhat does Out-of-distribution generalisation mean in Why models fail: hands, text, physics and permanence?
Performance on cases unlike anything in training, where video models have been shown to fail rather than extrapolate a rule.
Read the sourceWhat does Case-based generalisation mean in Why models fail: hands, text, physics and permanence?
Producing an answer by imitating the closest training example instead of applying an abstracted rule.
Read the sourceWhat does Temporal flickering mean in Why models fail: hands, text, physics and permanence?
Frame-to-frame instability in texture, colour or detail, tracked as its own evaluation dimension in VBench.
Read the sourceExplain why this claim is wrong: open weights means open source.
The Open Source Initiative's definition requires information about the training data, the training and inference code, and the weights. Most downloadable video models ship under custom licences that do not satisfy it.
Read the sourceExplain why this claim is wrong: open means free to use commercially anywhere.
HunyuanVideo's community licence states it does not apply in the European Union, the United Kingdom or South Korea, and Stable Video Diffusion's community licence directs commercial use to separate Stability terms. Licences differ per model and sometimes per model size.
Read the sourceExplain why this claim is wrong: the closed model is always better.
Hosted flagships often lead on quality, but the gap moves with every release, and quality is one axis alongside control, confidentiality, reproducibility and cost.
Read the sourceExplain why this claim is wrong: pinning the model name in an API call gives identical results forever.
Hosted endpoints are updated underneath the name. Only weights you hold can be genuinely pinned.
Read the sourceExplain why this claim is wrong: self-hosting is cheaper.
You swap a per-generation charge for GPUs, storage, engineering time and idle capacity. It pays off at volume and on confidential work, not automatically.
Read the sourceExplain why this claim is wrong: a court has already ruled that training on copyrighted work is fair use.
No US appellate court has ruled on training fair use. Thomson Reuters v. Ross was argued in the Third Circuit in June 2026 and remains undecided, and the most-discussed district-court analysis, in Bartz v. Anthropic, ended in a settlement approved in July 2026, so it was never appealed and binds nobody.
Read the sourceExplain why this claim is wrong: Getty lost in the UK, so training is lawful there.
Getty abandoned its training claim mid-trial for want of evidence that training took place in the UK, and the court made no finding on how many works were used. That is a narrow judgment about territoriality and evidence, not a ruling that training is lawful.
Read the sourceExplain why this claim is wrong: if the output does not resemble any single training image, there is nothing to argue about.
Output similarity and the copying performed to build the training set are separate questions, and both are being litigated.
Read the sourceExplain why this claim is wrong: open weights means the training data was licensed.
A licence on the weights says nothing about the provenance of the data used to produce them.
Read the sourceExplain why this claim is wrong: a model trained on licensed data removes the risk.
It narrows one risk. Indemnities carry caps and conditions, and prompting for a protected character or a trade mark can still produce an output problem.
Read the sourceExplain why this claim is wrong: these are bugs awaiting a patch.
The Sora technical report presents physical-interaction and object-state errors as current limitations of the approach, alongside long-sample incoherence and objects appearing spontaneously.
Read the sourceExplain why this claim is wrong: a more realistic-looking model has better physics.
Physics-IQ reports that physical understanding across current models is severely limited and unrelated to visual realism, so photorealism is no guide to whether an interaction will be correct.
Read the sourceExplain why this claim is wrong: scaling will fix physics.
A study on a controlled mechanics testbed found in-distribution generalisation but out-of-distribution failure, and concluded that scaling alone is not sufficient for these models to recover fundamental physical laws.
Read the sourceExplain why this claim is wrong: hands and text are the whole problem.
Hands and small text are the visible tip; continuity across shots, weight, contact and object permanence fail more often and are harder to spot in a single viewing.
Read the sourceExplain why this claim is wrong: one quality score tells you whether a model suits your shot.
VBench exists because a single number hides the trade-offs; models differ dimension by dimension, so the useful question is how a model scores on the dimension your shot depends on.
Read the sourceYour progress belongs to you
Course pages do not load analytics. Your answers are not transmitted to Shotcraft.
- No streaks, lives or leaderboard. Return whenever you want.
- No payment or account is required. Export your progress before clearing browser data.