Learn AIProject 6 · Saved comparison workbench
A controlled generative experiment.
Compare two actual generative mechanisms on the same four-mode target. Freeze numeric snapshots, keep failures, explain the objectives and make a limited decision from evidence. Report submission, independent assessment and human review are different outcomes.
Report: draft. Saved on this device and included in version 4 progress backups. Export before clearing browser data or changing devices. Edits reopen the draft without changing existing lesson achievements.
Use lab controls and explain the evidence; coding is optional. You can change paths; these project variants do not change published learning-path requirements.
1. Freeze matched evidence
Use the save button in a completed diffusion/flow run or the GAN lab. All three use the same four-mode training data. Four captures maximum; include at least two different trained mechanisms at the same seed. Sampling always produces 24 points. Matching seeds does not make training budgets, architectures or stochastic trajectories equivalent.
The VAE lesson uses a different target, and conditioning changes the target task, so those exports are not accepted as matched evidence here. They remain useful separate practice. Exported scores and claimed histories are ignored; samples are generated again from the captured parameters.
No snapshots saved yet. Train and capture in a lab, import its export, or begin with the supplied reference.
2. Explain and justify the comparison
Code remains text and is never executed on import or submission. Implementation quality requires independent review.
3. Submit a report and identify remaining evidence
Submission checks presence of required text and two distinct trained mechanisms with a matching seed. It does not assess reasoning, usefulness, code correctness, authorship or learning transfer. The five separate cases below check scalar arithmetic and evaluation design. Human review of reasoning and code remains pending. No project mastery or path achievement is awarded by this form.
- Capture two different trained mechanisms using the same sampling seed and shared four-mode target.
- Write the purpose report field.
- Write the protocol report field.
- Write the objectives report field.
- Write the evidence report field.
- Write the failures report field.
- Write the decision report field.
To restore saved work, export/import full progress through My Work. This report download is a review artifact, not a full-progress backup.
Try a different problem independently
These five fixed cases use different data and separate scoring arithmetic from the guided generation lessons. Work without the walkthrough first. Opening and submissions are saved; the first submission is retained, with at most two practice retries. Reopening, resetting or importing an earlier backup does not erase known exposure. This is a local objective check, not authenticated exam supervision or human review of your implementation.
0/5 cases passed on first submission. Latest/practice passes: 0/5. Project evidence, these outcomes and pending human review are separate.