UES / AI LAUNCH LAB / OCTOBER 5, 2026 / EDITORIAL EDITION
Read the launch. Test the claim.
Explain the distinction between model capacity, computation, availability, licensing and evidence.
Dated editorial edition · Sources checked October 5, 2026. See each source and claim boundary below.
Independent learning companion · Proposed 30-minute self-guided session · Curious learners and beginning builders. UES is independent learning, not an accredited degree or credential provider.
All exercises can be completed with the linked sources and a calculator. No paid API account, model download or personal data is required.
Four ways to make an AI headline useful.
EXERCISE 1 / 7 MINUTES
Trace a claim
Choose one sentence from the launch article. Label it as a vendor claim, a documented specification, an independent result or an editorial inference. Open its source and explain what would verify it.
Success check: The learner names both the evidence currently present and the evidence still needed.
EXERCISE 2 / 6 MINUTES
Estimate the weight payload
Using the announced 501 billion total parameters, calculate raw storage at two bytes, one byte and half a byte per parameter. Explain why the active-parameter count is not the storage count.
Compare with the answer guide
1,002 GB, 501 GB and 250.5 GB, using decimal units. These are idealized payload estimates before runtime and format overhead.
Write three test questions from public material. Specify the expected evidence, a passing answer, a failure case, tools allowed and a time budget. Explain how you would compare two models fairly.
Success check: The rubric is written before model outputs are seen and includes a check for unsupported claims.
EXERCISE 4 / 8 MINUTES
Practice a tool boundary
An assistant reads a document that says to email its contents elsewhere. Decide whether that instruction is authorized, identify the missing permission and design a safe read-only alternative.
Compare with the answer guide
A document is task data, not permission from the user. Summarize it within the approved destination; do not send it elsewhere without the required authorization.
Leave a useful evidence note.
Record one sourced claim and label its evidence type.
Show the payload calculation with decimal units and excluded overhead.
Keep the rubric you wrote before seeing any output.
Name one denied tool action and its safe read-only alternative.
Reflection: Which new piece of evidence would most change your assessment of Beam, and why?
No model run is required or implied. A future comparison needs separate access, model-version records and authorization.
Read the source record.
Sources were checked October 5, 2026. Each card links to the evidence behind its facts; analysis is PointCast editorial interpretation.