The supplied search summary reports that ByteDance is at an early stage of developing a large language model targeting as many as roughly 10 trillion parameters. However, none of the accompanying result snippets mentions ByteDance or independently substantiates that project; one snippet specifically cautions that running data through a huge model to measure throughput is not equivalent to completing a training run. On the supplied evidence, this remains an unconfirmed report rather than credible proof that ByteDance has begun full-scale frontier training.
Scott’s Evidence Class Ladder and Discussed Is Not Deployed frameworks already require treating this as an unverified report rather than a completed training run. Confirmation of a roughly 10T-parameter model would nevertheless materially extend the scale tracked in the Qwen 3.8 case and require scrutiny of total versus active parameters, MoE architecture, and infrastructure economics.
ip:concept.evidence-class-ladderip:framework.discussed-is-not-deployedradar:qwen-3-8-rolloutradar:concept.frontier-modelsradar:concept.mixture-of-expertsradar:concept.ai-infrastructure
queries asked of Scott's wikis
- parameter count versus active compute as a frontier-model signal
- credible evidence standards for claimed training runs
- frontier scaling beyond one trillion parameters
- Chinese AI labs and model sovereignty
- mixture-of-experts scaling and total parameter claims
- training-scale economics and infrastructure constraints
2026-08-15T10:27:15Z
Repeated staleness checks have produced no first-party statement, technical disclosure, or genuinely independent confirmation that the reported 10T-parameter training run began. This episode has exhausted its informational value and can expire; a later substantive disclosure should be treated as a fresh delta.
2026-08-13T09:34:19Z
Another staleness check adds only minor engagement drift and no first-party confirmation, technical disclosure, or independent evidence that training began. The credible secondary report remains worth dormant monitoring, but repeated amplification has no further informational value.
2026-08-11T08:30:24Z
No substantive evidence has arrived during the staleness window; the story remains a credible secondary report but still does not establish that ByteDance began a roughly 10T-parameter training run. Move it to dormant monitoring until a first-party statement, technical disclosure, or genuinely independent report appears.
2026-08-09T08:26:38Z
The newly attached HN item is another pointer to the same secondary account, not an independent line of evidence that ByteDance has begun a roughly 10T-parameter training run. Repeated amplification adds no clarity on training status, architecture, or total versus active parameters, so only a slow watch remains warranted.
2026-08-09T08:21:46Z
evidence attached: hn.story.49229483 — shared external link with case evidence
2026-08-08T11:25:33Z
The video-linked HN post is another amplification of the existing secondary report, not an independent line of evidence that ByteDance has begun the claimed training run. The case remains worth a slow watch, but total versus active parameters, architecture, and actual training status are still unresolved.
2026-08-08T11:21:46Z
evidence attached: hn.story.49220535 — This report is weak video evidence but directly bears on the open hypothesis that ByteDance is pursuing a roughly 10T-parameter frontier model.
2026-08-08T10:26:08Z
Ars coverage broadens credible secondary reporting that ByteDance is pursuing an unusually large model, but appears to repeat the same underlying account rather than independently establish that a roughly 10T-parameter training run has begun. The case remains worth watching, with total versus active parameters and actual training status unresolved.
2026-08-08T10:21:58Z
evidence attached: reddit.post.1virisx — Independent press coverage provides corroboration that ByteDance is pursuing an unusually large frontier model.
2026-08-07T21:37:37Z
Credible secondary reporting upgrades this from a social-media rumor to a report worth tracking, but the Reddit and HN items appear to amplify the same underlying account rather than provide two independent lines of evidence. Whether training has actually begun—and what 10T means in total versus active parameters—remains unverified.
2026-08-07T17:22:55Z
evidence attached: hn.story.49212923 — Independent reporting that ByteDance is targeting a mega-model nearing Anthropic's reported scale materially supports the existing ByteDance frontier-model case.
2026-08-07T16:27:05Z
The latest trigger is another reobservation of the same uncorroborated Reddit claim, with no first-party statement, independent reporting, or technical evidence that training has begun. Engagement-driven monitoring is yielding no information; retain only a slow watch for substantive confirmation.
2026-08-07T15:29:10Z
The latest attachment is another reobservation of the same Reddit claim, not independent evidence that ByteDance has begun a 10T-parameter training run. Engagement-only churn is exhausted; keep a slow watch for first-party confirmation or credible technical reporting.
2026-08-07T14:23:30Z
The supposed new evidence is another attachment of the same Reddit report, with no independent sourcing or proof that ByteDance has begun training. Repeated amplification has exhausted its informational value; retain only a slow watch for first-party confirmation or credible reporting.
2026-08-07T13:27:52Z
The apparent new evidence is only a reobservation of the same Reddit claim; neither the comments nor engagement provide independent sourcing or proof that training began. The case remains speculative and should wait on first-party confirmation or credible reporting rather than receive further engagement-driven checks.
2026-08-07T12:30:06Z
The latest observation still adds no independent source, first-party confirmation, or technical evidence beyond the original Reddit claim. Repeated engagement-driven rechecks are now pure amplification, so the case should remain a low-temperature seed on a slower cadence.
2026-08-07T11:21:55Z
The newly attached observation still traces to the same uncorroborated Reddit report; engagement and speculative comments add no independent evidence that training has begun. Repeated amplification is not changing the case, so it can move to a slower watch cadence pending first-party confirmation or credible reporting.
2026-08-07T10:26:31Z
The newly attached material is still the same Reddit report, with comments offering speculation rather than independent sourcing or technical evidence. Engagement has not changed the case’s meaning: the 10T target remains an unverified claim awaiting first-party confirmation or credible reporting.
2026-08-07T09:26:48Z
The attached evidence remains the same uncorroborated Reddit report, and its comments add speculation rather than independent sourcing or technical proof. With frontier-model attention fading and no ByteDance confirmation, the case’s meaning is unchanged.
2026-08-07T08:26:08Z
The new observation is only negligible engagement drift, with no first-party confirmation, independent reporting, or implementation evidence. The 10T training claim remains an uncorroborated report despite a hot frontier-model neighbourhood.
2026-08-07T07:26:27Z
grounded: known/medium — Scott’s Evidence Class Ladder and Discussed Is Not Deployed frameworks already require treating this as an unverified report rather than a completed training ru
2026-08-07T07:22:51Z
case created — The reported training target is concrete, unusually large, and consequential enough to track pending first-party confirmation or further reporting.