Monday, November 2, 2026HotTea archive editionVerified 12:31 AM PST

The lead

SemiAnalysis's new AI work test tops out at 12 percent

Claude Opus 5.5 leads a 206-task benchmark. Its weighted score measures performance on selected tests, not a share of all jobs.

Listen to this edition

Prefer audio? The briefing has chapters and a full transcript.

The briefing

The rest of the morning

4 more stories

02

KAIST reports up to 69.6 percent less fine-tuning time

KAIST announced CluSTER on November 2, after a September preprint and an October conference presentation. The method picks less-repetitive training data and divides it among GPUs. The team tested models with about 7 billion and 13 billion parameters on three coding and medical datasets, using eight performance benchmarks.

KAIST reports nearly the same performance with about 31 percent of the training data, cutting time by up to 69.6 percent. These are results from interested researchers on selected workloads, not a measured saving for every model or proof of medical benefit.

KAIST ↗CluSTER authors, arXiv ↗
Read article →
03

ADNOC signs an AI development deal with OpenAI and G42

ADNOC announced a multi-year partnership with OpenAI and G42 on November 2. The partners plan to link oilfield development planning with commercial decisions through OpenAI Frontier on UAE-based G42 infrastructure. ADNOC says human experts will keep oversight of consequential decisions.

ADNOC is an interested company source describing joint development and deployment plans. Its aim to prepare plans in hours rather than months is not an independently measured result from this new agreement. A signed deal does not show better investment decisions or higher returns.

ADNOC ↗
Read article →
04

Grok API text watermarks are scheduled for November 7

SpaceXSI's documentation, last updated October 31, says supported Grok models will watermark API text by default from November 7. The mark changes token choices instead of adding visible characters. Team administrators can request model-specific opt-outs, subject to the documented limits.

This is a company description of a scheduled rollout, not proof the new defaults are active today. Some models can remain locked on. Detection is statistical, can return an uncertain result, and requires approved access to the detection API.

SpaceXSI ↗
Read article →
05

Seven OPEC+ countries keep December oil targets unchanged

Seven OPEC+ producers agreed on November 1 to keep December required production at September levels. OPEC names Saudi Arabia, Russia, Iraq, Kuwait, Kazakhstan, Algeria and Oman. Reuters also reported the decision. The group meets again on November 29.

Required production is a policy target, not a measure of oil produced or delivered. The decision alone cannot show whether actual supply, shipping costs or fuel prices will change.

OPEC ↗Reuters via StreetInsider ↗
Read article →

The watchlist

Signals that could change the read

Agent workIndependent task reruns and costs including review and correction.Tracking
Training efficiencyCluSTER replication on new data, including selection costs.Tracking
Industrial decisionsADNOC deployments and evidence of human oversight in practice.Tracking
November datesGrok watermark settings after November 7 and OPEC actual output before November 29.Tracking
Across the desks Check the completed work

New task scores, faster training, and industrial AI plans each need a different test.

Agent task scoresHotSemiAnalysis reports a 12 percent leading score on its new work benchmark.
Training efficiencyWarmKAIST reports shorter fine-tuning on selected coding and medical datasets.
Industrial AIWarmADNOC, OpenAI and G42 announce joint development and deployment plans.
Rules and supplyWarmGrok documents a future watermark rollout; seven oil producers hold December targets.

Editorial direction, not a financial index. Each signal is tied to this edition’s reporting.

Edition validated · 5 stories · 8 unique sources

About HotTea & our sources →