Trump tweaks the Canada auto-tariff list again, third time since July

Good morning 👋 Third time since July, the White House has rewritten which Canadian parts eat the 50% auto tariff. The rate's not moving. The list is.

In today's issue:

• 🔭 Trump tweaks the Canada auto-tariff list again, third time since July
• 🧠 AWS ships auto-scaling fixes and GeoJSON route maps; Bayer builds an agent on Bedrock
• 🔬 How much of a network you actually need to train, and a watermark for stolen art data
• 🛠️ Turn today's tariff proclamation into a supply-chain exposure check
• 🔗 A GPU cloud buyout, a NYSE compliance scare, two drillers merging

🔭 THE ONE THING

🚗 Trump tweaks the Canada auto-tariff list again, third time since July

The administration issued a new proclamation yesterday, swapping which Canadian products eat the 50% Section 338 duty on motor vehicles. 22 items come off the list, about 100 go on, effective October 30 at 12:01am ET. It's the third modification to Proclamation 11048 since July, and Canada is still sitting on roughly $20 billion in retaliatory tariffs from the original fight. Nobody's rate is moving. What's moving is the product list, which means anyone sourcing auto parts from Canada now has to track a shifting target instead of a fixed number, and this administration clearly isn't done editing it.

🧠 MODELS & RELEASES

• 🧩 AWS Published a reusable CDK construct that deregisters a SageMaker endpoint's Application Auto Scaling before a CloudFormation update and re-registers it after, so deploys stop failing just because scaling is attached. AWS Machine Learning blog
• 🗺️ AWS Amazon QuickSight's layer maps now read Point and LineString geometry from one GeoJSON dataset, so you can plot locations and the routes between them without juggling two separate layers. AWS Machine Learning blog
• 🌱 Bayer Crop Science Built an "Agentic Decision Engine" on Amazon Bedrock AgentCore, a multi-agent system that lets breeders and analysts query their analytics platform in plain English. Worth reading for the five design decisions AWS walks through, not just the demo. AWS Machine Learning blog
• 📚 Ai2 and UPenn built LitPivot, a tool that checks a research proposal's problem, contribution, and evaluation plan against relevant literature and flags exactly where it's weak. In a 17-researcher study it beat a plain editor-plus-chatbot setup on grounding and perceived novelty. Hugging Face blog

🔬 RESEARCH HIGHLIGHTS

• Partial fine-tuning, quantified A theory paper on two-layer networks found that training just 4.1% of the parameters, with the rest frozen as random features, captures 90% of the benefit of training the whole thing. The pattern held up through their 15-layer experiments too. If you've been doing LoRA-style partial fine-tunes on instinct, there's now a number behind it: most of the gain lives in a sliver of the weights. Paper
• A budget for human review Gibbs, Han, Tibshirani, and Zrnic built an online auditing scheme that decides, decision by decision, whether a human needs to check an AI system's output. It guarantees the error rate on everything that skips review stays under a fixed threshold, no assumptions about the data required, while auditing as little as possible. They tested it on harmful-request screening for LLMs, text annotation, and data labeling. For anyone running a moderation queue behind a model, this is a way to shrink review volume without flying blind. Paper
• A watermark for stolen training data SpectralCanary blends a low-frequency signal into images before release. Invisible to the eye (SSIM 0.989), but detectable at 92.4% accuracy with a 1% false-positive rate if a text-to-image model trains on the set. Scrubbing it takes enough distortion (25+ FID points) to wreck the image anyway. Worth knowing if you license image data and want proof when someone trains on it without asking. Paper

🛠️ TRY THIS

Turn a tariff proclamation into a supply-chain exposure check

1. Pull the full text of the proclamation modifying Canada's additional-duty scope (the Federal Register entry, not a news summary).

2. Feed it to an LLM and ask it to extract every HTSUS code added or dropped from the list, with the rate attached to each.

3. Paste in your own BOM or supplier code list and ask it to flag overlaps.

4. For each hit, have it estimate the landed-cost delta at your current order volume.

Prompt: Here is the full text of a US proclamation modifying tariff scope on Canadian-origin products. Extract every HTSUS code it adds or removes from additional duties, with the duty rate for each. Compare against this list of my product codes: [paste codes]. Flag matches and estimate added or removed cost per unit.

Worth a look

• Folkbench tracks uptime, latency, and price across AI API relay services, then lets you one-click swap the Base URL and key for Claude Code, Codex, and similar tools from a desktop app. Open source, Apache-2.0, past 3,200 stars. github.com/Folkbench/folkbench

🔗 QUICK LINKS

• Reliance Global Group (EZRA) signed a letter of intent to buy a GPU cloud and marketplace operator, funded entirely in cash, and will rebrand as GPU Exchange under the ticker XGPU once it closes in early November. Filing
• American Express filed its Q3 earnings, though the actual numbers are tucked into the exhibits rather than the 8-K itself. Filing
• Stem drew a NYSE noncompliance notice after its market cap and equity fell under the $50 million floor, and now has 45 days to file a cure plan. Filing
• Transocean and Valaris locked in the merger calendar: shareholder votes on November 24, a Bermuda court hearing two days later, and a close around December 1 at 15.235 Transocean shares per Valaris share. Transocean filing · Valaris filing
• Gorman-Rupp posted a record quarter: $193.3 million in sales, up almost 12%, and net income more than doubled to $22.7 million. Filing

See you tomorrow.

Pradeep Perugu