November 10, 2026

BerLabs — November 10, 2026

New Democratic Congress will likely strip the gilding from Trump's Golden Dome

Ars Technica reports that two leading congressional supporters of Golden Dome are leaving as Democrats prepare to take control of Congress. Military officials have also signaled a possible retreat from intercepting missiles shortly after launch. Ars expects funding limits and efforts to remove controversial elements, though those changes remain prospective.

Why it matters
A change in budget oversight could reshape the missile-defense program's technical ambitions and funding.

Read the source →

Instagram’s AI age detection is expanding globally

Meta is expanding Instagram's age-estimation system globally, The Verge reports. It uses signals including interactions, birthday messages, and visual cues in photos and videos to identify users who may have misstated their age. Suspected under-13 accounts are removed, while teenagers are placed into more restrictive accounts; Meta says millions more teens will receive those settings.

Why it matters
Inferred age will increasingly determine users' privacy settings, messaging access, and ability to stay on Instagram.

Read the source →

Here’s the full text of Jensen Huang’s memo to Nvidia

The Verge reports that Nvidia's board has opened an independent investigation involving CEO Jensen Huang's relationship with an AI executive. In a memo reproduced by the publication, Huang acknowledges an affair and says he will cooperate with the investigation into national-security concerns raised in news reports. The memo provides no findings about those concerns.

Why it matters
An investigation involving Nvidia's chief executive raises questions about leadership and protection of sensitive company information.

Read the source →

Your Experts Have Partners. There's HOPE for Keeping Them Together

Amazon researchers describe HOPE, a released tool that removes model experts while accounting for interactions between them. In tests across three models, two calibration datasets, and coding and general benchmarks, the authors report stronger average rankings than four pruning baselines. One Qwen model retained roughly its unpruned coding score after half its experts were removed.

Why it matters
For teams hosting these models, preserving cooperating experts could make memory reduction less damaging to task performance.

Read the source →

Datacenters, Elon Musk, and AI are all on the chopping block once Congress changes hands in 2027

The Register's Kettle panel argues that opposition to datacenters channels broader frustration with technology companies. The Verge's Elizabeth Lopatto points to secretive development deals and limited public input as reasons communities resist projects. The discussion considers how incoming Democratic lawmakers could turn that frustration into investigations and tighter scrutiny.

Why it matters
Local consent and political scrutiny could become bigger constraints on infrastructure expansion.

Read the source →

Hide your slop PR descriptions behind <details><summary>

Simon Willison recommends putting a human-curated explanation at the top of agent-written pull requests, then collapsing the longer description underneath. GitHub supports expandable details in its Markdown, so reviewers can get the purpose quickly while retaining access to implementation and verification notes. He labels the expanded section with the model that wrote it.

Why it matters
A small formatting habit can preserve useful documentation while reducing the reading burden on reviewers.

Read the source →

parse_number() as a SQLite extension: C vs Rust

Simon Willison used Claude to prototype SQLite extensions that convert formatted numeric strings into sortable, summable values. His initial Rust versions were faster but much larger than the C extension. After replacing C's standard parser with a custom decimal parser, the reported C implementation beat both Rust versions across three tested data mixes.

Why it matters
The experiment offers a concrete reminder to examine parsing algorithms before drawing conclusions about language performance.

Read the source →