Some feeds I want to read every single story. Others I’m happy to skim once a week. And a few high-volume feeds I only check occasionally, so stories older than a day or two aren’t worth catching up on. Premium Archive subscribers get the site-wide “days of unread” setting, but it was too blunt, applying the same rule to everything. Now you can set how long stories stay unread on a per-feed and per-folder basis.

How it works

Open the feed options popover (click the gear icon in the feed header) and you’ll see a new “Auto Mark as Read” section. Choose how many days stories should remain unread before NewsBlur automatically marks them as read:

The slider goes from 1 day to 365 days, with a “never” zone at the far right for feeds where you truly want to read every story regardless of age. Choose “Default” to inherit from the parent folder or site-wide setting, “Days” to set a specific duration, or “Never” to disable auto-marking entirely.

Folder inheritance

Settings cascade down from folders to feeds. Set a folder to 7 days, and all feeds inside inherit that setting unless they have their own override. This is perfect for organizing feeds by how aggressively you want to age them out:

• Must Read folder: Set to “Never” so nothing ages out
• News folder: Set to 2 days since news gets stale fast
• Blogs folder: Set to 30 days for long-form content worth revisiting
• Individual feeds can still override their folder’s setting

The status text below the slider shows where the current setting comes from: the site-wide preference, a parent folder, or an explicit setting on this feed.

Site settings dialog

You can also configure auto-mark-read from the site settings dialog (right-click a feed and choose “Site settings”). The same controls are available there, redesigned to match the popover style.

Availability

Per-feed and per-folder auto-mark-as-read settings are a Premium Archive feature, available now on the web. They work alongside the existing site-wide “days of unread” preference in Manage → Preferences → General → Days of unreads, which is also a Premium Archive feature.

If you have feedback or ideas for improvements, please share them on the NewsBlur forum.

The Intelligence Trainer is one of NewsBlur’s most powerful features. It lets you train on authors, tags, titles, and text to automatically sort stories into Focus, Unread, or Hidden. But until now, there were limits—you couldn’t train on URLs, regex support was something power users had been requesting for years, and managing hundreds of classifiers meant clicking through feeds one by one.

Today I’m launching three major improvements: URL classifiers, regex mode for power users, and a completely redesigned Manage Training tab.

Train on URLs

You can now train on story permalink URLs, not just titles and content. This opens up new filtering possibilities based on URL patterns.

The URL classifier matches against the full story permalink. Some use cases:

• Filter by URL path: Like or dislike stories that contain /sponsored/ or /opinion/ in their URL
• Domain sections: Match specific subdomains or URL segments that indicate content types
• Landing pages vs articles: Some feeds include both—filter by URL structure to show only what you want

URL classifiers support both exact phrase matching and regex mode. The exact phrase match is available to Premium subscribers, while regex mode requires Premium Pro.

When a URL classifier matches, you’ll see the matched portion highlighted directly in the story header, so you always know why a story was filtered.

Regex matching for power users

For years, the text classifier only supported exact phrase matching. If you wanted to match “iPhone” and “iPad” you needed two separate classifiers. Now you can use regex patterns in the Title, Text, and URL classifiers.

A segmented control lets you switch between “Exact phrase” and “Regex” mode. In regex mode, you get access to the full power of regular expressions:

• Word boundaries (\b): Match \bapple\b to find “apple” but not “pineapple”
• Alternation (|): Match iPhone|iPad|Mac in a single classifier
• Optional characters (?): Match colou?r to find both “color” and “colour”
• Anchors (^ and $): Match patterns at the start or end of text
• Character classes: Match [0-9]+ for any number sequence

A built-in help popover explains regex syntax with practical examples. The trainer validates your regex in real-time and shows helpful error messages if the pattern is invalid.

Regex matching is case-insensitive, so apple matches “Apple”, “APPLE”, and “apple”. This mode is available to Premium Pro subscribers.

Manage all your training in one place

Over the years you may have trained NewsBlur on hundreds of authors, tags, and titles across dozens of feeds. But when you wanted to review what you’d trained, you had to open each feed’s trainer individually and click through them one by one.

The new Manage Training tab provides a consolidated view of every classifier you’ve ever trained, organized by folder. You can see everything at a glance, edit inline, and save changes across multiple feeds in a single click.

Open the Intelligence Trainer from the sidebar menu (or press the t key). You’ll now see two tabs at the top: “Site by Site” and “Manage Training”. The Manage Training tab is available everywhere you train—from the story trainer, feed trainer, or the main Intelligence Trainer dialog.

The Site by Site tab is the existing trainer you know—it walks you through each feed showing authors, tags, and titles you can train. That’s still the best way to train new feeds with lots of suggestions.

The Manage Training tab shows only what you’ve already trained. Every thumbs up and thumbs down you’ve ever given, organized by folder just like your feed list. Each feed shows its trained classifiers as pills you can click to toggle.

Filtering made easy

The real power comes from the filtering options. At the top of the tab you’ll find several ways to narrow down your training:

Folder/Site dropdown — Only folders and sites with training appear in this dropdown. Select a folder to see all training within it, or select a specific site to focus on just that feed’s classifiers. This is especially useful when you have hundreds of trained items and want to review just one area.

Instant search — Type in the search box and results filter as you type. Search matches against classifier names, feed titles, and folder names. Looking for everything you’ve trained about “apple”? Just type it and see all matches instantly.

Likes and Dislikes — Toggle between All, Likes only, or Dislikes only. Want to see everything you’ve marked as disliked? One click shows you all the red thumbs-down items across your entire training history.

Type filters — Filter by classifier type: Title, Author, Tag, Text, URL, or Site. These are multi-select, so you can show just Authors and Tags while hiding everything else. Perfect for when you want to audit just the authors you’ve trained across all your feeds.

Edit inline and save in bulk

Click any classifier pill to toggle it between like, dislike, and neutral. The Save button shows exactly how many changes you’ve made, so you always know what’s pending. Made a mistake? Just click again to undo—the count updates automatically.

When you click Save, all your changes across all feeds are saved in a single request. No more clicking through feeds one at a time to clean up old training.

Subscription tiers

All three features are available now on the web. If you have feedback or ideas for improvements, please share them on the NewsBlur forum.

Big Sur is one of the most beautiful and iconic parts of California — Highway 1 curves along the Big Sur coastline, featuring stunning ocean views from the dramatic bluffs. Unfortunately, the road is prone to landslides, and a major landslide in February 2024 closed a key portion of the highway through Big Sur. California Transit has announced that the iconic highway is officially open again, ahead of schedule. The road’s initial reopening date was March 30, which makes the repairs 90 days ahead of schedule.

“This reopening will bring much-needed relief to small businesses and families in Big Sur and the surrounding communities who have shown remarkable resilience and strength. I’m grateful to Caltrans crews for delivering Highway 1 ahead of schedule—deploying remotely controlled heavy equipment, stabilizing slopes with thousands of steel reinforcements drilled up to 60 feet deep, and using cutting-edge monitoring systems to overcome extraordinary geological challenges while keeping workers safe. This vital corridor is the gateway to California’s coast and the lifeblood of the Big Sur economy—and today it’s restored.” Said Governor Gavin Newsom regarding the reopening. 

The 2024 landslide that caused the closure — called Regent’s Landslide — resulted in a 6.8-mile closure on the road. Highway 1 along Big Sur is very prone to landslides, and 75 miles of the road is one of the most landslide-prone areas of the Western United States, with 1500 landslides on record. Repairing the road was difficult for crews, requiring daily monitoring from drones and remotely controlled machinery.

Big Sur is a popular day trip from San Francisco, offering iconic coastal views and many delightful stops along the drive. There are beautiful hikes, like a hidden valley filled with calla lilies, and many lovely hotels. We also recommend stopping by the historic restaurant Nepenthe, which features a deck with an incredible view of the coast.

You can learn more about the reopening here.

The post Highway 1 In Big Sur Is Officially Open After 2024 Landslide — Here’s What To Know appeared first on Secret San Francisco.

I have a lot of folders. Over the years I’ve organized my feeds into categories like News, Tech, Cooking, and Comics. But when I’m scanning my feed list, they all look the same—just folder icons with text. I wanted a way to make certain folders stand out at a glance, especially the ones I check most often.

That’s why I built custom icons for both folders and feeds. You can now personalize any folder or feed with an emoji, a preset icon in any color, or even upload your own image.

How it works

Right-click on any folder or feed in your feed list and select “Folder settings” or “Site settings”. You’ll see a new “Folder Icon” or “Feed Icon” tab where you can customize the icon.

There are three ways to set a custom icon:

Preset icons: Pick from over 240 icons (a mix of outline and filled styles) and colorize them with any of 84 colors organized by hue. Want a red heart for your favorites folder? A blue code bracket for programming feeds? It’s all there.

Emoji: Choose from 180 emojis organized by category. A basketball for sports feeds, a fork and knife for cooking, a newspaper for news—you get the idea.

Upload your own: Have a specific image in mind? Upload any image and it will be automatically resized to fit perfectly in your feed list.

Great for feeds without icons

Many feeds don’t have favicons, or they have generic RSS icons that all look the same. Custom feed icons let you give these feeds distinctive icons so you can spot them instantly. I’ve been using this to add icons to older blogs and newsletters that never bothered setting up a proper favicon.

Custom icons are available now on the web for all NewsBlur users. Folders and feeds both support the same icon options of emoji, preset icons with colors, or uploaded images.

If you have feedback or ideas for additional icon options, please share them on the NewsBlur forum.

Sam Rose explains how LLMs work with a visual essay

Sam joined ngrok in September as a developer educator. Here's his first big visual explainer for them, ostensibly about how prompt caching works but it quickly expands to cover tokenization, embeddings, and the basics of the transformer architecture.

The result is one of the clearest and most accessible introductions to LLM internals I've seen anywhere.

Tags: ai, explorables, generative-ai, llms, sam-rose, tokenization