114 stories
·
0 followers

Firefox 147 Will Support The XDG Base Directory Specification

1 Comment
A 21 year old bug report requesting support of the XDG Base Directory specification is finally being addressed by Firefox. The Firefox 147 release should respect this XDG specification around where files should be positioned within Linux users' home directory...
Read the whole story
motang
9 hours ago
reply
After 21 years this bug is finally being addressed!
Share this story
Delete

Some of our favorite gifts will cost you less than $25

1 Share

Holiday shopping on a budget can feel constricting, especially if you've been invited to a white elephant or need a last-minute stocking stuffer. The good news is that, with this guide, you can leave that stress behind.

Whether you're aiming for something practical or fun, we've found plenty of great gifts for under $25 - heck, we found several under $10 - that don't scream "I bought this at the gas station on my way over." You might not expect to find a worthwhile Nintendo Switch / iPad-compatible controller, a 4K-ready streaming stick, or a waterproof speaker in this guide, but here we are. And that's just the start.

8BitDo Micro contro …

Read the full story at The Verge.

Read the whole story
motang
3 days ago
reply
Share this story
Delete

Under the hood: How Firefox suggests tab groups with local AI

1 Share
Browser popup showing the “Create tab group” menu with color options and AI tab suggestions button.

Background

Mozilla launched Tab Grouping in early 2025, allowing tabs to be arranged and grouped with persistent labels. It was the most requested feature in the history of Mozilla Connect. While tab grouping provides a great way to manage tabs and reduce tab overload, it can be a challenge to locate which tabs to group when you have many open.

We sought to improve the workflows by providing an AI tab grouping feature that enables two key capabilities:

  • Suggesting a title for a tab group when it is created by the user.
  • Suggesting tabs from the current window to be added to a tab group.

Of course, we wanted this to work without you needing to send any data of yours to Mozilla, so we used our local Firefox AI runtime and built an efficient model that delivers the features entirely on your own device. The feature is opt-in and downloads two small ML models when the user clicks to run it the first time.

Group title suggestion

Understanding the problem

Suggesting titles for grouped tabs is a challenge because it is hard to understand user intent when tabs are first grouped. Based on our interviews when we started the project, we found that while tab groups are sometimes generic terms like ‘Shopping’ or ‘Travel’, over half the time users’ tabs were specific terms such as name of a video game, friend or town. We also found tab names to be extremely short – 1 or 2 words.

Diagram showing Firefox tab information processed by a generative AI model to label topics like Boston Travel

Generating a digest of the group

To address these challenges, we adopt a hybrid methodology that combines a modified TF-IDF–based textual analysis with keyword extraction. We identify terms that are statistically distinctive to the titles of pages within a tab group compared to those outside it. The three most prominent keywords, along with the full titles of three randomly selected pages, are then combined to produce a concise digest representing the group, which is used as input for the subsequent stage of processing using a language model.

Generating the label

The digest string is used as an input to a generative model that returns the final label. We used a T5 based encoder-decoder model (flan-t5-base) that was fine tuned on over 10,000 example situations and labels.  

One of the key challenges in developing the model was generating the training data samples to tune the model without any user data. To do this, we defined a set of user archetypes and used an LLM API (OpenAI GPT-4) to create sample pages for a user performing various tasks. This was augmented by real page titles from the publicly available common crawl dataset. We then used the LLM to suggest short titles for those use cases. The process was first done at a small scale of several hundred group names. These were manually corrected and curated, adjusting for brevity and consistency. As the process scaled up, the initial 300 group names were used as examples passed to the LLM so that the additional examples created would meet those standards.  

Shrinking things down

We need to get the model small enough to run on most computers. Once the initial model was trained, it was sampled to a smaller model using a process known as knowledge distillation. For distillation, we tuned a t5-efficient-tiny model from the token probability outputs of our teacher flan-t5-base model.  Midway through the distillation process we also removed two encoder transformer layers and two decoder layers to further reduce the number of parameters.

Finally, the model parameters were quantized from floating point (4 bytes per parameter) to integer 8 bit. In the end this entire reduction process reduced the model from 1GB to 57 MB, with only a modest reduction in accuracy. 

Suggesting tabs 

Understanding the problem

For tab suggestions, we identified a couple of approaches on how people prefer grouping their tabs. Some people prefer grouping by domain to easily access all documents for work for instance. Others might prefer grouping all their tabs together when they are planning a trip. Others still might prefer separating their “work” and “personal” tabs.

Our initial approach on suggesting tabs was based on semantic similarity. Tabs that are topically similar are suggested.

Browser pop-up suggesting related tabs for a Boston trip using AI-based grouping

Identifying topically similar tabs

We first convert tab titles to a feature vector locally using a MiniLM embedding model. Embedding models are trained so that similar content produces vectors that are close together in embedding space. Using a similarity measure such as cosine similarity, we’re able to assign how closely similar a tab title or url is to another.

The similarity score between an anchor tab chosen by the user and another tab is a linear combination of the candidate tab with the group title (if present) of the anchor tab, the anchor tab title and the anchor url. Using these values, we generate a similarity probability and tabs that have a high probability threshold are suggested to be part of the group.

Mathematical formula showing conditional probability using weighted similarity and sigmoid function

where,
w is the weight,
t_i is the candidate tab,
t_a is the anchor tab,
g_a is the anchor group title,
u_i is the candidate url
u_a is the anchor url, and,
σ is the sigmoid function

Optimizing the weights

In order to find the weights, we framed the problem as a classification task, where we calculate the precision and recall based on the tabs that were correctly classified given an anchor tab. We used synthetic data generated by OpenAI based on the user archetypes above.

We initially used a clustering approach to establish a baseline and switched to a logistic regression when we realized that treating the group, title and url features with varying importances improved our metrics.

Bar chart comparing DBScan and Logistic Regression by precision, recall, and F1 performance metrics

Using logistic regression, there was an 18% improvement against the baseline.

Performance

While the median number of tabs for people using the feature is relatively small (~25), there are some “power” users whose tab count reaches the thousands. This would cause the tab grouping feature to take uncomfortably long. 

This was part of the reason why we switched from a clustering based approach to a linear model. 

Using our performance framework, we found that the p99 of running logistic regression compared to a clustering based method such as KMeans improved by 33%.

Bar chart comparing KMeans and Logistic Regression using percentile metrics p50, p95, and p99

Future work here would involve improving F1 score. These could be by adding a time-related component as part of the inference (we are more likely to group tabs together that we’ve opened at the same time) or using a fine-tuned embedding model for our use case.

Thanks for reading

All of our work is open source. If you are a developer feel free to peruse our source code on our model training, or view our topic model on Huggingface.

Feel free to try the feature and let us know what you think!

Take control of your internet

Download Firefox

The post Under the hood: How Firefox suggests tab groups with local AI appeared first on The Mozilla Blog.

Read the whole story
motang
3 days ago
reply
Share this story
Delete

How Private Equity Killed the Media Industry - TPM – Talking Points Memo

2 Shares

I learned many surprising lessons from my 20 months as editor-in-chief of Deadspin, the skeptical, irreverent, hilarious, trailblazing sports outlet that entertained, offended, and educated audiences in roughly equal measure. 

I learned from a cease-and-desist letter that Jacuzzi is a trademarked brand, and that the hotel room in which a world-famous soccer star was alleged to have raped a woman contained a mere “spa” or “hot tub.” I learned from inhaling Chartbeat that our very dumbest stories and our very smartest stories would always be our biggest traffic drivers. I learned from our general counsel more than I ever wanted to know about the precise limits of fair use. I learned from my coworkers — all of them brilliant and entirely deranged — that there is no limit to how hard I can laugh in a soul-suckingly bland Times Square cubicle farm. Even knowing how it all ended, I’d still take the job 100 times out of 100.

The most consequential lessons I learned, though, were about the ways in which I had misunderstood “free market” capitalism, and about what that meant for the industry that gave me my career. Those are the lessons I haven’t stopped agonizing over six years later, the ones that led to my first book but also caused scores of sleepless nights. 

Until Deadspin, I naively believed that a company making money required the company to sell goods or services. Boosting profits, under this line of thinking, requires either selling more of the company’s existing goods or services, or finding new goods or services to sell. This misunderstanding made me genuinely optimistic when Deadspin and our sister sites were purchased by a private equity firm and renamed G/O Media in 2019. The numbers my bosses had shown me indicated that Deadspin was profitable, even if our parent company was not. I knew we could be more profitable if we made basic moves like developing a subscription program, but our previous owner — the Spanish-language broadcaster Univision — never seemed to care enough to put in the time. 

Our new bosses at Great Hill Partners promised real financial expertise: Its executives told me on their first day in charge that they agreed we needed a subscription program. As far as I could tell, things were looking up.

Had I paid better attention to what was happening in the rest of the media industry, my read on the situation would have been far less rosy. I did know that Alden Global Capital had been devouring and decimating local newspapers; just a year earlier, The Denver Post’s editorial board had published a package of articles begging for their newspaper to be rescued from the hedge fund’s clutches. Yet even as I rooted for my colleagues in Colorado to escape, I accepted the conventional wisdom that the root problem was newspapers not adapting fast enough to the digital age, that Alden was a just vulture feasting on the scraps. Embarrassingly, I thought that being digital-only, that being profitable, that being cool would protect us.

Being cool didn’t protect Deadspin, as has been thoroughly chronicled by my colleagues and me. I quit my job three months after the acquisition. The rest of the staff followed me out the door two months after that. The site sat dormant for months, then went Weekend at Bernie’s mode for a couple of years, during which its most notable story was one that called a Native American 9 year old racist for wearing a headdress at a Kansas City Chiefs game, leading to a defamation suit that is still ongoing. In 2024, Deadspin was sold to a Maltese company and became a referral site for online casinos. This July, G/O Media began “working towards a full wind down,” in the words of CEO Jim Spanfeller. One of the most influential companies in the history of digital media was finally, mercifully dead.

Being cool didn’t protect Vice News either. Once the swashbuckling rogue of digital media, the site was destroyed by a combination of its madman founder and its greedy private-equity investors-turned-owners. When Vice News stopped publishing in February 2024 — nearly eight years after Gawker’s demise, five after OG Deadspin’s —  it marked the final nail in the coffin of the era in which any media outlet was thought of as cool. On one level, that’s for the best; I can think of exactly one Deadspin employee in the site’s history who could accurately be categorized that way. But it also makes clear just how much private equity has taken from us: not just local newspapers providing invaluable information about communities, but also blogs willing to get weird, to try things no one else would. 

After several years of reporting on and obsessing over how private equity works and why, I finally understand the root of my misconceptions about capitalism. I had thought that the point of buying a beloved, profitable publication was to make it more profitable, to strengthen the fundamentals of its business model in hopes of a lucrative exit years down the road. 

That is not the point of buying a beloved, profitable publication (or any business). The point is to make the private equity firm more profitable. The Denver Post and Deadspin and Vice News are just widgets, endlessly interchangeable in the service of maximizing shareholder value. Only chumps make money by selling goods or services these days; the real geniuses rely on management fees, deal fees, dividend recapitalizations, real estate deals, and the like. That allows — requires! — a private equity firm to divorce its incentives from that of its own portfolio company, making it, at best, agnostic to whether the company lives or dies. In many cases, the best decision for the firm is the one that directly undermines the company it controls. The reason there are no weird blogs anymore is that it’s more fruitful to drive them out of business.

When G/O Media began “working towards a full wind down” this summer, Spanfeller wrote a 2,300-word “epilogue” to the company’s existence, the main thrust of which seemed to be to demonstrate that coherent writing skills are not a prerequisite for running a media company. (Sample sentence: “At one and the same time the two are clearly linked and yet can also be goals at cross purposes.”) He barely mentioned the journalism produced by the sites he oversaw for six years except to criticize it for being biased; his few references to the people who worked for him were mostly dedicated to railing against unions. Yet he couldn’t resist bragging about what a success he’d been, despite all those mean bloggers and bargaining committees plotting against him. 

Every one of the eight sites Spanfeller had taken over six years earlier had either gone out of business or been reduced to a shell of its former self, but that didn’t matter. The key point came seven paragraphs in:

“We will exit having increased shareholder value.”

Read the whole story
motang
8 days ago
reply
Share this story
Delete

Steam store pages get a mini makeover to better suit wide screens

1 Share

Store pages on Steam are looking a lot less cramped thanks to a new update. Pages have been made wider, with support for higher resolution images and new viewing modes for trailers and screenshots. You'll notice changes in the top carousel and in the "About the Game" section, where some new formatting options should make things look a bit more organized. The update just rolled out to the public after first being tested among beta users.

With this update, pages have been widened to 1200 pixels, which Valve says "felt like a good balance where we can show more content on screen without overwhelming the page and making it hard to navigate." There's now the option of a large pop-up view called theater mode in the carousel, as well as full-screen mode. In addition to games' store pages, Valve has slightly tweaked the appearance of search results and recommendation pages to be wider, and made store hubs, Steam Charts and the News Hub look more uniform.

You may also notice some more colorful backgrounds on games' store pages and in bundle detail pages. Where you won't see changes yet, though, is the homepage. While Valve says it's working on "similar adjustments" for the homepage, those aren't rolling out with this update.

This article originally appeared on Engadget at https://www.engadget.com/gaming/pc/steam-store-pages-get-a-mini-makeover-to-better-suit-wide-screens-200142506.html?src=rss

Read the whole story
motang
9 days ago
reply
Share this story
Delete

ClickFix may be the biggest security threat your family has never heard of

1 Share

Over the past year, scammers have ramped up a new way to infect the computers of unsuspecting people. The increasingly common method, which many potential targets have yet to learn of, is quick, bypasses most endpoint protections, and works against both macOS and Windows users.

ClickFix often starts with an email sent from a hotel that the target has a pending registration with and references the correct registration information. In other cases, ClickFix attacks begin with a WhatsApp message. In still other cases, the user receives the URL at the top of Google results for a search query. Once the mark accesses the malicious site referenced, it presents a CAPTCHA challenge or other pretext requiring user confirmation. The user receives an instruction to copy a string of text, open a terminal window, paste it in, and press Enter.

One line is all it takes

Once entered, the string of text causes the PC or Mac to surreptitiously visit a scammer-controlled server and download malware. Then, the machine automatically installs it—all with no indication to the target. With that, users are infected, usually with credential-stealing malware. Security firms say ClickFix campaigns have run rampant. The lack of awareness of the technique, combined with the links also coming from known addresses or in search results, and the ability to bypass some endpoint protections are all factors driving the growth.

Read full article

Comments



Read the whole story
motang
9 days ago
reply
Share this story
Delete
Next Page of Stories