PATENT DROP: Coinbase’s crypto watchdog
Discover more from PATENT DROP
PATENT DROP: Coinbase’s crypto watchdog
Plus: Meta’s adds to it’s ads; Salesforce skims
Happy Monday and welcome to Patent Drop!
Today, we’re checking out a filing from Coinbase for tech that susses out your crypto requests; Meta’s system to get a better sense of your scrolling habits in real-time; and Salesforce’s AI tool for people who don’t have time to read.
Let's dive in.
Coinbase brings the heat
Coinbase wants to make sure it’s letting in the right people.
The crypto exchange filed a patent application for a risk-analysis system for “cold restore requests” for digital wallets. With a cold restore, a user transfers cryptocurrency from “cold storage,” or a physical hardware wallet kept offline, to “hot storage,” or online storage, such as within a crypto exchange. (Think of it like storing money in a safe at home, versus in a bank.)
This system aims to determine the risk level associated with that transfer. Using a machine learning model, Coinbase’s system comes up with a “risk score” associated with transferring crypto from a cold state to a hot state, analyzing a host of different factors.
First and foremost, the system analyzes a user’s account history, which includes authentication credential usage, frequency of cold restore requests, transfer limits associated with hot wallets, and whether or not they’ve experienced an attack in the past. It also analyzes the crypto or asset that the user is trying to access specifically, such as looking at the volatility of the assets and “sentiment data corresponding to public discussion” around the asset. It also considers time of day and geographical location of the requester.
If the calculated risk score doesn’t satisfy a certain threshold, the system promptly denies the request.
Because crypto is a common target for theft, cold storage wallets provide additional protections that a typical digital wallet doesn’t, making them “significantly safer” for long-term investors, Coinbase noted. But cold restores can be initiated for malicious purposes, such as in the case of customer impersonation or stolen authentication credentials.
Coinbase is the largest crypto exchange in the U.S. by volume with a quarterly trading volume of $145 billion and $130 billion in assets on its platform. When managing that much in assets, cybersecurity is paramount.
Jordan Gutt, Web 3.0 Lead at The Glimpse Group, said that boosting security measures with tech like this could reflect its efforts to “increase user adoption in preparation for the next bull run.”
“Not only does the patent safeguard their users’ crypto assets, but enables Coinbase to build trust with their user base,” Gutt told me.
Coinbase, however, hasn’t always had the best run with security. In February, The 0ktapus hacking group, which had targeted more than 130 tech companies, attacked Coinbase, stealing the login credentials of one of its employees to try accessing its internal systems. The company also revealed a multi-factor authentication breach in October 2021 that impacted 6,000 customers. And in March of this year, a Coinbase user sued the company after a hack caused him to lose “90% of his life savings,” which the company refused to make whole.
While boosting cybersecurity could help keep users’ assets secure and restore user confidence, the patent doesn’t deal with what happens to the collected data after the fact, said Ali Allage, CEO of BlueSteel Cybersecurity. The machine learning model collects a lot of sensitive user data to make its risk analysis determination, Allage said, and that data needs to be protected in any case.
“They're trying to figure out how to protect the end user, and my gut feeling tells me that intentions are good,” said Allage. “I think it's just one of those things where the full picture needs to be thought through.”
Risk analysis using machine learning isn’t exactly uncommon, noted Allage, and plenty of companies use AI fraud detection. This patent may run into obstacles in the approval process unless Coinbase can prove its offering is unique.
Meta’s click tracker
Meta wants to know what posts make you linger and which ones you scroll past.
The company seeks to patent a system for displaying ads based on “recent user engagement signals” and “balancing ad load across surfaces.” Meta’s system tracks what it calls “recent user engagement signals” to determine where to place ads.
Meta noted that these engagement signals track pretty common user behaviors, monitoring what it calls browsing signals, click signals, impression signals and conversion signals (aka, the possibility that your clicks and impressions will lead to you spending money.)
The system then places ads based on those signals. For example, if you look up “shoes” on Instagram’s search bar, you may start to see real-time advertisements for shoes. If you like an influencer’s post tagging a specific dress brand, you may get advertisements for those dresses within an hour. Meta notes that this system places ads in an “explore context,” likely referring to Instagram’s explore page.
Current advertisements are also intermixed with “historically determined advertisements,” or those based on a users’ historical activity.
Additionally, this system also tracks user feed engagement in real time to figure out where to place ads in a video context, aiming to balance “feed and video ads” to not overload the user. Meta said that “since there is uncertainty in knowing what videos will be clicked on by users, it is difficult to pre-calculate where to place ads.”
Meta is no stranger to inventive digital ad tech. The company has sought patents in recent months for a tool to test which content (including ads and brand deals) will go viral, a system which displays ads based on what type of content you consume (i.e. videos versus photos versus stories), and even tech to monitor user engagement with ads in the metaverse.
This focus on ad tech makes sense given that digital ads make up the bulk of Meta’s business. And now that its ad sales are rebounding, this tech may be put to use even more.
In Meta’s earnings report last week, the company reported higher-than-expected revenue and profit for the second quarter as ad sales rebounded from their post-pandemic slump. Ad impressions, a key ad sales metric, rose 34% compared to the same quarter last year.
Those gains stand in contrast to Twitter and Snap, both of which have seen drop offs in ad revenue in recent months. Mark Mahaney, senior managing director at Evercore ISI, told CNBC that after Apple’s iOS advertising privacy change in 2021 threw social media companies for a loop, Meta buckled down to revamp its ad tools in a way other companies didn’t.
Now that it’s ad work is finally paying off, the company may have the freedom to focus attention on its more outlandish ventures, such as figuring out how to make money from the metaverse, which is still costing the company a pretty penny — $3.74 billion in the most recent quarter, to be exact.
Salesforce skims the pages
Salesforce wants to make skimming much easier.
The company is seeking to patent an AI-based “collaborative reading assistance tool.” Salesforce’s tool breaks down documents in a way that’s hyper-specific, allowing a user to choose exactly how much time they want to spend reading or what they want answered.
The system comes with a time filter or a question filter. With the time filter, a user picks exactly how much time they want to spend reading the document, and the system highlights “summative portions of the document” that will take around that time to read. The reading rate is determined either from average user reading rate or based on the individual user.
The question filter gives the user a list of questions that are generated by an AI model based on the document’s content, and the system highlights portions of the document answering the user’s chosen questions specifically.
This system also offers what’s called “focus mode,” which presents one paragraph at a time by “blurring, dimming, or otherwise obscuring the other paragraphs.” This mode also comes with the “ability to highlight, take notes, and answer reflection questions” that are also generated by an AI model.
Finally, the system tracks user activity such as notes, highlights or “dwell time,” or how long a user spends reading each paragraph, to determine which sections of the document are the most important to the reader, and generates a “human-AI summary” based on those metrics.
This isn’t the first time Salesforce has taken an interest in making reading easier. In 2017 – before every headline, product development and startup pitch seemed to encompass AI – Salesforce researchers developed a machine learning algorithm which automatically summarized text by picking out keywords.
The company also is working on a system called “CTRLsum,” which takes unique user preferences into consideration when using AI to generate summaries. The tech in this patent seemingly adds to this work, accounting for both explicit and implicit user preferences to create highly tailored summaries.
Given that Salesforce owns enterprise communication staple Slack, which it acquired two years ago for more than $27 billion, it adds up that the company is beefing up its productivity tech. At the Salesforce World Tour event in May, the company announced SlackGPT, a suite of AI productivity tools that will soon be integrated into the messenger.
When released to the public, the product will add summarization, workflow automation and integration of EinsteinGPT, the company’s enterprise-focused generative AI chatbot it announced in March amid the chatbot boom. Slack also is reportedly looking to hire back staff to work on generative AI tools less than six months after layoffs, Fortune reported in late June.
All this said, the market for this kind of tech definitely isn’t slim. Plenty of startups are pushing out AI summarization tools that claim to be the most efficient or accurate on the market, and many people defer to widely used chatbots like ChatGPT to sum up their work. While Salesforce’s system is likely more customizable than others, and the company does have the advantage of a built-in customer base, it may still have some competition in this arena.
Extra Drops
A few more before you go.
Roku wants to make sure your food gets here right on time. The company is seeking to patent a system for processing food orders based on “timing of media-content presentation.”
Are you still watching? Google wants to know. The company is seeking to patent a system for changing “TV mode” in accordance with viewership. Essentially, while you watch the TV, the TV watches you, too!
If that 10-K filing is a little too complicated, JPMorgan Chase has something for that. The company is seeking to patent a system for “understanding financial documents” using an AI algorithm to break data down into “knowledge graphs” conveying relevant information.
What else is new?
Amazon says it’s made major progress in its one-day and same-day delivery goals, delivering 1.8 billion units to Prime members in that time frame so far this year amid shifts in its fulfillment network.
The iPhone 15 is reportedly going to feature some big changes, including USB-C charging compatibility and titanium edges instead of stainless steel.
Apple has finally greenlit Twitter’s rebrand to X on its app store, a process that took so long because Apple generally doesn’t allow apps to be named just a single character.
Have any comments, tips or suggestions? Drop us a line! Email at admin@patentdrop.xyz or shoot us a DM on Twitter @patentdrop.
 | A guest post by
|