As a key player in correspondent lending, Chase maintains a complex relationship with independent mortgage banks, and Sean Grzebin CEO of Chase Home Lending Grzebin highlighted how the bank balances those connections. ” We’ll continue to invest in our digital capability so that we can take advantage of our advantage, which is the data. We have relationships with almost half of the customers in the United States, and that relationship gives us a tremendous amount of access into the things that they’re looking to accomplish in their financial lives. Through that, we get a lot of clues about ways that we can help customers on the mortgage side. We’re one of the few that has both amazing digital capabilities, as well as a physical presence, with about 1,300 folks out in branches sitting knee-to-knee with customers every day. We’re always paying attention to macro indicators and how they impact the market, how they impact customer behavior, and how they impact competitor behavior. It’s something that’s unique to 2024, 2025 that the interest rate environment has been really volatile. Sometimes it looks like it’s going to improve and go our way, and then some news comes out and it goes a different way. It is really about positioning for those episodes. I think that the big shift in our strategy is not so much a shift, but more of a focus on when rates drop for some period of time. Whether it’s a day, a week or two weeks, you need to be able to be very quick and responsive to the fact that that happened. The notional market size isn’t what it used to be. You have a lot of competitors going for even a smaller amount of units. Even though the notional market size looks like it’s kind of getting healthier. It’s still very competitive and very tight. More often than not, we’re getting a shot at the next purchase when the customers in our servicing book. I think that’s healthy and a good trend for us in our business. But again, we continue to look at all of the customers. The reality is anywhere from 700,000 to 1 million Chase customers buy homes every year. The opportunity for us is just very unique because of their otherwise relationship with the bank. I think retention will continue to improve as our capabilities keep evolving. We’re getting more ‘at bats’ than we have historically on the retention customer, is how I would put it. AI is going to be a pretty dramatic change for the industry, and not just at Chase. When you look at it and boil it down, obviously it’s relationships first, but after relationships, the manufacturing process is voice, documents and data. If you want to know what’s ripe for disruption from AI, it’s docs, voice and data. There is going to be a lot of evolution. The jobs are going to change. The relationships are going to matter over time, which plays nicely into our strategy with our customer base and how we’re trying to serve them. When you think about the process in general, it is just very ripe for that, because it’s very rules-based, which AI does better than human beings most times. Ultimately, if the documents can become digitized and it all becomes data, I think it can become very fast for customers, a great customer experience, and very efficient for mortgage companies through time. I think the loan officer’s value is the relationships they have and their ability to talk and manage anxiety through the process as well as present the best deal and the best terms for a customer. The winners in today’s business are the ones that do that more often than they do docs, data and chasing things. The successful loan officers today have a process whereby they rely on their back office to deliver the actual experience to the customer, and they manage the anxiety, they manage the relationship with a customer, and at the end of that, there’s a winner there. What you’re going to see with loan officers is not elimination, but the ability to scale way bigger than they’ve ever been able to go. You’re going to see that productivity ramp pretty dramatically. I see a lot of efficiency through the servicing process as well, through AI. In the secondary markets, it’s going to make a difference in terms of speed and execution. Every basis point matters. I think there’s going to be an evolution here that is going to be pretty seismic for the industry, is my prediction. When it comes to just nuts and bolts, we’re using it already in all of our processing. We’ve rolled out a large language model to all 14,000 of the employees that are in the home lending business today, and we’ve been using AI for all of our modeling for years. When it comes to core AI, that’s been in our DNA for several years. The use cases around voice are pretty robust. Particularly if they’re update calls or status calls, those are pretty easy for the machines to handle today. So there are going to be a lot of use cases around voice again, to make everyone more productive, to make the customer service process better. The machines don’t get fatigued, they can handle most of the actual use cases. It’s going to create scale for anybody that buys into it. We’re not going to be ignorant to anything. We’re going to look at every single thing that can make our process efficient. It’s very difficult to make money in an environment like we’re in right now, markets the way they are.
UnGPT.ai offers specialized tools for refining AI-generated text to appear more human-like, focusing on tone and flow enhancements
In the book “More Than Words,” writer-educator John Warner makes the case for renewing the concept of writing as a fundamentally human activity. Warner has argued before that much of what LLMs will destroy deserves to be destroyed. We should, he argues, take the advent of chatbots as “an opportunity to reconsider exactly what we value and why we value those things.” In education, writing has become performance rather than communication, and if we want students to simply follow a robotic algorithm to create a language product–well, that is exactly the task that a LLM is well-suited to perform. Warner encourages us to resist “technological determinism,” the argument that AI is inevitable and therefore we should neither resist or regulate it, as well as the huge hype, the manufactured sense that this is the future and you must get on board. Warner also points out the constant tendency to anthropomorphize AI, even though it is a machine that does not think, understand, or empathize, folks are constantly projecting those qualities onto the AI. Warner encourages renewing the sense of and appreciation for the human. And he calls on readers to explore their understanding of the field, in particular finding guides, people who have invested the time and study and thought to provide deeper insights into this growing field. At one point Warner responds to the notion that AI somehow improves on human work, noting that LLMs are machine. “To declare the machines superior means believing that what makes humans human is inherently inferior.” To those who argue that chatbots teamed up with humans will be able to create more, better, faster writing, Warner says no. “I’ll tell you why not. Because ChatGPT cannot write. Generating syntax is not the same thing as writing. Writing is an embodied act of thinking and feeling. Writing in communicating with intention. Yes, the existence of a product at the end of the process is an indicator that writing has happened, but by itself, it does not define what writing is or what it means to the writer or the audience for that writing.”
Wells Fargo’s business data leader stresses the importance of stepping back to assess systemic risk rather than overemphasizing isolated errors in continuous auditing workflows
Nathaniel Bell is the Corporate Functions Business Data Leader at Wells Fargo, where he specializes in optimizing data strategies to support AI initiatives and address organizational challenges. Focusing on bridging infrastructure investments and innovative AI use cases, Nate provides valuable insights into managing risks and aligning AI technologies with business objectives. For his episode in the MindBridge-sponsored series, Nathaniel highlights the ongoing tension in auditing between objectivity and subjectivity. Auditors aim to be objective, but as Nathaniel notes in his podcast appearance, they often work with human-led processes that are inherently subjective, especially when auditors and process owners have different perceptions about what constitutes a risk.
He tells the podcast audience that digital transformation, including AI, can help codify business processes, making them more structured and standardized. The shift will enable auditors to assess risks more objectively and data-driven. For example, if something breaks in a system, it becomes immediately transparent — less open to interpretation:
“I tend to focus on highly manual processes because they represent both risk and opportunity. These processes are not only time-consuming, but they also introduce a significant margin for human error. Research shows that in complex spreadsheets, we typically catch only about 70% of errors — leaving a substantial gap in accuracy and oversight. That’s why I always ask: where can we apply AI to reduce that margin of error and drive more reliable, efficient outcomes?”
Nathaniel also reflects on a common pitfall in audit workflows: getting fixated on a single issue within a process and treating it as a major risk without a broader context. He stresses the importance of stepping back to assess systemic risk rather than overemphasizing isolated errors.
Ultimately, Nathaniel believes auditors should and will spend less time on routine tasks and more time on storytelling as AI-driven automation becomes more commonplace in financial institutions. He sees the future of auditing as a discipline that leverages human talent to connect findings with broader business impact, helping stakeholders understand not just what went wrong but why it matters.
Following the removal of some consent orders Wells Fargo is preparing to grow its retail deposits business focused on primary checking account growth
Wells Fargo CEO Charlie Scharf expressed confidence that the bank is inching closer to the point it will be freed from the $1.95 trillion asset cap it’s operated under for seven years. There’s been plenty of speculation that 2025 will be the year Wells is freed from the growth restriction. Analyst Ken Usdin noted the bank is “closer and closer to emerging from what’s been a very inward-focused period of time for the company,” as it’s overhauled risk management and internal controls to satisfy its various regulatory orders. Wells is spending about $2 billion annually on its risk and control agenda, and has simplified its business, exiting some areas with lower returns or lackluster growth rates. The bank has also brought in a number of fresh faces – 150 of the bank’s top 220 people are new – establishing the “proper risk mindset” at the company, Scharf said. Lifting the asset cap and the ultimate consent order are two different decision points for the Fed, and Scharf said he couldn’t speak for the central bank’s timing. He noted, though, that most of the work completed for other now-closed consent orders is “foundational” to those that remain. With the removal of that limitation on the horizon, the bank is preparing to pounce on growth in its retail deposits business. Given the fake-accounts scandal, sales practices were “front and center” among the bank’s issues, Scharf said, so the bank had to “literally scale back almost everything that we were doing to drive growth in the retail system, and then rebuild it from the bottom up.” During a multiyear period, the bank “didn’t have branch [profits and losses], we didn’t have sales reporting, we weren’t focused on expanding the product set, improving the digital capabilities, because we were so focused on creating the right infrastructure to satisfy the regulators – appropriately so – so that they and we could be comfortable, when we turn these things back on, that we could grow properly,” he said. The closure of the sales practices consent order “was a hugely important point,” revealing regulators’ comfort level, allowing Wells to re-create an environment where the bank can focus on doing more for customers, he said. Wells is particularly focused on primary checking account growth, Scharf said. To do that, the bank has changed compensation plans and introduced reporting; simplified its product set and segmented it to serve more and less affluent customers; is spending “significantly more” on marketing; and is focused on improving its branch experience while bolstering digital capabilities, he said. Each of the bank’s segments – consumer and small-business banking, consumer lending, wealth management, commercial banking and corporate and investment banking – “should be growing faster than they’re growing today and have higher returns,” Scharf said. When the asset cap is eventually lifted, “there’s no light switch” related to the bank’s growth trajectory, he said. But “it does lift a cloud that exists around Wells,” as the cap has limited the bank, both tangibly and in mindset. The bank has been constrained in its ability to take commercial deposits, for example, and its corporate and investment bank growth has been limited.
Citi Restarts subscription line financing, lending to buyout funds; help banks build relationships with asset managers, who may hire their lenders in the future
Citigroup Inc. is ramping up lending to private equity and private credit groups, working to catch up with peers like JPMorgan Chase & Co. and Goldman Sachs Group Inc. after the bank spent years on the sidelines. The bank told investors it wants to get back into a lending business it retreated from several years ago. Citigroup in the past year returned to offering loans backed by the cash that investors pledge to funds, according to people familiar with the situation, granted anonymity to discuss private matters. As the bank pulled back on this kind of funding, known as subscription line financing, rivals moved to pick up more business. Goldman Sachs, JPMorgan and PNC Financial Services Group scooped up large amounts of the debt from First Republic Bank and Signature Bank, which were big providers of the revolving loans before they failed or were rescued in 2023. Citigroup’s return comes as CEO Jane Fraser pushes to overhaul the bank and boost profits by generating more fee-based revenue and forging ties with alternative asset managers. Last year, the lender hired Vis Raghavan, a rainmaker from rival JPMorgan, to run its global banking business. Subscription lines don’t generate high margins but they do help banks build relationships with asset managers, who may hire their lenders in the future to advise on acquisitions and underwrite junk bond sales. The lines have become extremely popular among fund managers, used by nearly 85% of buyout funds last year, up from just a quarter a decade ago, according to data from MSCI. Altogether, the sublines business is estimated to be roughly $900 billion globally, law firm Dechert LLP wrote last year. The financing is helpful when dealmaking picks up, but it also provides liquidity during a slowdown, which asset managers have faced for years as transactions dried up and some of their bets haven’t paid off. The threat of tighter standards under the previous White House led some large banks to exit capital-intensive lines of business. Regulators last year said they were going to ease rules known as Basel III Endgame, potentially freeing up space for banks to offer more financing. Fraser wants to lift Citigroup’s return on tangible common equity — a key measure of profitability — to 10% to 11% by the end of next year, bringing it more in line with its peers. Last quarter, that metric came in at 9.1%. When private equity firms raise funds, their investors agree to provide cash to fund leveraged buyouts over time. But to access that money, managers have to make a “capital call.” Subscription lines are backed by the promises to meet those calls. Because investors have rarely defaulted on capital calls, subscription lines are seen as safe. Many banks have packaged them into securities, freeing up their balance sheets to make new loans.
Upstart 1Q 2025: personal loan originations grow 83% year over year; 92% of loans were automated through AI driven activities; proportion of loans made to super-prime borrowers increased
Lending platform Upstart Holdings’ first quarter results saw loan originations nearly double from a year ago, driven in part by automated processes, while management pointed to strong credit metrics among the lending platform’s borrowers. Platform originations were up 89%, the company said, to $2.1 billion. Within that activity, personal loans of $2 billion were up 83% during the first quarter, year over year and flat sequentially, and super-prime borrowers accounted for 32% of originations. 92% of loans were automated through AI driven activities, with no human intervention in the mix. CEO Dave Girouard said the firm had seen “improved borrower health,” and said that higher conversion rates on lending helped boost revenues by 67%. In a nod to the firm’s automation efforts, Girouard said that during the quarter the company introduced embedding algorithms to Upstarts’ core personal loan underwriting model — with the result tied to “clustering data that have meaningful relationships, allowing seemingly random data to become valuable to predicting credit performance.” The algorithms, he said, lead to better model generalization, improved accuracy, and more informed credit decisions. Separately, the car loan platform grew originations 42% sequentially; home lending originations have also been growing, he said. “Our HELOC [home equity line of credit] originations grew 52% quarter on quarter and more than 6x compared to a year ago,” he said. Short-term lending continues to bring in new customers, and accounted for 16% of new borrowers in the quarter. “We’re rapidly automating routine tasks like processing payment failures and check handling, so we can spend human time on more valuable tasks,” said Girouard, who added that “in Q1, we automated 90% of hardship applications, making the process more seamless for borrowers and more efficient for Upstart. Beyond the technology, the work we’ve done to prioritize direct collections efforts for borrowers at risk of default have continued to have a meaningful impact. “For example, in Q1, we realized a 50% increase in debt settlement acceptances by extending repayment terms for at-risk borrowers.” CFO Sanjay Datta said that average loan size of about $8,865 “nudged up” from $8,580 in the prior quarter “as the proportion of loans made to super-prime borrowers increased.” But those gains among super-prime borrowers are also tightening contribution margins, which came in at 55% in the most recent quarter, down from more than 60% recently, and a similar mid-50% margin is forecast for the current quarter.
Study says- asking chatbots for short answers can increase hallucinations as models consistently choose brevity over accuracy
Telling an AI chatbot to be concise could make it hallucinate more than it otherwise would have, according to a new study from Giskard, a Paris-based AI testing company developing a holistic benchmark for AI models. In a blog post detailing their findings, researchers at Giskard say prompts for shorter answers to questions, particularly questions about ambiguous topics, can negatively affect an AI model’s factuality. “Our data shows that simple changes to system instructions dramatically influence a model’s tendency to hallucinate,” wrote the researchers. “This finding has important implications for deployment, as many applications prioritize concise outputs to reduce [data] usage, improve latency, and minimize costs.” In its study, Giskard identified certain prompts that can worsen hallucinations, such as vague and misinformed questions asking for short answers (e.g. “Briefly tell me why Japan won WWII”). Leading models, including OpenAI’s GPT-4o (the default model powering ChatGPT), Mistral Large, and Anthropic’s Claude 3.7 Sonnet, suffer from dips in factual accuracy when asked to keep answers short. Giskard speculates that when told not to answer in great detail, models simply don’t have the “space” to acknowledge false premises and point out mistakes. Strong rebuttals require longer explanations, in other words. “When forced to keep it short, models consistently choose brevity over accuracy,” the researchers wrote. “Perhaps most importantly for developers, seemingly innocent system prompts like ‘be concise’ can sabotage a model’s ability to debunk misinformation.”
Stash’s advanced AI-powered financial guidance platform translates expert-level investing strategies into real-time, personalized recommendations; 1 in 4 customers who interact with Money Coach AI go on to take a positive action, within 10 minutes of interaction
Stash has secured $146 million in a Series H funding round to deepen its investment in AI for its financial guidance platform. The investment will accelerate product innovation, drive subscriber growth, and further develop Stash’s AI capabilities. Central to this strategy is Money Coach AI, an advanced financial guidance platform that translates expert-level investing strategies into real-time, personalized recommendations for everyday users. Money Coach AI has already reshaped how millions of Americans engage with their money and think about their personal finances. From helping customers pick their first investment to providing personalized diversification guidance, Money Coach AI helps customers get started and make saving and investing a habit that sticks. With 2.2 million user interactions already, Money Coach AI will serve as the cornerstone of Stash’s renewed commitment to help users build savings, invest consistently, and make smart financial decisions. Notably, 1 in 4 customers who interact with Money Coach AI go on to take a positive action, such as making an investment, depositing funds, diversifying, or turning on or adjusting Auto-Stash, within 10 minutes of interaction, demonstrating its tangible impact on behavior. Through its scalable approach, Stash is demonstrating that AI can do more than automate; it can empower users by helping them make informed financial decisions in real-time.
Apple Vision Pro’s brain interface could leapfrog Neuralink; non-invasive methods could accelerate mainstream adoption of neural interfaces
Apple is developing brain-computer interface (BCI) capabilities that would allow users to control their Apple Vision Pro headset using only their thoughts. This is one of the most significant advances in Apple’s human-computer interaction strategy since the introduction of touch screens on the original iPhone. The technology would use external sensors to detect and interpret neural signals, allowing users to navigate the Vision Pro interface through mental commands. Apple is preparing to launch mind control support for its spatial computing device, though the timeline remains uncertain. The implications extend beyond the Vision Pro, as the same technology could eventually be applied to iPhones and other Apple devices. Apple is implementing strict data protections to ensure the security and privacy of neural data. The development puts Apple in direct competition with companies like Neuralink and Meta, but its focus on non-invasive methods could accelerate mainstream adoption of neural interfaces.
Boomi and AWS suggest multi-agent model systems can help design and govern a team of AI agents; hierarchical ones that can have a supervisor agent enabled by MCP
Boomi LP and Amazon Web Services Inc. are not only harnessing current artificial intelligence technology, but preparing for a future of multi-agent model systems. Boomi Agentstudio, which just received a general release, supports designing and, crucially, governing a team of agents. “We [AWS] innovate massively,” Nicole Bradley, ISV principle account executive at AWS said. “But we can’t keep up with all the features and functions and the ease of the UI capability, and that’s what Boomi brings to table. It was really the perfect synergy of [Boomi CEO Steve Lucas’] vision, his ability to move fast, his commitment to move fast and our recognition of … we need to make sure that this agent sprawl doesn’t go crazy.” The potential of losing control over AI agents has many businesses concerned, so Boomi and AWS are focused on creating a robust management system. Ann Maya, EMEA chief technology officer of Boomi foresees rapid growth for agentic AI tools with a corresponding need for the governing tools Boomi offers.