Additional Reading

Neurosymbolic AI is a rising technique that aims to solve the hallucination problem. It’s a hybrid approach that marries gen AI’s neural networks for pattern recognition with symbolic reasoning’s logic that can prove whether outputs are correct, with mathematical precision. Amazon has become one of neurosymbolic AI’s most prominent adopters. Its Automated Reasoning Group, built tools to verify security policies in the AWS cloud. Those methods now underpin new systems such as the Vulcan warehouse robots, which combine neural networks for perception with automated reasoning for precise planning. Amazon and Imandra are advancing the approach in parallel, Amazon with warehouse robots and shopping assistants, Imandra with finance, code verification and regulatory compliance. Imandra built a mathematically precise language for specifying the FIX (Financial Information eXchange) protocol. Instead of relying on hundreds of pages of PDFs, clients can now use automated reasoning to ensure precise communication each way. To scale its approach, Imandra launched Imandra Universe, a platform it describes as the first marketplace for neurosymbolic agents. It hosts symbolic reasoning engines specialized in domains from geometry to logistics. A “Reasoner Gateway” lets developers plug them into agent frameworks, augmenting AI systems with logical checks. One flagship tool, Code Logician, targets the flood of AI-generated code. Once installed in coding assistants like Cursor, Code Logician builds a mathematical description of what the AI generated code is doing and use Imandra to verify it. Passmore said around 60% of AI-generated code contains bugs and Code Logician can make 96% of the code correct within three iterations. Imandra is expanding Code Logician beyond Python to Java and even COBOL, reflecting enterprise demand for code migration. It is also developing agents for geometry reasoning, among other pursuits.

Read Article

 

A new study from Arizona State University researchers suggests that the celebrated “Chain-of-Thought” (CoT) reasoning in Large Language Models (LLMs) may be more of a “brittle mirage” than genuine intelligence. Researchers of the new study argue that “a systematic understanding of why and when CoT reasoning fails is still a mystery,” which their study aims to address. As the paper notes, “theoretical and empirical evidence shows that CoT generalizes well only when test inputs share latent structures with training data; otherwise, performance declines sharply.” The ASU researchers propose a new lens to view this problem: CoT isn’t an act of reasoning but a sophisticated form of pattern matching, fundamentally bound by the statistical patterns in its training data. They posit that “CoT’s success stems not from a model’s inherent reasoning capacity, but from its ability to generalize conditionally to out-of-distribution (OOD) test cases that are structurally similar to in-distribution exemplars.” In other words, an LLM is good at applying old patterns to new data that looks similar, but not at solving truly novel problems. Based on their findings, the researchers conclude that CoT reasoning is a “sophisticated form of structured pattern matching, fundamentally bounded by the data distribution seen during training.” When tested even slightly outside this distribution, performance collapses. What looks like structured reasoning is more of a mirage, “emerging from memorized or interpolated patterns in the training data rather than logical inference.” The breakdown was consistent across all three dimensions. On new tasks, models failed to generalize and instead replicated the closest patterns they had seen during training. When faced with reasoning chains of different lengths, they struggled, often trying to artificially add or remove steps to match the length of their training examples. Finally, their performance proved highly sensitive to superficial changes in the prompt, especially variations in core elements and instructions. The researchers offer a direct warning to practitioners, highlighting “the risk of relying on CoT as a plug-and-play solution for reasoning tasks and caution against equating CoT-style output with human thinking.”

Read Article

A new report from MIT has sent shockwaves through the enterprise AI world. According to the State of AI in Business 2025 study, 95% of generative AI pilots deliver zero return on investment. The findings, based on 300 public deployments and more than 150 executive interviews, suggest that billions of dollars have been spent on AI experiments that never scale — and that most organizations are stuck on what MIT researchers call the “GenAI Divide.” The numbers are stark. Forty percent of organizations say they’ve deployed AI tools, but only 5% have managed to integrate them into workflows at scale. Most projects die in pilot purgatory. Meanwhile, headlines are warning of an “AI bubble,” and investors are shorting AI stocks on the idea that generative AI’s big enterprise moment is already stalling out. “Confidently wrong is the problem,” says Tanmai Gopal, co-founder and CEO of PromptQL, a unicorn AI company that counts OpenAI, Airbus, Siemens, and NASA as customers. “If the system is not always accurate even the tiniest percent of the time, I need to know when it’s not. Otherwise, my minutes turn into hours; the ROI disappears.” In his blog post, Being “Confidently Wrong” Is Holding AI Back, Gopal describes what he calls the “verification tax.” “I don’t know when I might get an incorrect response from my AI. So I have to forensically check every response.” This tax explains much of what MIT labeled as the GenAI Divide. Enterprises eagerly launch pilots, but employees end up spending so much time double-checking outputs that the promised efficiencies never materialize. It’s not that generative AI lacks raw horsepower — the models can be dazzling. It’s that their confidence is uncalibrated. In regulated or high-stakes industries, one bad answer can outweigh ten good ones. As Gopal puts it: “For serious work, one high-confidence miss costs more credibility than ten successes earn.” MIT’s researchers framed the same issue differently. They found that most enterprise AI tools don’t retain feedback, adapt to workflows, or improve over time. Without those qualities, they stall. Gopal agrees. “Without high-quality uncertainty information, I don’t know whether a result is wrong because of ambiguity, missing context, stale data, or a model mistake. If I don’t know why it’s wrong, I’m not invested in making it successful.” That insight matters because it reframes the entire conversation. If AI isn’t failing due to lack of capability, but because it hasn’t been designed to communicate its limits and learn from corrections, then the fix is less about building bigger models — and more about building humbler ones. PromptQL has built its entire platform around solving this exact problem — what Gopal calls the difference between being “confidently wrong” and “tentatively right.” Instead of presenting outputs as gospel, PromptQL calibrates confidence at the response level:

Quantifies uncertainty. Every answer comes with a confidence score. If the system is unsure, it abstains — effectively saying “I don’t know.

Surfaces context gaps. Rather than hiding uncertainty, the system flags why an answer may be unreliable: missing data, ambiguity, or lack of context.

Builds an accuracy flywheel. Each abstention or correction becomes training fuel. PromptQL captures those signals, letting the system improve continuously — closing the “learning gap” MIT identified as the number one cause of pilot failure.

Integrates into workflows. Instead of sitting in a chatbox, PromptQL embeds directly into enterprise processes like contracts, engineering, or procurement, so uncertainty flags and corrections appear exactly where the work is happening.

“The starting point of this loop is if an AI system could tell the user when it’s not certain about its accuracy in a concrete and native way,” Gopal writes. That loop — abstain, get corrected, learn — is what he calls the accuracy flywheel. “We don’t need perfection; we need a loop that tightens.” This humility-first approach has led to adoption in some of the most skeptical corners of the enterprise market. While 95% of pilots stall, PromptQL is closing seven- and eight-figure contracts with Fortune 500s, government agencies, and regulated industries — the exact places MIT says AI has struggled to gain traction. The company is living proof that enterprise AI is not failing. The wrong kind of enterprise AI is. As Gopal puts it: “No amount of solving any other problem — integration, data readiness, organizational readiness — will change the fact that AI’s tendency to be confidently wrong keeps it out of real-world use cases.” The takeaway, then, is not that AI is doomed to fail. It’s that enterprises must demand a different kind of AI: one that is transparent about its uncertainty, tightly integrated into workflows, and capable of improving with every interaction. The MIT report is right to highlight the GenAI Divide. But if we only focus on the 95% that failed, we miss the 5% that are actually scaling — and why. The companies that build and adopt AI that admits when it doesn’t know are quietly rewriting the story. PromptQL is one of them. And if their traction holds, the conclusion isn’t that enterprise AI is a bubble. It’s that a small handful of companies have already figured out how to burst it.

Read Article

State Street is pushing deeper into digital assets by joining JPMorgan’s blockchain-based tokenized asset platform Digital Debt Service as the first third-party custodian. The first transaction State Street anchored was a $100 million tokenized commercial paper issuance by the Oversea-Chinese Banking Corporation (OCBC), a Singapore-based banking group.  State Street Investment Management, the bank’s asset management arm, purchased the debt. J.P. Morgan Securities acted as placement agent. The tokenized asset market could grow could balloon in the next few years, though projections vary from McKinsey’s $2 trillion by 2030 to Ripple and BCG’s almost $19 trillion by 2033. By joining JPMorgan’s blockchain platform, State Street can now offer clients custody of tokenized debt securities without changing its traditional servicing model. In this particular case, State Street manages client holdings in a digital wallet directly connected to JPMorgan’s system, eliminating manual steps in settlement and recordkeeping. The infrastructure supports delivery-versus-payment settlement, with the option for same-day (T+0) settlement, and automates corporate actions such as interest payments and redemptions through smart contracts.

Read Article

 

Diana Robinson, Head of Investments & Advice, UK, Ireland & Channel Islands, J.P. Morgan Private Bank says, “Our mission is simple: to help clients define and articulate their goals and build portfolios that meet these objectives. By analysing client data and preferences, we leverage AI to enrich our traditional, research-driven approach, crafting personalised portfolios tailored to achieve their dreams. JPMorgan Chase’s investment of $17 billion in technology has facilitated the creation of a flexible, common toolchain for software development, integrating external IT innovations to drive operational excellence. These modernisation efforts have positioned the firm to adapt swiftly to technological advancements.  The implementation of the firm’s generative artificial intelligence (GenAI) toolkit, known as the LLM Suite, across 200,000 employees has been transformative. This suite has streamlined processes, enabling significant cost savings through improved fraud prevention, personalised services, trading efficiencies and enhanced credit decision-making. The widespread adoption of these tools has simplified and accelerated tasks, empowering employees to concentrate on more strategic activities. A standout innovation is the Coach application, integrated into the platform we use with our clients. This GenAI tool automates and scales information delivery, empowering advisors to deepen client relationships and expand their client rosters. By providing personalised insights and data, Coach enables advisors to offer tailored advice, enhancing the overall client experience.  2025 has been quite the year to stress-test these innovations. Amid the news-flow whiplashes and steady stream of policy curveballs, our AI tools have helped us deliver timely information to our clients. On ‘Liberation Day’, our advisors were able to provide insights quickly, offering advice that not only responded to current market moves but also aligned with clients’ long-term goals. Equally important, AI plays a crucial role in helping clients understand the risks associated with their investments. Through sophisticated stress-testing models, various market scenarios can be simulated to assess their impacts on client portfolios, empowering informed decision-making. Understanding downside risks allows clients to prepare for and remain invested during adverse periods, benefiting from the compounding of wealth over time.  JPMorgan’s decision to build an internal large language model (LLM) platform underscores our commitment to protecting data assets. By ensuring that data is not used to train external models, we maintain our competitive edge while adhering to evolving regulatory requirements across various jurisdictions. Internally, AI has transformed our operations by accelerating administrative tasks and other execution-based work. The automation of routine processes enhances efficiency and reduces the potential for human error, allowing advisors to focus more on strategic decision-making and client engagement. The integration of AI into our internal processes is a testament to our commitment to operational efficiency and regulatory compliance. A key factor in the successful adoption of GenAI has been the emphasis on making AI tools engaging and accessible. By promoting peer competition and showcasing success stories, we have encouraged the widespread use of AI technologies.

Read Article

State Street is pushing deeper into digital assets by joining JPMorgan’s blockchain-based tokenized asset platform Digital Debt Service as the first third-party custodian. The first transaction State Street anchored was a $100 million tokenized commercial paper issuance by the Oversea-Chinese Banking Corporation (OCBC), a Singapore-based banking group.  State Street Investment Management, the bank’s asset management arm, purchased the debt. J.P. Morgan Securities acted as placement agent. The tokenized asset market could grow could balloon in the next few years, though projections vary from McKinsey’s $2 trillion by 2030 to Ripple and BCG’s almost $19 trillion by 2033. By joining JPMorgan’s blockchain platform, State Street can now offer clients custody of tokenized debt securities without changing its traditional servicing model. In this particular case, State Street manages client holdings in a digital wallet directly connected to JPMorgan’s system, eliminating manual steps in settlement and recordkeeping. The infrastructure supports delivery-versus-payment settlement, with the option for same-day (T+0) settlement, and automates corporate actions such as interest payments and redemptions through smart contracts.

Read Article

 

The past few days delivered some big-name announcements that have triggered a renewed focus on blockchain architecture. One came from stablecoin issuer Circle, which used the occasion of its Q2 results to reveal plans to launch a blockchain called Arc, powered by Circle’s stablecoin USDC. According to Circle’s statement, Arc will be “designed from the ground up to serve the next era of stablecoin-native applications,” such as tokenization, foreign exchange swaps and derivatives, to name just a few. Another was more inadvertent — payments giant Stripe posted an open role for a product marketer for its Tempo blockchain, still in stealth. A journalist noticed, wrote about it and although the ad was swiftly taken down, Stripe has not denied the report. The posting described Tempo as a “high-performance, payments-focused blockchain” being developed in conjunction with VC firm Paradigm. Its head will reportedly be Paradigm co-founder and managing partner Matt Huang. Both announcements raise an obvious question: Why build a new blockchain from scratch when there are already many to choose from in the market? Assuming that customization is the answer, why not at least work within an existing ecosystem? For instance, ethereum is the largest smart

Read Article

During the past decade, J.D. Power has supported automakers in understanding user experiences across more than 100 new technologies, demonstrating a commitment to advancing automotive innovation through consumer insights. The J.D. Power 2025 U.S. Tech Experience Index (TXI) Study enhances these insights with the addition of a smart vehicle category that highlights the growing use of artificial intelligence (AI) in vehicles.  

• Genesis ranks highest overall and highest among premium brands for innovation for a fifth consecutive year, with a score of 538 (on a 1,000-point scale). Cadillac (526) ranks second and Lincoln (523) ranks third.
• Hyundai ranks highest among mass market brands for innovation for a sixth consecutive year, with a score of 493. Kia (474) ranks second and Mitsubishi (471) ranks third.
• Land Rover Defender is the premium model receiving the comfort and convenience award for its advanced air purification system. Toyota Land Cruiser and Toyota Sequoia, each in a tie, are the mass market models receiving the comfort and convenience award for camera rear-view mirror technology.
• Genesis GV80 is the premium model receiving the connected vehicle award for phone-based digital key. Hyundai Santa Fe is the mass market model receiving the connected vehicle award, also for phone-based digital key.
• GMC Hummer EV SUV receives the award for driver assist in the premium segment for active lane change assist. Hyundai Santa Fe receives the award for driver assist in the mass market segment for its blind spot camera.
• Kia EV9 receives the award for electric vehicle in the mass market segment for one pedal driving. The premium segment in the electric vehicle category is not award eligible this year.

Following are some key findings of the 2025 study:

• Smart technologies can improve vehicle quality and satisfaction scores: Certain smart technologies are beginning to positively affect initial vehicle quality scores. For instance, smart climate control systems—which can automatically adjust the vehicle’s heating, ventilation and air conditioning (HVAC) system to improve comfort and efficiency—have shown a notable improvement year over year in user experience, with a reported drop of 6.3 problems per 100 vehicles (PP100), according to the J.D. Power 2025 U.S. Initial Quality Study (IQS).SM This reduction in issues has also driven higher satisfaction in the J.D. Power 2025 U.S. Automotive Performance, Execution and Layout (APEAL) StudySM among users who have this feature. This offers a much-needed workaround to address the increasing complexity of climate controls being migrated into the infotainment system.
• Car wash mode a shiny promise with muddied execution: Car wash mode, a new feature tracked in the TXI Study this year, is intended to automatically prepare the vehicle to go through a car wash, initiating actions such as closing all windows and disabling the windshield wipers, among others. It is a feature that sounds good in theory, but is often buried within the infotainment system, making it difficult to find, which causes delays and frustration, especially when in line at a car wash. This results in more than one-third (38%) of owners who say they need better instructions to use it. Another 15% say the feature is too slow to perform the necessary tasks to prepare the vehicle for a wash. Ultimately, despite its potential benefits, the complicated accessibility of the car wash mode and lack of user guidance are preventing it from becoming a widely adopted feature.
• In-vehicle payments making a comeback: Previous executions of in-vehicle shopping and payment systems within the infotainment system have been hindered by limited app offerings, complex interfaces and lengthy purchasing processes, all of which have contributed to low adoption. However, in-vehicle payment capabilities are gaining interest, with 62% of owners expressing a desire for this feature. At present, the most common uses are paying for charging, fuel, parking and tolls. “The next generation of in-vehicle shopping and payment systems has the potential to succeed if automakers focus on convenience purchases that directly enhance the vehicle experience, such as parking and fuel, and design access within the infotainment system to be simple and user-friendly,” Kathleen Rizk, senior director of user experience benchmarking and technology at J.D. Power said.
• Recognition technologies now next challenge drivers face: Owners’ perceptions of vehicle quality have shifted from broken components to issues with usability, mainly due to complex infotainment and advanced driver assistance systems (ADAS). Currently, problems increasingly focus on technology requiring connectivity. In the future, according to findings in the TXI Study, the next evolution of vehicle quality problems owners will likely face will be related to inconsistent performance of recognition and authentication technologies. These technologies include biometric authentication (29.2 PP100); touchless or hidden controls (19.6 PP100); and direct driver monitoring (19.4 PP100), which are the technologies with the highest number of problems reported in the TXI Study. As automakers navigate challenges with recognition technologies, the focus must remain on ensuring that advancements enhance—rather than hinder—the overall driving experience.
• Blind spot camera is a customer must-have technology: A large majority (93%) of customers say they use the blind spot camera most of the time, and 74% say they desire the feature in a future vehicle. Notably, vehicles that include this feature also spend less time on dealer lots compared with those that lack it. 

Read Article

Walmart Inc. has issued second-quarter results showing strong revenue growth for each of the company’s business segments. Globally, e-commerce increased 25%, led by store-fulfilled pickup and delivery and marketplace, with digital mix up across all segments. Meanwhile, Walmart U.S. comp sales were up 4.6%,with solid growth in grocery and health and wellness.  Other highlights of the quarter included: Revenue of $177.4 billion, up 4.8%, or 5.6% in constant currency; Global advertising business growth of 46%, including Vizio; Walmart Connect in the United States up 31%; Membership and other income up 5.4%, including 15.3% growth in membership income worldwide; Gross margin rate up 4 basis points, led by Walmart U.S.; Operating income decrease by $0.7 billion, or 8.2%, affected by discrete legal and restructuring costs; adjusted operating income up 0.4% in constant currency, due to strong sales and continued execution on the company’s financial framework; growth also affected by about 560 basis points from higher self-insured general liability claims expense; Adjusted EPS of 68 cents, excluding the effect, net of tax, of a net gain of 26 cents on equity and other investments, 5 cents from charges related to certain legal matters and 1 cent from business restructuring charges . E-commerce sales saw 26% growth, with sales through store-fulfilled delivery channels rising nearly 50%. Momentum in advertising continued, including a 31% increase in Walmart Connect sales (ex-Vizio). The gross profit rate rose 26 basis points, with membership income up by double digits. An operating income increase of 2.0%  was attributed by Walmart to strong inventory management and improved e-commerce economics, aided by continued improvement in business mix, although largely offset by a 620-basis-point headwind from increased self-insured general liability claims expense.

Read Article

The second half of 2025 is already shaping up to be a decisive period for banking mergers and acquisitions (M&A). Transaction pipelines are filling, balance sheets remain strong, and technology, particularly artificial intelligence (AI), is redefining what gets bought and how buyers behave. New global deal launches on Datasite, which facilitates about 19,000 new deals annually, rose about 3% compared with H1 2024, showing that appetite for transactions has not faded. On the target side, AI firms have become acutely desirable, even for traditional banks. Acquiring AI capabilities is about survival as much as growth. Banks are buying firms that can reduce compliance costs, predict customer needs, and automate lending or risk decisions. Smaller fintechs with specialized AI capabilities, such as fraud detection, know your customer (KYC) automation, or personalized wealth tools, are now at the front of the line. On the acquirer side, AI is changing how deals get done. Banks are using AI-driven analytics to screen targets, simulate synergies, and stress test proforma financials under different economic scenarios. In due diligence, natural language processing tools scan loan books, contracts, and regulatory filings in days rather than weeks. Execution is becoming faster, more transparent, and in some ways more disciplined. AI is speeding both the why and the how of banking M&A. It sharpens deal rationale and quickens execution. That is why it is no longer only technology-driven banks making acquisitions; traditional lenders of all sizes are joining in.

Read Article