Matt Abrams on LinkedIn: Financial Statement Analysis with Large Language Models (2024)

On this #memorialday this picture from Arlington National Cemetery's Memorial Amphitheater (https://lnkd.in/gZbjPkdp) hits me on so many levels. 📍 It is situated on the Western side of the Tomb of the Unknown Soldier (https://lnkd.in/gNKbvQ2D) where you can see people from all backgrounds and perspectives come to solemnly honor and pay witness to the guarding of the Tomb. 📍 It was designed with historical reference to the Greeks & Romans for whom modern Democracy and our Republic descended. 📍 The design, the outdoor setting, and the solemn place in which it is situated where core norms of decency, honor, and kindness are everywhere throughout (and demanded), is a reminder as to what we have failed to hold dear and strengthen. It is a reminder, on hollowed ground, that diverse [groups of] people can and should and, indeed need to, come together in places. Arlington National Cemetery broadly is not simply a place to honor & remember those who gave the greatest of themselves in sacrifice and service to us all, but rather a pointed reminder that, if we are to honor & remember here; if we're capable of sharing in the common norms of decency, honor, and kindness on hallowed ground, then to truly pay our respects we must do so in every aspect of our lives. More simply put perhaps, it represents a place to come together in common bonds (not common agreement) to remember what we're here for and it is in that coming together where we are able to solve our greatest challenges and be perhaps most who we want to be. It often seems we've lost site of that and we all are in our own silos, groups, and identities. Perhaps because it is hallowed ground each of us spends a moment reflecting on our time as finite and what it means, if even for a fleeting moment, to truly honor and pay our respects as Americans, as people, to those that sacrificed all. That is a gift to truly reflect and appreciate as, in so doing, perhaps it allows us to act differently or treat others with a bit more kindness where we can be worthy of those that sacrificed to benefit the rest of us. And most importantly, what it means, as Larry Sampler wrote, "to do better in order to be that American and rebuild an America that is worth the sacrifices of so many." Arlington National Cemetery and the Memorial Amphitheater are designed to incentivize some of the best aspects of us. That is worth understanding and remembering and doing our best to live up to every day of our lives...as we remember and honor those died serving the rest of us.

2

A reminder...on this #memorialday weekend

1

https://lnkd.in/g3wSYHmk The Economist "Baby-boomers were born between 1946 and 1964—and are the luckiest generation in history. Most of the cohort, which numbers 270m across the rich world, have not fought wars....They grew up during strong economic growth. Not all are rich, but in aggregate they have amassed great wealth, owing to a combination of falling interest rates, declining housebuilding and strong earnings. American baby-boomers, who make up 20% of the country’s population, own 52% of its net wealth, worth $76trn. Now the generation is moving into retirement, what are they going to do with their money...it turns out boomers are remarkably stingy—not just in America but across the rich world. They are not spending their wealth, but trying to preserve or even increase it. ...recent evidence has cast doubt on the notion that a baby-boomer spending splurge is on the way...Our analysis suggests that the wealth-decumulation puzzle is becoming still more puzzling, for boomers are more miserly than previous generations...We have looked at household data from a range of rich countries. Turn first to America. In the recent past old people behaved like old people behave in economic models. In the mid-1990s, those aged between 65 and 74 spent 10% more than they took in, eating into their wealth. But since 2015 people this age have saved about 1% of their income....In Canada the saving rate of people over the age of 65 fell during the 2000s. But around 2015, after boomers had started to retire, the decline stopped. More recently the rate has risen. In South Korea from 2019 to 2023 the saving rate of over-65s jumped from 26% to 29%, a bigger rise than in other age groups. In Britain retirees are spending an ever-smaller share of what is coming in. In Australia in the early 2000s, people over 65 saved next to none of their income. In 2022 they saved 14% of it. In Germany from 2017 to 2022 the saving rate for retired folk rose from 17% to 22%. And in Japan the old-age saving rate is soaring. Pensioners account for about 40% of total consumer spending in Japan, less than they did a decade ago, even though there are far more of them.Three factors stand out: "Bequest motives"Many boomers recognise how lucky they are to have accumulated such enormous wealth. They want to pass it on to their children, many of whom are struggling to buy a house or pay school fees. Pandemic Then there is the pandemic. In 2022 American boomers spent 18% less in real terms on dining out than they did in 2019. Bosses at Darden Restaurants, which runs chains including Olive Garden, a purveyor of pasta, recently noted that custom from people aged over 65 “is still below pre-covid”.“Longevity risk” Many boomers will live to 100 and beyond, meaning lots will spend a third of their lives in retirement. This poses a financial burden, especially for those who may eventually need round-the-clock medical care."

4

1 Comment

https://lnkd.in/gqp-RCMh Scientific AmericanTLDR: we need a panoply of solutions that:🎯 strengthen the integrity, quality, and transparency of our information and information supply chains🎯 improve our discourse & dialogue and reduce "us vs them"🎯 provide the foundational backbone for effective civic services🎯 protect human agency in the digital sphere (ie, privacy, security, etc)🎯 implement policies & incentive structures that support the aforementioned and incentivize the attributes and investments critical for trust "...misinformation “solutions” are still based on a rational and ethical figure, an idealized human with no background or social ties, who carefully weighs up all the facts and arrives at “the truth,” contributing to a more civil public discourse. Real people operate on hunches, loyalties and grudges. To combat misinformation, we need to start from this actually existing human, someone who is emotional, factional and frictional.First, humans are emotional, intuitive and instinctual rather than rational. Faced with an incredible volume and velocity of conflicting information, we turn to the nonthought of feeling, reactions and routines. Preferences come before inferences.Secondly, humans are factional. Individuals show favoritism to those within their group while ignoring or ostracizing those deemed to be outside this circle. Factionalism shapes how we assess and even approach information. Group belonging trumps accurate judgment. Partisan differences matter. The “truth” is shaped by identity and sociality.Finally, humans can be frictional. This doesn’t imply everyone exhibits virulent racism or sexism but simply acknowledges that humans are highly attuned to difference, and that favoritism and discrimination can manifest in systemic and subliminal ways...These divisions shape the production, consumption and circulation of information. “Identity propaganda,” for instance, uses othering narratives to delegitimize or even dehumanize out-groups and reinforce membership in the in-group. Us versus them.The messy human tramples through the comforting assumption that misinfo can be solved with more or “better” information. Instead, we need multipronged solutions, combining interdisciplinary insights from media, race and cultural studies, psychology, political science and education to understand what makes anti-immigration or climate denial compelling. Rather than claiming silver bullet solutions, we should acknowledge misinformation is a wicked problem, one without easy answers."

4

1 Comment

https://lnkd.in/gCmNAqFd Theodore Tablante

"...we conducted...analysis of OpenAI #ChatGPT [see limitation(s) below)] answers to 517 programming questions on Stack Overflow and examined the correctness, consistency, comprehensiveness, and conciseness of ChatGPT answers...Our analysis shows that 52% of ChatGPT answers contain incorrect information and 77% are verbose. Nonetheless, our user study participants still preferred ChatGPT answers 35% of the time due to their comprehensiveness and well-articulated language style. However, they also overlooked the misinformation in the ChatGPT answers 39% of the time.Findings:1: More than half of ChatGPT answers contain incorrect information, 77% of ChatGPT answers are verbose, and 78% of ChatGPT answers contain inconsistencies with human answers. 2: Many answers are incorrect due to ChatGPT’s incapability to understand the underlying context of the question being asked. Yet ChatGPT makes fewer factual errors compared to conceptual errors.3: ChatGPT rarely makes syntax errors for code answers. The majority of the code errors are due to applying wrong logic or implementing non-existing or wrong API, library, or functions.4: Popularity, Type, and Recency of programming questions affect the correctness and quality of ChatGPT answers. Answers to more Popular and Older posts are less incorrect and more verbose. Debugging answers are more inconsistent but less verbose. Conceptual and How-to answers are the most verbose.5: Compared to human answers, ChatGPT answers are more formal, express more analytic thinking, showcase more efforts towards achieving goals, and exhibit less negative emotion.6: ChatGPT answers portray significantly more positive sentiments compared to human answers on Stack Overflow.7: Participants can correctly discern ChatGPT answers from human answers over 80% of the time.8: Users overlook incorrect information in ChatGPT answers 39.34% of the time due to the comprehensive, well-articulated, and humanoid insights in ChatGPT answers.9: Participants preferred human answers from Stack Overflow more than ChatGPT answers (65.18% of the time). Participants found human answers to be more correct, concise, and useful.LIMITATION: this work has used the free version of ChatGPT (GPT-3.5) for acquiring the ChatGPT responses for the manual analysis...to understand how differently GPT-4 performs compared to GPT-3.5, we conducted a small analysis on 21 randomly selected SO questions where GPT-3.5 gave incorrect answers. Our analysis shows that, among these 21 questions, GPT-4 could answer only 6 questions correctly, and 15 questions were still answered incorrectly. Moreover, the types of errors introduced by GPT-4 follow the same pattern as GPT-3.5. This tells us that, although GPT-4 performs slightly better than GPT -3.5 (e.g., rectified error in 6 answers), the rate of inaccuracy is still high with similar types of errors. Marty Narayan Jill Sean Thomas Samia Bonan David Tianyi Dennis Gleeson

8

4 Comments