The Analyst #10: Sentient
Each month, we’ll tell the story of an analyst’s journey to become one of the greats.
The sunglass frames one could talk to but did not provide any visual insights were never more than a stop gap on the way to more integrated Gen AI. And we got there much sooner than expected. With just about a year to go before the game show called “2028 Presidential Election” to see who follows Trump, Char-lee’s new, fully-air pod-integrated AI OS was becoming as ubiquitous as the iPhone. Data was projected visually without the fake Ray Bans, just right and left of line of sight. You could see Bloomberg terminals, on one side, spread sheets or applications on the other. All while walking the dog. The verbal prompts got better too. Eventually, with the aid of a small implant, user experience continued to flourish. There would be no need for desk, a phone, television. Everybody would have screens and nobody else could see them.
For the ADHD population of social media absorbing teens, it was heaven. Parents had no idea what they were doing. Eventually, there would be a version linked to the implant that would easily detect lying, but getting over the dystopian, thought-police concerns was challenging. Always a senator happy to mug for cameras and tell their voters he was looking out for them. Crypto made it that much easier to pad those sanctimonious pockets off the grid. Just needed to know who take care of on the regulatory side and who to elect the next president - big tech was on it.
Founded in 2025 by Lee Criso, Char-lee combined the OS primary functionality Capture, Harness, Address, Reveal – C. H. A. R. with his first name into a construct people could easily remember and associate. Like his own moniker, he sensed that a gender-neutral, culture neutral, country neutral name would grab the biggest demo. And it didn’t hurt that the software integration was flawless.
The advent of fully integrated, personal AI set off a whole new tech eco-system of winners and losers. The enterprise turned upside down. Any application or software confronted an apple store type-moat, where the hardware and software marriage was indelible. Now, with the power of a strong and growing Gen AI model in Lee’s hands, Gemini, GPT, Claude, and Co-Pilot were rapidly disintegrating in value. The business case for multiple winners failed on the simple premise that the thing that captured everything through the eyes of the user would deliver the best, most comprehensive data. It would also do so with meaningfully less energy need for compute than the hyper-scaler-based farms and farms of data centers the first Gen AI needed.
Char-lee’s hold on Wall Street was menacing the money crowd. How were the traditional mechanics of finance going to adapt? Exchanges, brokers, deals, even research, the supposed last bastion of free thought were becoming increasingly captured and harnessed. Or so it would appear. The biggest hedge funds had spent heavily on AI, convinced they would have proprietary insight to help them trade markets better. Char-lee evaporated that spend in a year. Hundreds of millions were written off, wasted, redundant. Maybe there would be only one…
Inside the big banks, the better nerds saw the writing on the projected walls to the left and right of their line of sight and frantically looked to exploit it via client products. The most successful products were, in the end, remarkably un-creative. They just followed the long line of CMO, CDO’s, private credit and SPACs into developing new things that created their own demand via the magic of scarcity power for IPOs. Nobody cared if they worked or generated alpha, as long as the deal was oversubscribed. They just needed some kind of funky AI use case. And with everyone now getting queued for the deal flow via implants in near real time, demand was never a problem.
Oddly, the market stubbornly refused to fall in line. Sure, the street would create new products that veered close to financial malfeasance. And the big funds would generate attractive returns. That’s what they did. But figuring out a bottom-up, fundamental basis for stock price moves remained an obstinate barrier to Char-lee and its users. As it turned out, markets were not efficient, and the growth in data availability – a trend in place for a very long time – did absolutely nothing to make them more so.
As she walked out of a downtown showing of the sci-fi classic “Bladerunner”, Sydney thew the empy popcorn and soda in the bin, and looked at Matt, “Rutger Hauer, that dude was bad ass.”
“Yeah,“ he said, making a big-eyed nod and continued, “that was the first sentient android character I’ve seen in a movie that really scared the shit out of me. Hauer was almost too good looking to be a bad guy. How could a non-human dare to think or dream about human things. Was that even possible?”
“I don’t know,” she said,” why don’t we ask Char-lee?”
Matt had the implant, so the response came back in seconds on his right screen. He read it to Sydney, who did not want the implant, but instead used a desk-top version of Char-lee out of habit:
“With the advent of LLM’s, most of the writing in human history has been captured and tagged, so that its objectives could be harnessed, to address intent and reveal its meaning. By so doing, these models ably project what humans call emotion or pathos with little to no discernable variance from human-derived content.”
Char-lee was effectively saying, “yes”, they concurred.
But Sydney did not agree, “I think because Char-lee and other bots produce things that look and act like us, it doesn’t make them us. Take markets, for example.” She continued, “What it refers to as ‘human derived content’ is not just numbers, it’s all the cognitive bias that goes into buying and selling stocks. I’ve been watching these models swing and miss at stocks moving in a direction that’s different from the quant forecasts now for three years.”
Matt enjoyed watching her embrace the nerdy market-analyst side, so he poked the idea more, “Are you saying that quant models miss on stocks because they don’t understand humans? Why can’t they just log the outcomes, train the model better and get closer-and-closer?”
“No, they understand us just fine. But they are not sentient. They don’t understand the palpable fear someone has when a short starts ripping their face off. They don’t understand why firms that are supposedly hedge funds have books filled with index shorts and baskets on the opposite side of trades, despite an overwhelming body of evidence highlighting the value of single stock alpha on the short side. They don’t understand why it works to trade momentum factor from the long side and consistently make money. A machine doesn’t really get bias that can’t be explained by logic or math. It’s too random. Too human. And thankfully for us, too inefficient.”
“Ok, maybe I’m just a movie buff that loves acting, but this seems like a solvable problem, no?”
“No.” she continued, “think about it this way: you audition for a part that calls for heart-wrenching break-up after years of caring for someone. You think about the love, disappointment, anger, sadness you need to express. Is there one-way to do that scene? Is there a best way? Or is there just your way and how it’s perceived? We love what Rutger Hauer does because of the actor, not the part. Market participants are actors. They bring their shared experiences, including their biases. And sometimes, they completely fuck up the scene.”
He loved that she got him. And he got her, “So markets don’t need to be right or wrong?”
“Nope. They just need to price in expectations, and if you want me to nerd-out a bit, those are the expectations for future cash flow in today’s terms. It’s 100% subjective until somebody decides to buy the whole company, then you know the price. Otherwise, this lovely, ephemeral thing we refer to as value is going to move around quite a bit.”
The next morning, Sydney headed over to her seat on the desk of a new, but growing fund in Hudson Yards. She was hardly anti-Ai. Char-lee was great at creating code from her desk-top. She used it to track her research notes, get updates on daily news from her watch list of tech names, write code for scenario analysys and to analyze crowding data for identifying good shorts. But selfishly, she was less inclined to use Char-lee for the “reveal”. She liked to do that herself.
Char-lee’s daily suggestion box posted:
“Hi Sydney, if you’d like I can replace your keyboard with a retinal scan to alleviate typing?”
She laughed, “Char-lee is persistent. I ‘ll give’em that,” she mumbled, “but he is not always right.” Then she typed:
Cha-lee, please create a position strategy for the following, “Mr. X is long 600,000 F, and wants to flip into the print on 7.30 as he thinks the quarter will be weaker. When is the best time to start reducing, based on available liquidity data, and likely estimate changes?
Char-lee responded:
“Assuming, Mr. X correctly flags the miss on earnings, he is 65% likely to achieve alpha both long and short by flipping 16 days ahead of the print, based on available trading data for slippage, liquidity and available borrow. Would you like to see a graphic portrayal of projected alpha decay?”
She traded better with this kind of probabilistic feedback but found a few kinks in the underlying code Char-lee produced to explain the forecast. Her view was healthy skepticism: use the tools to get better and know what goes into the processing for recommendations.
The fund leaders noticed Sydney’s consistent PNL generation. She had found a good balance between woman and machine. And they sensed that the machine part in the new Char-lee dominated world was increasingly not within their control. They needed to keep her happy, which in Sydney’s case meant, giving her the space to work quietly at the office and to recognize that during time away from work she was not ‘always reachable’. The nerds liked their away time. Somehow it made them better at the job.
Her prior fund had failed to see that. Despite posting consistent high single digit alpha on a growing but not mega-sized long/short book, her old fund pushed her to :
Run more capital.
Hire a bigger team.
Trade sectors outside of her subject expertise.
The zeal the big funds had for PMs to run ever-larger gobs of gross was done to drive return. Until it stopped working.
She remembered the conversation she had with her boss, when it was time to leave the firm that gave her the first seven years of experience and training, “Look, if this about money, can you at least let us know what we’re up against from the competition? “
“It’s not about that”, she said, “In simplest terms, I’m not a robot. I can’t just keep adding people and trading more gross like you want. I need to sense the pulse of my names, how they move, why they move, what research needs to be done by the team to understand the stocks. And I can’t farm that out to new people or systems. I need time to calibrate all of this to be comfortable with my process.”
Her boss looked stunned,” We provided you with the best training in the hopes that you would get to the place we’re standing now. You are a consistent money-maker, and we want to make this your home for a long time.”
At that moment, she took a deep breath. They just wanted a bigger cog in their money machine. But work was not home for her. She understood her human limitations very well – it was precisely why she was so good. Both sentient and self-aware. In a week or two, she would send a note to her old boss and HR, maybe the firm would might consider a different approach with future talent?