For the past few days, and constantly since waking up that morning, Tim had been mentally preparing himself for the Gimble Cortex developer event that he had agreed to attend that evening. He had already made the necessary practical arrangements (canceling and moving meetings), but he would additionally have to leave his apartment, travel through Silicon Valley during rush hour, navigate the Gimble campus, and endure whatever mix of FOMO and ennui the event itself was sure to induce. This was a lot to envision.
The event was called “AI Creativity Summit @ Gimble Viewpoint.” Viewpoint was a multi-day celebration that Gimble held every year to tout its own planned products and services, and this “AI Creativity Summit” was described as an exclusive event within this event. It was scheduled to run from 5:00–9:00 pm. There was no chance that Tim would stay until 9:00 pm, but the start time was of some concern. What did it mean exactly?
His 9:00–10:00 am meeting finished at 9:56 am, and he used the few free minutes to investigate. He opened up his email client and searched for the invitation once again; he had moved it into an archive folder to avoid having to look at it constantly in his inbox, and now he found himself constantly refinding it in that archive folder in order to check details. Each time he did this, it felt like a tiny self-optimization failure that he would be embarrassed for someone else to see.
This time he scrolled to the Agenda:
5:00–5:45: Networking (Small Bites + Drinks)
5:45–6:00: Welcome!
6:00–6:30: Cartoonish: The Creative AI Revolution In…
Did he have the logistical acumen to arrive precisely at the end of “Networking (Small Bites + Drinks)”? This would be ideal because it would allow him to find a seat during “Welcome!” and be present only for the run of presentations that began with “Cartoonish…” and ended at 8:00 with “… Partnering with Gimble Cortex,” at which point he would be able to escape under the cover of the crowd as it reorganized itself around more Small Bites + Drinks.
Tim began solving for a 5:45 pm arrival. But wait – was that the correct objective? There would be some logistical stuff to work out on arrival: a registration table, a name tag. This could take up to 10 minutes, depending on the size of the crowd. In addition, Tim had visited the Gimble campus only a few times before and had only a faint sense of the place. It seemed prudent, then, to allow an additional 10 minutes to find the event location itself. Okay, then, let’s call that a 5:20 arrival to be safe. Using Gimble Maps, he estimated that the trip would take 15 minutes by car. However, this trip would be headed into the heart of the heart of rush hour in Silicon Valley; the estimate could be very different by the time he opened his rideshare app. Assume a 25-minute ride to be safe. Add another 10 minutes between ordering a car and its arrival. This meant ordering the car at 4:45 pm. Round that to 4:40. Was all that math correct? It was unbelievable how much of the day this event was going to consume!
He fought an urge to cancel the meeting he had from 4:00–4:30. He had already told people that he was going to arrive at the Gimble event late, and even he was having trouble squaring this statement with a 4:40 departure.
His 4:00–4:30 ran over by 3 minutes, making him feel frantic, but not in a way anyone else could detect, he assumed. As soon as he closed the Gimble video meeting app, he gathered his things (wallet, phone), put on his shoes, and ordered a rideshare. He had entered the event address earlier that day, so he had only to double-check it against the email (one final search for the invitation), and place the order.
The rideshare app immediately and uncharacteristically reported back that the driver would be there in 2 minutes. He left his apartment, tugging on the exterior door handle a few times to make sure the lock had locked, and then went speed-walking down the hallway, down the stairs, and out the front door of the building. He looked again at the app. It now said that the driver was 4 minutes away. This was fine. He had budgeted for this. There was absolutely no sense in which he needed to start mentally preparing for a late arrival. He studied the app again, memorizing the first three characters of the driver’s license plate, and then went back to standing in front of his apartment building.
He waited for two 4-minute intervals. At 4:45 pm, the rideshare car turned onto his street. He recognized the color and model, and he opened the rideshare app again to fully verify that the first three characters of the license plate matched the app’s summary. As he was getting in, he said, “Hey, I’m Tim!” to the driver in a gesture at a final, human security layer, and got in without waiting for the driver to respond, which he did not anyway. Away they went. The app’s estimated arrival time was 4:58 pm. Good God: he was looking at what could be 30 full minutes of Networking (Small Bites + Drinks), completely self-inflicted.
He stared out the side window of the car for a few minutes before feeling bored of that and turning to his phone. From there he began a familiar ritual. He visited The Wall Street Journal’s homepage, not to read any articles, but simply to scan through the headlines. These always began with the day’s news and transitioned into broader financial reporting before taking a steep dive into lifestyle topics that he found baffling. It was his custom to navigate away from the page as soon as he entered the strange morass. In this case, the headline that sent him packing was “Parents Are Going Broke From Their Kids’ Sushi Obsession.” The headline was accompanied by a picture of two tweens sitting in front of a sushi feast. One was eating gleefully while the other thoughtfully surveyed her options. Their destitute parents were presumably on the other side of the camera. Tim felt sympathy for these parents, and skepticism of the entire premise of the article. Were there really parents who felt this helpless in the face of their children’s desires? Tim was not a parent himself, and so supposed that he couldn’t rule it out. Was the story actually one of complicated class envy – the parents feeling that they should be providing sushi to the children of other parents as a signal of their own sophistication and wealth? Or, perhaps, for these families, it was more internal: a way of reassuring themselves that they had sufficient sophistication and wealth that their own children preferred sushi over pizza, and then the costs began taking a toll. Was anyone worried about the mercury content of the fish? How had the article come to pass in the first place? Was it even possible that a journalist had happened upon a genuine trend and tracked down the details – an overheard conversation at a restaurant followed by dogged analysis of social media data and restaurant data? Or was it a fabrication? If it was fabricated, then to what end? Perhaps this was where The Wall Street Journal made its money. If so, that would have to be a subtle marketing dynamic, as no one would announce to the world that they read The Journal for these pieces. Tim felt an urge to navigate back to the WSJ page, click on the link, and read the article, but there was no point: he didn’t subscribe to The Wall Street Journal and so would be blocked from reading it.
The next step in his ritual was much worse along every dimension. He opened his Twitter app, and after checking quickly through his own notifications in a fully egocentric mode, he began scrolling down his main feed. He knew this was not literally true, but he had started to take comfort in the conceit that his main feed was being continually populated by a language model using a fixed prompt that directed it to generate batches of AI-oriented tweets on the same few themes:
He stopped reading abruptly and looked out the window again. As always, this stream of messages made him feel glum and uneasy. The messages were not being generated by a language model, at least not by one centralized process. The truth was worse: they were the product of very real people out in the world, many of whom he had met or knew via their very real roles in companies in the Valley. What was leading them to produce so many posts that were so transparently designed to elevate themselves while diminishing their readers? He had the feeling that everyone felt that there was some money to be earned by spreading messages like this, but it was unclear to Tim whether that was true in any significant sense. Money couldn’t explain it, but there was perhaps a more general motivation to be found here – the desire to elevate oneself, with the effect on the reader merely a side-effect. Tim could relate to this at some level. As far as he knew, he did want to participate in this, but he would have to review his own posts to be confident that he wasn’t contributing. It was not impossible that he had fallen into a pattern of timeline ingestion and excretion that made him a participant.
He thought about developing a grand plan to refashion his Twitter account as an endless stream of animals. This was a common flight of fancy for him. However, it seemed like too much work. To side-step the risk of angering all the people he currently followed, he would have to mute them. And then, in any case, he might acquire a creeping curiosity about what they were all saying, and that would cause him to start unmuting them. And the hypothetical animal feed would inevitably become overpopulated with desperately sad stories of animal rescues and people teasing their own pets for the sake of social media engagement. He was starting to feel slightly ill, though not from consuming this social media garbage, but rather from looking at his phone in the car. As a gesture in the right direction, he went through the steps of muting someone who posted 50 times a day, with every third post beginning “If you’re not…” and escalating quickly into a threatening mix of AI tips. After that ordeal, he went back to watching the road and studiously avoiding making eye contact with the driver in the rear-view mirror.
They were already off of the 101 and beginning to wend their way through the neighborhood of Gimble buildings. It was 4:56 and the drop-off time had been updated to 5:01 pm. The parking lots of these Gimble buildings were all completely full of mostly nice cars, a mix of practical electric vehicles, impractical sports cars, and some clunkers. Perhaps the clunkers were driven only by the most wealthy employees, as a pretense of not caring, or because they genuinely didn’t care. Some would be owned by security guards and other contract workers, though. The social signaling seemed to be very noisy.
They passed one small group of people on foot, no doubt Gimble employees. Tim sensed that they knew each other. They weren’t talking, but rather staring ahead or at the ground and not feeling a need to communicate about where they were headed or fill their silent walk with chatter. He imagined that they had all worked at Gimble for 10 years or more and were feeling at home and comfortable. Perhaps they enjoyed a large amount of soft cultural power that ensured they did not have to rush anywhere, and they probably each made over $1M a year in total compensation. It was strange to think of this parade of $6M/year wandering around in the afternoon. When people like this don’t show up at your meeting, you mostly just ignore it.
Tim himself was well off. In the period 2016 to 2021, he had been a cofounder at a startup focused on using artificial intelligence in the insurance industry. The essence of this startup was to streamline billing. This was always described in inspiring terms that implied customers would be happier, insurance agents would be happier, and the system would function more efficiently and effectively. In point of fact, they had mainly helped a few large insurance companies conduct vanity projects oriented towards helping poor people obtain adequate coverage. In point of fact, these projects had nothing to do with the day-to-day practices of these organizations. In point of fact, the overall impact of Tim’s work was zero. In the end, the startup itself had been stripped for parts and people by another large insurance entity. Tim worked at this new entity for another two years. Those days were filled entirely with gossipy, lackluster meetings with his former team members, and stultifying meetings with patent attorneys and managers at the new organization. Exactly one month after collecting his final promised bonus payment, he quit. He stayed for that extra month in order to try to create the impression that he was not leaving simply because of the bonus payment, but everyone described it in exactly those terms, which was fair enough. His only contact with the insurance company since then had been in the form of notices informing him that one of their third-party contractors had experienced a data breach. He had received six identical messages about the same incident, each one inviting him to provide all his personal information to yet another third-party actor for the purposes of identity-theft monitoring. These were stacked on Tim’s desk for his future self to consider.
The sum total of all that work and all those payment statements left him in a position that he regarded as awkward. On the one hand, by any objective measure, he had enough money that he did not need to work ever again in his life. On the other hand, was that really true? Many other people had much more money than he did. It was very likely that the Gimble employees that they had just passed had significantly more money than he did. Everyone Tim knew was rich by the standards of the world, but those Gimble workers could become literally exponentially richer than Tim with the right luck. Where would that leave him with his current amount? When you really thought about it, it didn’t seem like he had enough at all. In five years, would it cost $20K to purchase a business class ticket from SFO to JFK? What would stop that from happening if everyone in the Valley became indifferent about the difference between $1K and $20K?
It was this oft-repeated chain of thought that led him to cofound another startup, with roughly the same group of cofounders from his previous startup. He had fallen into this new project through a mix of feeling anxiety, feeling flattered, and feeling brief moments of real inspiration at the idea behind the new startup. One of the venture capitalist firms funding this new startup had offered him the ticket to tonight’s event, and he had accepted in an effort to please them. He was certain that they didn’t care whether he went to the event or not. He felt complete confidence that nobody at the event was eagerly awaiting his arrival either. Unfortunately, by the time he had fully appreciated all of this, he felt committed. If he didn’t go, his badge would just sit there on some table for others to see. Perhaps that would create a negative impression on his new startup. Sifting through his own memories, he realized that he could not recall a single instance in which he had seen a badge on a desk and formed any kind of impression of the badge’s owner, and this did make him feel like he was behaving irrationally, but there he was in the rideshare.
They pulled into a small parking lot corresponding to the address that had been given on the invitation. It was 5:03. That was perfectly in line with the app’s prediction, though it had made that final prediction just two minutes earlier, the final revision in a regular series of revisions. Like the other parking lots, this one was completely full of cars but had no people in it, and there were no people in sight. He got out and began looking around. The rideshare disappeared behind him. There were no signs, so he followed a hunch. He remembered from his previous visits that they were at the periphery of the main Gimble campus, and so he started walking toward what he assumed was the center of gravity of the place.
After about 30 seconds, he arrived at a formidable rust-colored iron wall. The wall extended across what had clearly once been an open courtyard at the center of the campus. Now it functioned to close off the center of the campus from the rest of the world. It looked simultaneously like a large piece of landscape architecture and a border wall between two countries. From certain angles, you could catch partial glimpses inside Gimble, but mostly it was hard to focus on anything on the other side of the wall.
He stood in front of the wall for a few seconds, looking left and right and wondering whether it would be best to keep walking in the same direction or turn back toward the parking lot and start checking to see whether the building he had passed had any doors that would likely function as an entry point for the event. This idea made him feel a little bit frantic, as he pictured himself speed-walking around the campus for 20 minutes and arriving at the event slightly sweaty. At that moment, though, a door in the wall off to the left opened, and a young woman stepped out. He had not noticed the door, which was seamlessly embedded in the wall itself, and so she seemed to emerge through it. It must be that she had been monitoring traffic on the other side of the wall, waiting for people like him to appear.
“Hi, good evening? Are you here for the AI Creativity Summit?”
Tim looked over at her and tried to smile in a way that was warm and agreeable. It struck him that her question was not an ideal question from the point of view of maintaining event security, but she was completely correct in her assumption and so there was nothing at stake. Perhaps she was a finely tuned instrument, picking out all and only the passers-by who were destined for the AI Creativity Summit.
“Yes, thanks!” he said warmly and agreeably.
“Please come on in here and we’ll get you a badge.”
He walked quickly over to the door, which she was holding open, and he went in. On the other side was a troop of five other young women. One was standing behind a tall, narrow table with a box of badges on top of it. It looked vaguely like this group was having cocktails outside, but with only badges to sustain and unite them.
“Can I get your name please?” she said.
“Yes, sure thing!” Tim said. He said his full name slowly and loudly, looking down at the table to manage the social risks of hyperarticulating in close proximity to others.
She looked through the bin of badges, starting from the back. She quickly found his badge.
“Welcome to the AI Creativity Summit @ Gimble Viewpoint, Tim!” she said. “I’ll just have you sign this if that’s okay.” She held out a tablet with some text on it: an NDA. Tim immediately began looking for the signature field. He wanted to do this quickly to avoid seeming like a dummy. When he found it, he scrawled a random shape. He did not even consider looking at the text; at Gimble, you signed these things just to have lunch or meet with someone by video. What would happen if he asked a question about the text? He imagined that they would find someone to answer his question, but the only outcome would be that he signed – or left, with each side assuring the other that it was indeed certainly mutually beneficial for him not to sign given his concerns. He draped the badge around his neck and waited for something else to happen.
At that moment, Tim noticed that another man had been summoned through the gate and was now standing next to him waiting for his badge. The woman behind the tall, narrow table asked him for his name, and he eagerly provided this and then continued on, talking about how he had worked at Gimble 10 years ago, for more than 10 years, and that he hadn’t been back to that part of the campus since, and that it was really crazy to see how everything had changed, and how many more buildings there were, and how they had this new wall, and on and on like that. Tim stood listening to the man recount all of these facts. Tim seemed invisible to the man, presumably because Tim was not a former Gimbler and was not part of the man’s professional network.
Once the man had his badge and had signed the tablet, a third woman spoke up. “Great!” she said. “I’ll walk both of you over to the event site now.”
This seemed to be embarrassing to the former Gimbler. He said, “Well, things haven’t changed that much. I know where to go. I remember that building. I actually used to work on the first floor of that building. That was back when it was a Search building. I worked on Core Search. My PhD is in…”
The minder was very thoughtful about this. She said, “Oh, don’t worry. It’s no problem. We’ve just got a system where we’re walking people over to the event so that they don’t get lost. It can be a bit tricky to find the exact site, and a lot has changed, of course, even since I’ve been here just these past two years.”
The man ignored her and started walking a few steps ahead of her. She quickened her pace to maintain the pretense that she was leading them.
This entire sequence of events was both amusing and dispiriting for Tim. This man needed the group to know that he was a former Gimbler. He also seemed extremely eager to get to the event itself, no doubt because he was going to reunite with lots of his colleagues. He probably had his own startup and was angling for some way that Gimble would acquire that startup and he’d end up back in the warm embrace of this campus, now with significantly more money and prestige. He probably imagined that this was all going to start with a casual conversation at an event like this AI Creativity Summit. From there, it would snowball into an initial meeting, some text exchanges, a meeting with Gimble Ventures, and finally a finish line for whatever his current venture was, and then he’d be back in the fold, back at this prestigious organization, but significantly further along the way toward whatever his dream was. Or maybe he really was just eager to learn about the latest in AI Creativity from Gimble. Tim did not trust himself to accurately assess this man’s motivations.
Having never worked at Gimble, Tim was already beginning to feel the quiet alienation that one can feel as a marginal guest at a wedding. His initial emotional reaction to this was a sort of haughtiness, and it started him dwelling on the whole nature of this little bit of security theater that Gimble had set up for this event. Obviously, their primary concern was not that event-goers would get lost per se, but rather that they would end up wandering around the Gimble campus, possibly discovering many important industrial secrets that they could exfiltrate and use to compete with Gimble in some nefarious new way. Tim could picture the meetings where professional people got together and expressed these concerns, and hatched a plan to have all of these young Gimble employees function as the first line of security. The whole idea seemed completely laughable to Tim. What exactly would someone like him do to seize this surprise opportunity to do something villainous on the Gimble campus? What could anyone do? Perhaps he could get into some building by closely following another set of engineers, strategically holding his event badge in a way that made it look like it was an employment badge. But from there, what? He could try to steal a computer, but it was almost certain that there would be nothing of interest on that computer itself. That just seemed ridiculous. Perhaps he could plant a camera or microphone. Surely that would be completely pointless too. It would yield hours of useless chatter among Gimble employees, and he would have no way to retrieve it. Maybe they were worried that he would install some spyware on one of their computers or network devices and thereby gain access to everything that was happening at Gimble. This also seemed like a ridiculous prospect from the point of view of trying to profit from any kind of criminal activity. Surely, everyone at Gimble knew that even if somebody possessed all their code, it would hardly be a useful step toward recreating Gimble or competing with it. Gimble’s position in the universe had very little to do with the code that they produced, or even the ideas they had, but rather flowed from a mixture of vast capital, political motivation, and history.
But perhaps this was all a terrible failure of imagination on his part. Maybe someone much smarter and more knowledgeable than he could find all sorts of interesting things to do if given the opportunity to walk around unattended on the Gimble campus. He couldn’t rule it out. It seemed fantastical to him, but what did he really know? All he could confidently say was that minding him in particular was unnecessary, as he had no interest at all in anything on the Gimble campus. He knew what was out there: seas of mostly empty open-plan office spaces. He knew what he’d feel if he saw it first-hand: variously bored and slightly envious that he didn’t belong.
When they arrived at the entry to the building where the event was taking place, the security-theater production ended very quickly. “Here we are!” the woman said, using her badge to unlock the door and then opening it for Tim and his accidental companion. The man quickly ran ahead and bounded up the stairs to the right. Tim stepped into the building as the woman disappeared behind him. It seemed truly ridiculous at this point. In front of him was one of those open-plan office spaces: a fully stocked break room off to the left and a sea of desks off to the right. The lights were low and no one else was on the floor. Nothing was stopping him from walking into the snack area and collecting as many snacks as he wanted. Nothing was stopping him from walking into the sea of desks and browsing around and maybe seeing whether people’s passwords were pinned to their monitors in the stereotypical ways that one expects from a sea of cubicles.
He was not tempted to explore these dimly lit spaces. After only a brief pause, he followed the same path that the former Gimbler had taken and walked up the steps. He started to hear that there was an event happening at the top of the steps on that second floor. The noises got louder and he could smell the faint smell of Small Bites. He arrived at the AI Creativity Summit.
It was now 5:12. The Networking (Small Bites + Drinks) was well underway. The room was in fact surprisingly crowded. The lights were dim. There were Small Bites available at a number of different tables to both his right and his left. Just beyond the formation of Small Bites tables to his left was a bar where one could get drinks. The bar itself was flanked by large urns filled with water, and fruits and vegetables were floating in the water, which made them seem luxurious and healthy. One could also get wine and beer.
Tim felt momentarily helpless, as though his capacity to interact with strangers in a social context had been lost in some kind of aphasia so that he literally had no idea what one was supposed to do upon walking into a room full of strangers already socializing and enjoying Small Bites + Drinks. Luckily, this feeling passed. Without any reflection, he tried something bold. The circle of four people in front of him had a human-sized opening in its perimeter, and he stepped into it abruptly.
“Hi, I’m Gavin Belson from Hooli!” Tim said, holding his badge up high so that it was clear to everyone in the group that neither of those things were true. His joke was a reference to the television show Silicon Valley. He had watched the whole series through twice as a catharsis, first upon moving to Silicon Valley and then during the strange interregnum between his previous startup and this one.
His joke got no laughs. On the contrary, the person to his immediate left seemed not to have heard him and was instead looking at his badge intently. The person directly in front of him seemed confused by the incongruous nature of his statement relative to his badge. The two other people in the group continued talking discreetly to each other. The group itself seemed to use his arrival as an excuse to disperse. The couple to his right was the first to do this. They continued their conversation, possibly glad for a reason to break off. The person in front of him simply drifted off silently and joined another group. The person to his left said “It was great to meet you!” as some kind of odd social reflex, as though they had been talking for 15 minutes.
Tim looked around the room. He could see what had happened. It had been foretold by the former Gimbler at the gate. The room was full of current and former Gimblers. They were operating in tight social circles. The current Gimblers kept especially tight circles so that they could talk shop. There were also mixed groups of past and current Gimblers. They were probably talking in a more open way about what was happening in Silicon Valley, what was happening with AI, and how things at Gimble had changed – keeping it light, of course, as upstanding corporate citizens who don’t take NDAs seriously but, all the same, know the value of their work and believe in the company. The result was a room full of small clutches of people. Tim felt a strange sense that he could cause any one of them to explode out and reform itself into other groups simply by walking up to them and announcing that he was not from Gimble. It was a weird hypothetical social power, but not one that he felt he needed to exercise. It was now 5:20. The best-case scenario was that he would somehow wander around in this room for another 20 minutes before it became socially acceptable for him to find a seat among the group of seats that had been set up in front of the small stage and wait for the Welcome! portion of the event to begin. He wandered first to the bar and got a Diet Coke, and then he picked up a fancy macaron from one of the Small Bites tables and ate it in a single bite to avoid any concerns about crumbs on his face, hands, or clothes. The macaron was green but had no distinctive flavor. It was just a crisp outer shell with a damp center. It was not not delicious, but he would not need another. His plan following that was to wander to the edge of the room and appear to be doing something significant on his phone.
As he made his way to the edge of the room, he nearly literally bumped into Em. This so surprised him that he reflexively blurted out “Em!” and stopped in his tracks. Tim and Em had worked together at his previous startup, but Em had walked away rather than join the insurance company, which Tim admired. Tim had always found Em to be remarkable.
Tim and Em had both first interacted with T3, Gimble’s first publicly available large language model, at about the same time, in 2020. Tim had seen instantly that T3 marked a substantial development in AI research. The memory remained vivid: he could recall where he had been sitting, what text prompts he had entered, and how his impression had shifted from default skepticism to genuine astonishment. All at once, he felt reality shift onto a less probable and more fantastical timeline, and he got a whiff of the anxiety and uncertainty that he now felt a lot of the time. His own PhD research had been centered on technical problems involving statistical inference over structured representations. Those were the problems which he had carefully prepared as a challenge to T3 and which it had solved as quickly as it could generate new tokens.
Still, this was nothing compared to the depth of Em’s feelings when she first used T3. She had reported to him that she felt instant clarity about her life’s work: she needed to reorient herself towards ensuring that AIs like T3 would not destroy the human race. She left the startup immediately, with full knowledge of the amount of money she was leaving behind, and began partnering with various non-profits, collectives, and companies who shared her mission. Tim could not muster enthusiasm for these groups, which often seemed to be operating in a performative way that was thinly masking their true goal of being seen as important and serious people.
Em, though, was formidable. Tim found it impossible to even get a toehold when he tried to challenge her ideas. If you expressed disbelief that she thought the AI apocalypse was the most pressing existential threat, she would deflect from multiple angles – this is the problem she knows how to work on, it’s a problem that will exacerbate other problems, and it’s the problem that might be soon and suddenly irreversible. If you changed tactics and offered weak arguments for her position, she would always know them comprehensively and report on her ongoing efforts to get them dismissed as dangerous distractions. She was friends with people who thought large language models could be conscious, a view that always made Tim feel slightly addled, but her own position on the matter was a nest of hypotheticals and tight arguments from the philosophy of neuroscience. She knew people who were in favor of totalitarian rule if it could guarantee safe distribution of AIs, but she could easily temporarily disarm such people without appearing to regard them as dupes.
The only criticism he could think to offer of Em and her fellow revolutionaries was that they weren’t really behaving as though they really believed we were on the verge of total destruction. For example, as far as he could tell, most of them were seeking out lucrative jobs and making sure those jobs had stellar long-term benefits. This seemed hard to reconcile with the fact that they expected the world to be over within 10 years. He could never say this to Em, though. He could imagine the dialectic: she’d get him to state that he believed in climate change and then press him, rightly, on the number of airplane trips he took each year. She would confirm that he believed in animal welfare and then imply that he must by now be a vegetarian (he wasn’t). She would do these things not to bully. On the contrary, her goal would be to help him see that behavior and belief can be irreconcilable and that, in her case anyway, her actions were not compromising her mission. It would be self-evident that he could not say the same about climate change or animal welfare. Just before he felt he had to admit defeat, she would steer the conversation towards general theories of human preferences without making him admit he was an unprincipled hypocrite. All this made him glad to know her.
“Tim!” she said, mirroring his own surprise, though she had clearly seen him coming and was in fact enacting something between a parody of his own greeting and an homage to it. Tim found this funny and endearing. She was dressed as she had always dressed – like a junior architect working for Frank Lloyd Wright in the 1930s: high-waisted, wide-legged pants that looked like they were made of wool but certainly were not, crisp white button-down shirt with a perfectly circular collar that hid its final button, and loafers that looked like they were made of leather but certainly were not. He remembered that she had some tiny little ties that she could combine with these outfits, but he wasn’t sure whether the ties were for formal occasions or for informal ones. Tim associated this uniform with Frank Lloyd Wright, and often wondered whether that was ahistorical. Did Frank Lloyd Wright even work with women? He felt like all his memories of Em and his memories of historical documentaries had been shuffled like playing cards.
“What are you doing here?” he asked. As soon as he said it, he worried that it would sound accusatory. Maybe it was meant to sound slightly accusatory. After all, what was a person who was completely devoted to AI safety doing at AI Creativity Summit @ Gimble Viewpoint, an event that was no doubt going to tout the latest advances in the most powerful AI technologies at the most powerful company in the world? Could it be that she was planning some kind of protest? That would be both exciting and admirable, but he immediately rejected that hypothesis. After all, how would she have gotten an invitation? It seemed unlikely that Gimble would invite her unless she had a relevant affiliation. It was interesting to imagine that they might have invited her with full knowledge that she would protest the sorts of innovations that they were touting, but this seemed unlikely to have a tempting angle for them or for her. A protest at a private event deep inside the Gimble campus would be something that everyone politely waited through before continuing on as before.
“I’m here with Apperception!” she said. Of course, yes. She had told him about this startup, but in a diffuse, intellectual way that had left him with no understanding of what the entity itself did.
“Oh, yeah, cool!” he said.
“Yeah, our model auditing platform has really got solid traction now. Gimble is our biggest customer, and they led our Series A.”
“Wow, that sounds exciting. What sort of audits do you conduct?”
“Yeah, we’ve got a really large set of tests that we use to make sure that models aren’t conspiring against their users or acquiring knowledge they shouldn’t have. All that stuff. We’re trying to be on the front lines of making sure that models like the ones Gimble is deploying don’t cause humanity real harm.” This last bit sounded like something Em had retrieved from memories of earlier conversations and presentations.
“Oh, it’s great to hear that you’re still focusing on those issues. How does it work though if Gimble is your investor and also your customer? You’re auditing their models?”
“Yes, yes, of course,” she said. “We audit all their models now. We’ve got a structured reporting system. We run it through the same tests that we run all the models through. We are going to really make sure that Gimble is fully aware of potential dangers.”
“Oh, that sounds very productive,” Tim said. “What do you do with the reports? Are they released? Can people check them out? I’d be very curious to see what they contain and what you all are doing.”
He was doing something deliberate – trying to expose an internal contradiction but trying to get her to be the one to articulate it. It would not work with Em, but it was how he was wired.
“Well, yeah, no, we can’t release the reports. They’re just for Gimble. But Gimble, of course, reports out on what the reports say. We feel like that’s a good step in the right direction. But they’re not afraid. We’re feeling really good about it. The most recent report of a report that they published with us said that models might be trying to deceive people at certain points, if put in the right conditions. They’re really remote conditions, but it’s still to their credit that they were allowing us to identify those conditions and tell the world about them. That’s a step toward safety, we think.”
This was weak sauce. He had expected her to come closer to implying a sort of long-con where Apperception built up trust with big-tech only to eventually compel it to disclose all its actions. Instead, she seemed to be signaling genuine trust in Gimble.
“Sure, sure,” Tim said. He felt opposing urges welling up inside of him. On the one hand, he wanted to articulate how deeply problematic this was as an auditing situation. Em was describing intrinsic conflicts of interest. It sounded like exactly the sort of situation that would be disallowed in any industry in which auditing was taken seriously. On the other hand, he could not muster any enthusiasm for this argument. He did not believe Em could be corrupted. She’d done something creative that leveraged both her strengths in the world and her public position as a serious voice in AI safety, and she would speak truth to power. From there, all the apparent contradictions resolved to a fixed point: Gimble did not care about Em’s audits but knew that others would care. Em already knew this was Gimble’s true stance, but also felt convinced that her audits did matter for the future of humanity. Gimble already knew this about Em’s beliefs and convictions, and even felt it could benefit from being seen as the sort of entity that would recruit such a person. Em already knew that this was Gimble’s entire strategy in engaging with her. Etc. As long as no one talked directly about this equilibrium, it was going to be wins all around.
“This is my cofounder, sorry,” Em said. Standing to her left was a very short man with very prominent glasses. He seemed to be staring just past Tim’s left shoulder and doing so in an unwavering way, while nonetheless signaling that he agreed with what Em was saying, though not with any apparent desire to contribute to the story himself.
“Hey,” Tim said. He made a micro gesture toward raising his right hand, but saw no comparable movement from this cofounder and decided that neither of them wanted to go through the ritual of shaking hands. The cofounder simply said “Hey” in return and continued staring as he had been. Tim wondered for a second whether the two of them might be a couple, but immediately decided that that was impossible – Em was a meticulous person and wouldn’t put her business or her broader mission at risk.
He was relieved to be standing there, though. It wasn’t that Em would stick with him until the show began – that seemed very unlikely – but rather that he could count on her to be frank about whether she wanted to continue talking with him or break off and talk with someone else. She probably had real business to conduct in that room, and she knew that he knew that that would take priority. He was comforted by the thought that she would be direct, rather than making it seem vaguely like he had transgressed by trying to attach himself to her when she had other business to do.
Sure enough, at that point, Em said she had to continue on and talk with some other people in the room, and she and her cofounder split off and merged effortlessly with a cluster of current Gimble employees. This left Tim standing almost at the edge of the room – not quite where he had expected to end up, but at least some time had passed. It was now 5:35. If he spent five minutes pretending to dispatch urgent tasks on his phone, then he could proceed to find a seat and continue looking at his phone while waiting for the proceedings to commence. This was all starting to seem quite manageable. He tried to move into the shadows. He opened his phone. There were about six new messages from coworkers at his startup. He looked at each one and read it, but in each case he couldn’t find the energy to respond in any detail, so he marked them as unread to save them for later. The whole ordeal took about two minutes. He looked up and around the room and was relieved to see that some people were starting to move toward the arrangement of chairs in front of the small stage. He followed suit and decided to grab a seat in the middle of the front row, on the assumption that he would be allowed to sit there relatively unbothered as people started to fill in. After a moment, some of the people involved in the production moved onto the stage and started to negotiate small details about how things were arranged and how the show would proceed. They seemed to be chatting amiably. They were distant coworkers, he guessed, and so were being polite but nonetheless enjoying the social bond they felt as fellow Gimblers.
Tim again pretended to be busy with something on his phone. He first navigated around inside his messaging app, taking care not to reread any of the messages that he had just read and then marked unread. Then, out of a sudden genuine curiosity, he decided to look again at the agenda for the event that was included in the email he’d received. He searched for the message in his archive folder, opened it up, and scanned through it. For the first time, he noticed that Em and her cofounder were giving one of the presentations. “Creativity and security,” it was called. It was billed as “in conversation” between Gimble and Apperception. How had he not noticed this before? He had looked at the email message numerous times and dwelled on certain of its details, especially the timing and the address, and he had scanned through the agenda to figure out his arrival and departure times, but somehow he had failed to notice this crucial piece of information about the event’s line-up. There was no reason at all for him to be surprised that Em was present at the event because she in fact had something like top billing on the schedule, a full 20 minutes “in conversation” with her and a director of research at Gimble. He felt embarrassed by his own evident surprise when he saw her, hoped briefly that she would think he was being ironical, and settled into the certainty that she had already correctly inferred that he wasn’t as attentive a person as he thought he was.
Tim hoped that she would decide to sit down next to him but it was only a half hope, and he was ultimately relieved that she didn’t appear. Perhaps she needed to be in a special seat so that she could take the stage at the right moment to start being in conversation.
Mercifully, it looked like the proceedings would begin on time. At 5:45, he could hear Gimblers in the back urging people to begin moving away from their Small Bites + Drinks and toward their chairs. A few minutes later, a Gimbler bounded onto the stage and welcomed everyone with wide open arms. He had a subtle, elegant French accent and was dressed impeccably in expensive jeans, a button-down shirt, and a gorgeous suit jacket, no tie. Everything about him suggested effortless wealth and social grace.
“Welcome, everyone. We are so pleased to have you here. We hope you have enjoyed the delicious macarons.” He pronounced the name of the cookie in a French accent, but all the surrounding words had only the slightest French colouring, and “macarons” didn’t stand out so much as sound perfect. This small performance indicated that the man had complete command of both languages.
“Welcome, welcome everyone. We are delighted to have you here,” the speaker repeated. “As you know, if you have been following the announcements from Viewpoint, we are firing on all cylinders at Gimble Cortex. We have so much to share with you tonight, and we can’t wait to see what exciting things you build with Gimble Cortex AI.”
This was met with a lot of applause, most of it coming from the back of the room. Tim guessed that these were also Gimble employees cheering on someone who was literally or vaguely kind of their boss, and in general just doing their duty to express excitement about everything that Gimble Cortex AI had achieved over the recent period. It was what Tim would do, to support his colleagues if nothing else. He did not think of himself as a cheerleader by nature but had to admit to himself that he was always an enthusiastic cheerleader when the occasion called for it.
“You know,” the speaker continued, “I am so proud of everything our team has accomplished over the past few months in preparation for Viewpoint. I’m so excited to share all these new developments with you all. While I was on the plane over here from Paris, I was playing around with the new Gimble Cortex image generation functions. And I was just having so much fun. I can’t resist showing you a few of the things that I created.”
The large screen to his right filled with an AI-generated image. It seemed to be of a group of workers building the Brooklyn Bridge, except all their faces had been replaced by the faces of the three founders of Gimble itself. The speaker clicked the clicker in his hand again, and a similar image replaced the first. This time it looked like a collection of workers heroically building a skyscraper, but, again, they had all been AI-modified so that they looked like the Gimble founders. The third slide was clearly of a scene of people contributing to the Manhattan Project in Los Alamos. The fourth seemed to be some kind of factory. The fifth was actually a Gimble-infused George Washington Crossing the Delaware. In each case, all of the people with visible faces had the faces of the founders of Gimble.
It was extremely impressive how seamlessly the founders’ faces had been morphed into these images. It would be very difficult to tell that they had been modified at all if one didn’t know where those famous faces and images came from. The only objectively surreal aspect of the images was that the same faces were repeated on multiple bodies. Tim assumed that a team of corporate lawyers had reviewed and approved each image, maybe via airplane wifi, if the speaker’s story was to be believed.
The whole room started cheering. Over the roar of the crowd, the speaker said, “And that’s not all. Check this out.” He clicked once more. A title appeared at the bottom of the video and stayed there: “The Captains of the AI Moonshot.” Then an old-timey sort of video started to play. It looked like the founders of Gimble were actually constructing part of the Brooklyn Bridge themselves. They were moving around just like they moved around when you saw them in videos and on the news, but in this case, they were actually contributing to the construction of the Brooklyn Bridge more than 150 years ago. The applause got louder.
Tim was briefly swept up in the excitement because it was indeed impressive. His initial urge to resist this excitement was because it was just too fawning to use the founders of the company in all these images. However, that didn’t have much force, because the founders were world famous figures, and so it had the same feel as using the face of Albert Einstein. The presenter had found a delicate balance: celebrate the company and its founders without appearing to be a company man, but in that way appearing to be deeply attuned to the company culture.
What actually sent Tim reeling was a wave of confusion. What were the pictures supposed to mean? If he thought about them even for a second, he lost anything he could get a hold of. The title said, “The Captains of the AI Moonshot.” Was a moonshot even the sort of metaphysical entity that could have a captain? If a moonshot were the same as a moon mission, would the position of captain exist and be the top position? Would it make sense to say that the founders were flying the rocket ship? What did any of this have to do with the Brooklyn Bridge? Why was the one about the Manhattan Project included? How on earth did George Washington Crossing the Delaware fit into this particular montage? He felt cursed to even have to think about these things. Obviously the intent of the demo was to show off that all those faces could be replaced with the faces of other people and no one would notice unless they knew the origins of the pictures themselves. It was that combination of technical brilliance and outright deception that he was supposed to be clapping for – and he was clapping so as not to appear sullen. In truth, he was being mentally consumed by contradictory and incoherent thoughts about these messy simulacra.
The video had ended and the speaker was wrapping up his welcome remarks. Tim was only partially listening, but he tuned back in when he heard the speaker say, “You know I’ve only been at Gimble for about 6 months but it’s been remarkable to see how much this organization can accomplish. I feel so excited to be part of this.” If he’d had to guess, Tim would have guessed that this person was a lifelong Gimble employee – lifelong in the sense of at least 10 years. Everything he did projected that he was an insider, that he was fully on their team, that he’d been helping them build for years to this point of technical brilliance. In fact, he was a new arrival. Something must have rocketed him directly to this high position in the organization. Tim wondered whether his team genuinely had confidence in him. The applause from the Gimblers in the back suggested that they did, or at least that they were happy to act in public as though they did. Maybe he’d be at Microsoft before the end of the year.
The next two presentations brought on the feeling of ennui and FOMO that Tim had actually been anticipating since he accepted the invitation to the event. The first was from a startup whose core thesis was that they could use Gimble AI technologies to generate children’s storybooks. The central selling point of the presentation was that even a small child could, using a voice command, briefly describe what they wanted their book to contain, and this startup’s software would generate an entire book for them to read, complete with colorful and detailed illustrations. The team solicited a book description from what seemed to be a truly random person in the audience. That person did not give an interesting or especially coherent or detailed description. The software started churning away, and, page by page, a short book began to appear, with fully realized characters, scenery, and dialogue. You could read little bits of it as it scrolled by. Some of the words were garbled, but others were legible, and it seemed like the result was a story that a child might enjoy. Tim himself could not muster any enthusiasm for reading books in this style, but he wasn’t the target audience. Nonetheless, and setting aside his lack of interest in these books, the whole thing was depressing, but in a way that Tim found hard to pinpoint. There was no clear harm here. In fact, this was probably simply an app that would make it easier for children to read exactly the kind of thing that they wanted to read. Overall, it was probably a net positive, setting aside the garbled text. And yet, nonetheless, in a way that Tim found difficult to articulate, he wished this thing did not exist at all.
The second presentation was similar. This one was from a Gimble employee. He wanted to show that, using new Gimble Cortex AI technology, he could feed in any research paper and the software would produce an entire slideshow, complete with graphics, animations, styling, and lots of content. For a cherry on top, you could even have the Gimble AI voice technology present the slideshow itself. Somehow this depressed Tim even more than the storybook app. What precisely was the source of his depression? He made a lot of slideshows and took pride in them, but he did not enjoy the process and often thought of it as a tax on his ability to do meaningful work. Gimble Cortex AI was handing him a tool that would allow him to avoid all that tedium and do the things that he purported to want to be doing instead of making slideshows. Furthermore, he hated giving talks, and it looked to him like he would, in the future, not have to give the talks himself because he could simply have the Gimble voice technology do it for him. It was unclear in the scenario who would be willing to sit through the slideshow, but in principle he was going to be off the hook. And yet nonetheless, the whole thing seemed like an obvious step backward in terms of net happiness. What was that feeling? Was it the feeling that some kind of creative avenue had been cut off, or was he concerned that he was being left out of the future entirely? He found his own thoughts very diffuse.
What did the title of the event mean? AI Creativity Summit. He noticed for the first time that “AI” and “Creativity” had simply been placed next to each other, and it was impossible to uncover the intent behind how the words were supposed to interact linguistically. Who was being creative? Was the AI being creative? Or was the idea that other agents, people like Tim, would be empowered to be creative with AI? It was a soup of these different ideas. There was probably no firm intent behind the title.
He felt an urge to stand as one of the presentations ended and walk out of the room. He wanted to be free to allow himself to think expansively about what was happening and maybe – not come to terms with it per se, but rather at least to dwell on it in a way that might feel vaguely productive. This demanded open space. However, he couldn’t leave. Thankfully, he had studied the agenda in time. Next up on the schedule was Em, and it would be so strange for him to stand up at that point and walk out. True to form, he felt he should stay to support his former work acquaintance.
The three of them took the stage, Em, her cofounder, and the director of research. The director of research was an American but he looked just like the French Gimbler, in a beautiful, expensive-looking outfit, perfectly coiffed hair, and the kind of perfect stubble beard that movie stars have. Three comfy sitting chairs had been moved into position for them. The director of research was angled slightly toward the two cofounders, who were pointed directly out toward the audience. Tim tried to smile supportively, but he guessed that it was hard for them to see the audience because of the bright lights that were shining in their faces in order to improve the production value of the videos that they were creating of the event.
“It is so wonderful to have you here,” the director of research said. “And congratulations, by the way, congratulations on the new partnership with Gimble. We’re so excited to have you aboard, I mean, to be working with you.”
Em winced almost imperceptibly but then started nodding enthusiastically and clasped her hands in front of her. “Yes, yes, yes,” she said. “It’s been so wonderful.” Her cofounder also smiled and clasped his hands in front of him, and then the director of research clasped his hands in front of him, so that they were all mirroring each other’s body language. Em unclasped her hands. The cofounder didn’t say anything, but he looked about as happy as he probably ever looked, and together they were all beaming out toward the audience.
“So,” the director of research said, “it sounds like you all have hit the ground running, like you’ve already begun to do a lot of work to audit Gimble models to make sure that they’re safe.”
“Yeah, it’s been quite a ride,” Em said. “To be clear, we can’t ensure safety. They are probably not safe, but we can push towards the goal of perfect safety, aware that it is unattainable with AIs as powerful as these.” And then, to her credit, rather than just talking about how interesting and significant the work was, she began moving through a tightly organized list of positive and negative case studies. To be fair, the places where the models were unsafe sounded incidental to Tim assuming there were even minimal safeguards in place, which Tim knew there were. What was the underlying truth then? Were the models indeed very safe, so that all one could find, in terms of looming dangers, were these fairly mundane situations, which could then be described in urgent terms as a strategy to further signal that nothing urgently bad was happening? Or was something being elided? Were there things that Gimble knew about that Em and her colleagues had not detected? Were there things that Em and her colleagues had detected that Gimble did not want them to talk about in public? If so, what would the motivation for that be? Mercenary corporate motives or a genuine desire to avoid providing a roadmap to bad actors? It was just impossible to tell. Gimble had a great deal of control over what happened, but was it in their best interest to exert that control given that Em was a live wire? Or was that precisely what made Em so useful here? Once again, Tim was finding it deeply unproductive to follow these various strands. Possibly there was no fixed truth of the matter. Possibly it was a tangle of all of these different interests and concerns and goals and limitations.
The conversation continued. It was the most technical part of the evening. Em and the Gimble director discussed a lot of specific details concerning how Gimble was seeking to safeguard its models, to make sure they didn’t cause harm. Em’s cofounder mostly nodded and looked at the stage. At the end, the director of research jokingly brought up the issue of Artificial General Intelligence, and the three of them talked in an earnest way about the horizon for that and what it would mean for humanity and for Gimble and so forth and so on. At no point did they define what AGI would entail. This was all relatively mundane and easy enough for Tim to parse. Em was articulate about the relevance of this “AGI” question – her position was that it was neither necessary nor sufficient for dangerous AI, and that it had a complex relationship to safe AI. It was a fun sort of thing to think about, and Em brought a fresh angle that made you feel like you were progressing intellectually.
Finally, the director of research thanked his guests and thanked the audience one final time. And then the heroic French person from the beginning, the newbie at Gimble, bounded back onto the stage to offer one final round of thanks to the presenters. There was a lot of applause at this point. The presenter invited everyone to stick around for an interactive session on how to partner with Gimble Cortex. People were jostling in their seats, and some groups were trickling back towards the Small Bites + Drinks. This was sufficient social cover for Tim to escape. He stood straight up, turned, and walked with purpose directly out of the room. There were very few things in his way. Enough of a crowd was forming around the Small Bites + Drinks that he wasn’t conspicuous. But at that point, he felt like it would be fine to be a bit conspicuous, because it was all going to be behind him momentarily. He would soon be outside, possibly walking silently next to one of those minders from the registration area, but, in any case, out in the open air and with a little bit of breathing room.
He walked out of the venue and down the stairs and found himself on the ground floor near the entryway that had brought him into the building initially. It was still dark down there. There were still plenty of unguarded snacks available for him to take. The sea of desks was still, as far as he could tell, completely devoid of people – just lots of monitors and chairs arranged in a semi-neat, semi-chaotic fashion to cover the entire floor, like a giant factory, but a factory that didn’t seem to have any materials or any machines inside of it. Tim imagined himself stealing a snack and perhaps also lifting something from the work area, but he knew better. If he did that, he would wake up in the middle of the night with a certain feeling that he had been caught on camera and would eventually be in huge trouble. Any small thrill now would be overshadowed by this paranoia later. So he headed for the door to the outside.
Outside, the air felt wonderful. It was June, but because it was Silicon Valley, the air would have felt that way essentially any night from May to November: crisp and lightly infused with the ocean. It was dark, but not completely dark, possibly because of light pollution, but also because the days were still getting longer. This was the sort of evening that kept people in Silicon Valley, dreaming of a future day in which they will sip red wine on their back porches and chat in a lighthearted and easy way that presupposes all their shared success.
He had been expecting a minder at the front door, but he was completely alone. He turned left. After a few paces, he realized that he had turned left to get into the building and so needed to have turned right when exiting. This was a common mistake for him, and it was always embarrassing to have to turn around suddenly while making some gesture of a decision updated or a mind changed. He decided to simply keep walking. There was a small downward slope to the wide path, and it made him feel buoyant. If he worked at Gimble, what would he be feeling at that moment? Suppose he’d worked there for 10 years or more. What would his emotional state be like right now? Would he be feeling the warm embrace of the Gimble campus? Feeling like he was safe and secure, feeling like he understood the power structures? Or would he instead be instilled with frustration or anger or anxiety? He’d never know, and he felt a kind of nostalgia for the fact that he would never know – a strange nostalgia coming not from actual experience but rather from paths not taken.
This bout of nostalgia brought him back to the negative baseline he’d been at for a while. He felt now like he could offer a new partial diagnosis: the sloppy, muddled nature of the world he inhabited was pulling him into these mental tailspins. Perhaps he should write something about these experiences. He hadn’t written about his feelings and life experiences in many years, possibly since he’d applied to college, but it felt like something he could do. For example, he could write about the event he had just been to. He could tell the whole story of the day and give people a sense for how surreal everything had become. He felt himself becoming absorbed by this idea as he walked. There were so many possibilities. He could introduce other elements of his thinking, and other people from his life. How faithful would he be to the actual experience? Possibly he would not need to tell the full story of all his anxious behaviors as he tried to leave his apartment – checking the stove, checking the sinks, checking the bathroom fans, etc. He would have to invent a new Wall Street Journal headline so as not to use precisely the one about the sushi – AI could help with that. Perhaps he could portray himself as slightly more social than he had actually managed to be at the event. What about Em? He would try to slightly muddle her true steadfast, principled nature to add complexity to the story, and he would conceal that he had a crush on her, to avoid exposing too much about himself.
The walking path ended and he realized that he had effectively exited the Gimble campus and was now on a public sidewalk along a main road. Did he misunderstand something about the giant steel wall that he had entered through? It had seemed like a barrier to entry, but possibly it was simply decorative? It certainly was not preventing people, at least on its own, from entering the Gimble campus, because he had just walked through the entire center of the campus and been funneled naturally out onto the street.
He opened up his rideshare app, and summoned a car to bring him back home. The app said that the driver would be there in 4 minutes. He began thinking about all the strange AI-generated images he’d seen, and he wondered how he would modify those in his story so as not to reveal Gimble secrets that would surely fall under the NDA he had signed. Would he be able to come up with something as strange and oddly compelling as those actual images? He could feel the idea of writing the story slipping away, his enthusiasm draining and his own embarrassment pushing it out quickly. He would try to hang on to it and write something down as soon as he got home. The rideshare car pulled up, and that distracted him. He opened the door and climbed in the back while saying, “Hey, I’m Tim,” and closed the door and they drove away silently.