[Gardner Analytics Office, SoMa — July 2014, 3:00 PM]
The check cleared on a Tuesday. Three million dollars, wired from Andreessen Horowitz's escrow account to Gardner Analytics' Bank of America business account, appearing on the dashboard as a number so large that Ethan's browser initially displayed it as a formatting error.
$3,000,000.00.
Sarah refreshed the page twice. Marcus refreshed it a third time because he trusted nobody's browser but his own. The three of them stood around Ethan's desk — the same thirty-dollar Craigslist chair, the same laptop with the YC sticker and the GitHub octocat, the same office that smelled like pastrami and ambition — and stared at the number until it became real.
"Fifteen million valuation," Sarah said. Her voice was flat, diagnostic, the tone she used when processing data that exceeded her model's expectations. "Three million on a fifteen million pre."
"James Wright closed fast," Ethan confirmed. "Monica handled the negotiation. Benchmark and Greylock both came in with competing terms, which pushed Andreessen to accelerate. Final deal: three million, one board seat for James, one observer seat for Raviga, twelve-month milestone tied to revenue."
"Revenue, not paper?"
"Revenue. James was specific. He wants to see paying customers, not published research."
Sarah picked up her coffee — Blue Bottle, the three-dollar drip that had been their daily ritual since the financial crisis, back when the two-dollar savings per cup had been a meaningful fraction of their compute budget. She drank. The gesture was the same. The context had transformed.
"So the paper strategy is backup," she said.
"Monica negotiated it as an alternative milestone. Revenue preferred, paper acceptable. But James made it clear: he's betting on a business, not a research lab."
The GPT-1 training run had launched three days before the wire transfer. ChronoCloud's dashboard showed the model at hour sixty-two of an estimated hundred-and-fifty-hour run — loss at 4.7, dropping with the steady gradient that indicated healthy convergence. The architecture was working. The decoder-only design, stripped of the encoder overhead, was training faster per parameter than the original Transformer, exactly as the blueprint in Ethan's mind had predicted.
The early generation samples were already promising. At hour forty, the model had produced a paragraph about startup fundraising that Sarah had read three times before admitting it was "disturbingly good." At hour fifty-five, it had generated a technical explanation of encryption that Marcus declared "more coherent than my college textbook." The GPT path — generation over understanding, creation over classification — was proving its thesis with every epoch.
Monica arrived at four, carrying a bottle of champagne that was better than the Veuve she'd brought to the first celebration and significantly better than the fourteen-dollar sparkling they'd used to toast the production Transformer. Krug. The kind of bottle that communicated: this is a milestone that matters.
"To Gardner Analytics," Monica said, pouring four glasses — real glasses, which she'd also brought, because Monica Hall did not toast three-million-dollar funding rounds with plastic cups. "Which needs a better name, by the way."
"Working on it," Ethan said.
"You've been working on it for five months."
"It's a hard problem."
They drank. The Krug was exceptional — dry, precise, each bubble a small detonation of flavor that made every previous champagne in Ethan's second life feel like sparkling water. Through the window, the June sun slanted across Folsom Street, catching the chrome of parked cars and the glass of office buildings. Below, Manny had propped open the sandwich shop's door, letting the afternoon heat ventilate through his kitchen, and the combined smell of pastrami and sourdough rose through the floor like a benediction.
"I spoke with Linda this morning," Monica said. "Raviga's portfolio review flagged your company as their top-performing seed investment. Linda used the phrase 'category-defining.' Mark Rubenstein — the same Mark who barely voted yes three months ago — asked about co-investing in the Series A."
"Did he?"
"I told him the round was oversubscribed." Monica's expression carried the precise calibration of someone who'd delivered a satisfying rejection to a person who deserved it. "He can participate in the Series B."
Sarah laughed — a full, unguarded sound that Ethan heard so rarely that each instance registered as an event. Marcus raised his glass in a toast to Mark Rubenstein's regret.
The celebration was brief. Monica left at five — she had two other portfolio companies demanding attention, a reminder that Gardner Analytics was not the only startup in her orbit, merely the most unusual. Sarah and Marcus returned to their desks. The GPT training run continued, its loss curve descending in the background like a tide going out.
Ethan stood at the window. The same window he'd looked through on his first morning in this body — different apartment, different view, but the same city. San Francisco in the summer of 2014, golden and buzzing with the particular energy of a tech ecosystem that was about to be reshaped by something it couldn't yet see.
He pulled up his laptop and opened a browser. TechCrunch. Hacker News. The Verge. The tech news landscape had shifted since January — subtly, but measurably. AI was appearing in more headlines. Investment in machine learning had increased by forty percent quarter-over-quarter, according to a PitchBook report from the previous week. Two Stanford researchers had published a paper on word embeddings that was gaining traction in the academic community. Google had made noise about expanding its AI lab.
The ripples. His presence in this timeline — the Transformer built three years early, the HooliBot criticism that had gone viral, the Raviga investment that had signaled to the market that AI was worth funding — was accelerating the field's development. Not dramatically. Not in ways that most observers would attribute to a single company. But the butterfly effects were accumulating, each one nudging the timeline further from the script he'd half-remembered from a show he'd half-watched.
His meta-knowledge was degrading. Not the architectural blueprints — those were supernatural, anchored to the ability rather than memory. But the show knowledge, the plot predictions, the timeline of events he'd used to navigate the first months of his second life — those were becoming unreliable. Monica was investing in his company instead of focusing exclusively on Pied Piper. Gavin Belson had a personal grudge against a man who didn't exist in the show. The AI investment landscape was warming three years ahead of schedule, which would change the competitive dynamics of everything that followed.
The future he'd known was becoming a future he was creating. The distinction mattered. A remembered future was a map. A created future was uncharted territory.
The training run ticked to hour sixty-three. Loss: 4.63. The GPT model was learning.
Author's Note / Promotion: Your Reviews and Power Stones are the best way to show support. They help me know what you're enjoying and bring in new readers! You don't have to. Get instant access to more content by supporting me on Patreon. I have three options so you can pick how far ahead you want to be: 🪙 Silver Tier ($6): Read 10 chapters ahead of the public site. 👑 Gold Tier ($9): Get 15-20 chapters ahead of the public site. 💎 Platinum Tier ($15): The ultimate experience. Get new chapters the second I finish them . No waiting for weekly drops, just pure, instant access. Your support helps me write more . 👉 Find it all at patreon.com/fanficwriter1