[Greylock Partners, Sand Hill Road — June 2014, Tuesday, 10:00 AM]
Monica wore a new blazer — charcoal, tailored, the kind of garment that communicated authority without requesting permission. Her legal pad was fresh, the pages unbent, the pen uncapped and ready. She'd arrived twenty minutes before Ethan and had already arranged the conference room: their chairs angled slightly toward the investor's seat rather than facing it head-on, a psychological positioning she'd explained in the car as "presenting together, not petitioning."
"Follow my lead on the business case," she'd said during the drive down the 280. "Jump in on technical questions. If the cloud provider comes up—"
"Supply chain innovation. Proprietary hardware partnerships. I know."
"If your background comes up—"
"The technology speaks for itself. The model is the credential."
"Good. And Ethan?"
"What?"
"Let me sell. You built the thing. Let me sell it."
The Greylock partner was a man named Robert Keating. Mid-fifties, gray at the temples in the Sand Hill Road standard-issue way, wearing a blazer that cost what Ethan used to earn in a month. His handshake was dry and brief. He sat, opened a leather folio, and said: "Monica tells me you've built something interesting. Twenty minutes."
Monica took point. She'd prepared a narrative — not a pitch deck, not a slide presentation, but a story. The failed startup that pivoted. The technology that survived Marcus Webb. The model that generated text indistinguishable from human writing. She wove the facts into a trajectory that felt inevitable rather than improbable, each data point supporting the next, building toward a conclusion that made investment seem not just reasonable but obvious.
Ethan ran the demo when Monica nodded. Three prompts, three outputs, each one calibrated to demonstrate a different capability. Robert Keating read each one. His pen, which had been idle, began moving.
"The architecture," Robert said. "Monica describes it as 'attention-based.' What does that mean, specifically?"
Ethan explained. Multi-head self-attention. Parallel processing of input sequences. The ability to model relationships between words regardless of distance. He kept it accessible — analogies to how humans read, comparisons to existing approaches that made the innovation legible. Robert asked follow-up questions that demonstrated genuine comprehension. He'd done his homework.
"Cloud infrastructure," Robert said. The question Monica had predicted. "Your compute costs are unusually high for a seed-stage company. What's the provider?"
Monica intercepted smoothly. "Proprietary hardware partnership under NDA. The provider offers access to pre-release GPU architectures — next-generation silicon that won't be commercially available for several years. It's a supply chain advantage that directly translates to a competitive moat: Gardner Analytics can train models that competitors literally cannot replicate because the hardware doesn't exist on the open market."
Robert's pen paused. "That's a compelling narrative. Can you verify the hardware specifications independently?"
"We can provide benchmarking data from our training runs. FLOPs per second, memory bandwidth, training throughput — all verifiable against the model's output quality." Monica slid a prepared document across the table. Sarah had built it the previous week: a technical report showing training metrics, hardware utilization rates, and output evaluations. The data was real. The source was impossible.
Robert read the document. Wrote in his folio. Looked at Ethan.
"Your background concerns me. No ML credentials. No academic record in the field. A data visualization company that pivoted to cutting-edge AI research in three months." The words were direct but not hostile — the tone of a man accustomed to identifying risk without personalizing it. "How did you develop this architecture?"
"The architecture was developed through iterative research and a deep understanding of the mathematical foundations of language modeling." Ethan delivered the line Monica had helped him craft — technically true, deliberately vague, positioned to deflect without lying. "Our CTO, Sarah Chen, contributed significantly to the implementation. She has a Stanford background in neural network research."
"A background she left incomplete."
"Which she's been completing through building the most advanced language model in the world."
The ghost of a smile from Robert Keating. "Fair point."
The meeting ran twenty-eight minutes — eight over the allotted time, which Monica told him afterward was "the best possible sign." Robert had asked for a follow-up meeting with Sarah and Marcus. He'd requested access to the demo for his technical partners. He'd used the phrase "very compelling" twice, which in VC-speak was the equivalent of a standing ovation.
---
[Benchmark Capital — Same Day, 2:00 PM]
The second meeting was harder. The partner — a woman named Catherine Li — had a PhD in computer science and had run ML teams at Facebook before moving into venture. She asked questions that required real technical depth: gradient flow in deep decoder stacks, the computational complexity of causal attention, the specific training optimizations that enabled convergence at this scale.
Ethan answered each one. The Phase 2 clarity of the Transformer and the sharpening Phase 1 of the GPT architecture gave him fluency that went beyond prepared responses — he could describe the mathematical properties of attention heads, explain why specific architectural choices improved training stability, and predict the behavior of models at scales they hadn't yet trained. The knowledge flowed from the blueprint in his mind, translated through months of implementation experience, delivered with the confidence of someone who'd spent a decade working with the technology.
Catherine was impressed. Catherine was also suspicious.
"You describe scaling behavior with a specificity that implies empirical observation," she said. "But your training runs are limited to the models you've described. Where does this predictive knowledge come from?"
Monica's hand, resting on the conference table, didn't move. But her posture shifted — a micro-tension that Ethan recognized as the tell of someone preparing to intervene.
"Theoretical extrapolation from our training curves, combined with mathematical analysis of the attention mechanism's properties at different scales," Ethan said. The same deflection he'd used with Sarah. The same deflection that worked less well each time he deployed it.
Catherine wrote something in her notebook. Let it pass. Moved to the next question.
The meeting ended at 2:45. Catherine requested a technical paper — any published or pre-published work that documented the architecture. Monica promised a draft within the month.
---
[Andreessen Horowitz — Wednesday, 11:00 AM]
The third meeting was the shortest and the most promising. The partner — a junior partner named James Wright who'd been given the meeting because the senior partners were traveling — listened to Monica's pitch with the attentive hunger of a young investor looking for the deal that would make his career.
James didn't ask about the cloud provider. Didn't question Ethan's background. He asked one question: "If I give you five million dollars, what does the company look like in eighteen months?"
Monica answered: "Fifty employees, three product verticals, and the only production-quality language generation system on the market. First-mover advantage in a category that will be worth billions within five years."
James wrote a number on his notepad. Showed it to Monica. She showed nothing on her face, but her pen tapped once on the table — the tell Ethan had learned to read as suppressed excitement.
---
[Honda Civic, Driving North on 280 — Wednesday, 1:30 PM]
Monica sat in the passenger seat, her legal pad balanced on her knees, her phone in her hand cycling through emails. The Honda's AC struggled against the June heat — the car's one persistent mechanical failure, a compressor that worked at half capacity and filled the cabin with an intermittent rattle that Sarah had diagnosed as "a dying animal in your dashboard."
"Three meetings," Monica said, not looking up from her phone. "Two strong, one very strong. Robert wants a follow-up. Catherine wants a paper. James wants to move fast."
"How fast?"
"He used the phrase 'pre-empt.' In VC language, that means he wants to close before the other firms can make competing offers. He's afraid of losing the deal."
"Is that good?"
"It's the best possible outcome from a first meeting." Monica set the phone down. Looked out the window. The peninsula rolled past — brown hills, green medians, the occasional flash of a Tesla or BMW in the neighboring lane. "Ethan."
"What?"
"Catherine Li is going to be a problem. She's technical enough to identify the gaps in your explanations. The scaling knowledge, the architectural specificity, the way you describe training behavior you haven't observed — she sees what Sarah sees. What I see."
"But she's interested."
"Being interested and being suspicious aren't mutually exclusive. She'll do her own research. She'll look for the architecture in published literature and not find it. She'll compare your claims to the state of the art and realize you're describing capabilities that shouldn't exist yet."
"We'll publish the paper before she gets that far. The Transformer architecture — the original, not the GPT variant. It establishes a public record. If she searches for the architecture after publication, she'll find our paper."
"Which was published by a company she's evaluating for investment. That's not independent verification."
"It's the best we can offer."
Monica turned back to her phone. The silence between them was filled by the Honda's laboring compressor and the steady hum of highway traffic. Two people in a failing car, driving away from meetings that could fund a future neither could fully predict.
"The James Wright offer," Ethan said. "The pre-empt. What does the number look like?"
"He showed me five million at a sixty million pre-money valuation."
Five million dollars. Sixty million valuation. The numbers belonged to a different universe from the one where he'd eaten stale Cheerios in a dead man's apartment and calculated how many GPU-hours his savings could buy. Four months ago, he'd been negative two dollars and eighty-one cents. Now a Tier 1 VC was offering sixty million for something he'd built with supernatural abilities and a lot of bad coffee.
"That's real," he said.
Monica looked at him. The same evaluative focus, softened now by something that resembled fondness. "It's real. And it's going to happen. James is motivated, his fund has dry powder, and your demo is the most impressive thing he's seen this quarter. We'll have a term sheet by end of next week."
"You sound certain."
"I am certain." She picked up her phone again. Began typing an email. "Stop looking so surprised I'm good at my job."
The callback from their earlier meetings — the same words, recycled through a relationship that had evolved from professional evaluation to something neither of them had named. Ethan let the line land without responding. Some moments were better left unedited.
The Honda merged onto the 101 toward San Francisco. The city appeared through the heat haze — gray and white, stacked on hills, the skyline punctuated by cranes building towers for companies that would fill them with engineers working on problems less interesting than the one Ethan's three-person team was solving above a sandwich shop.
His phone buzzed in the cupholder. A text from Sarah.
GPT decoder implementation at 60%. Marcus found an optimization in the causal masking that reduces memory by 15%. Come back and review.
Ethan typed back one-handed while keeping his eyes on the road — a skill he'd developed out of necessity and retained out of bad habit.
On my way. Series A looks strong. Details tonight.
Monica glanced at his phone. "Sarah?"
"The team. Implementation is ahead of schedule."
"Good. Because once the Series A closes, the twelve-month clock from Raviga resets. New money, new timeline, new expectations. You'll need the GPT model working before the ink dries."
The Honda rattled north. The compressor wheezed. Two people drove toward a future that one of them had seen before and neither of them fully controlled, carrying the weight of five million dollars that didn't exist yet and a partnership built on trust that shouldn't have been possible.
Monica's phone buzzed. She read the message. Her pen tapped the legal pad — once, the tell of suppressed excitement.
"James Wright just emailed," she said. "He wants to schedule due diligence for Monday. He's moving to term sheet."
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