Chapter 93 — The Inseparability of Good and Bad
The Curve of Time, Chapter 93 —— The Inseparability of Good and Bad, in which the Saskias investigate what happened outside the bathroom.
Followed by a little bonus material on what `the whole package’ means in the context of machine learning.
Listen to full episode :
— 93 —
The Inseparability of Good and Bad
A second gunshot cracked through into the bathroom from Mica’s front entryway, and Fish, who had retreated beneath the vanity cabinet, let our a guttural yowl.
Then, everything went eerily silent, until Wassily’s wail pierced the air: “Noooo!”
Saskia and her twin catapulted to their feet, their legs and arms an automatic scramble of spaghetti. They disentangled by the door, and Saskia led them towards Wassily.
Rounding the hallway to the compact entry nook, Saskia saw Mica, hands covering her face.
Wassily faced Mica, fire in his eyes, his clothes covered in blood. “What story?” Wassily was demanding of her.
“My boss. Yesterday. I had notes on Saskia’s lottery win——and Charles Belfry’s.” She dropped her hands and turned to the two Saskias. “With time travelers everyhwere——it was a front page story.”
“So, you brought him here?” Wassily seethed as he gesticulated at the floor in the open doorway behind him.
“Wait.” Saskia held up her hands, begging for calm. And then she saw the crumpled man, blood pooling out from under his akimbo arm, leaking out onto the floorboards. “What the . . . ?”
Wassily turned to Saskia. “His wife was a time traveler. She’s gone——from this timeline——and he”——Wassily gestured at the dead man’s body——“he read——she”——he swung his finger on Mica——“She got her cover story. About you.” His eyes shifted to Native Saskia standing behind Saskia. “Or maybe you.”
Then, Saskia saw Sienna’s body and gasped again.
“And now Sienna is dead.” Wassily fell forward, onto Sienna’s body, which had been obscured by Mica from Saskia’s line of sight.
Saskia looked at Mica. “What is Wassily talking about? What article?”
“I called it: ‘Lotteries: a proving ground for time travelers’,” Mica said, her voice a monotone of shock. “My boss published it. I——I had a draft with Charles Belfry’s name. My boss added your name. I——time travelers are everywhere.”
Suddenly, Wassily turned on Saskia. “Was that why you disappeared to the bathroom? Escaped? You knew this was coming?”
“What?” Saskia recoiled. “This world looks nothing like mine.”
“Except, I’m here,” Wassily snapped as he thumped his chest. He then whirled his finger on Mica. “She’s here.” But, at that point, the fight left his body and he dropped his trembling finger, his hand opening as he laid it gently on Sienna’s hair. “She’s here.” Tears were streaming down his cheeks.
Sienna was gone. Dead. No longer a complication. What an awful thought! And yet, Saskia recognized her own emotion. She’d witnessed it before, in a colleague at work. The woman’s father had died, and she’d given Saskia a lop-sided smile while noting that she didn’t need to worry about him, or her mortgage, anymore.
“This isn’t the break I was hoping for,” Mica mumbled.
Ten minutes ago, via Sienna, Wassily had basked in the warmth of Saskia’s most intimate glow. But that light was gone.
Suddenly, Wassily looked up at Saskia. “You have to go back.”
“Where?” Saskia asked automatically.
“Just go back,” he implored, “and fix this.”
“Wassily . . . .”
“Then teach me to go back.” he glared at Saskia. “I can meditate.”
“It won’t change this world.”
“I can save her.”
“Not in this timeline,” Saskia shook her head. “And there would be two of you there. You don’t get a perfect do over.”
“I don’t want to be here.”
Native Saskia shook her head. “Your lived-life is whatever it’s been. Won’t change.”
Saskia nodded at her. “Incompatible changes just make new timelines.”
Her twin turned to Mica. “If you’re right about the Fermi Paradox——that time travel destroying civilizations is why we’ve never seen an alien——the important thing is going back. Stopping everything from ever happening.”
“Please help,” Wassily implored both Saskias simultaneously.
“It’s not a silver bullet,” Saskia reiterated with a heavy sigh. “It’s messy, like everything else. Lived-life is a trap-door algorithm,” she tried putting it in terms that Wassily would understand. “Once you’ve lived something it can’t change. Any Sienna’s that you save are Sienna’s that only exist because you’ve gone back and intervened.” She moved across to Wassily and held him.
“His wife isn’t even dead,” Wassily sobbed.
From the stairwell, frantic footfalls clattered their way, and preceding them, a shrill woman’s voice: “Mica?! There was a woman——downstairs. A crazy woman. She said you know her husband.” The woman reached Mica’s landing and stopped short as she sighted the glistening pool of red on the entryway floor. “What the——?”
Saskia recognized the woman as a neighbor from Mica’s building.
Mica just pointed at the man on the floor. “It was his own gun.” This, as if it made a difference. “He came——he saw my article. My first front page. Suzy——”
“He killed Sienna.” Wassily interrupted her.
Suzy turned from one living Saskia to the other, and from them to their dead doppelganger. “You’re time travelers?”
Saskia reached forward and riffled through the man’s pockets. She extracted his wallet, and read his name off his driver’s license: “Dennis Wachtel.”
“Dennis?” Suzy gasped.
Saskia tossed the wallet in Suzy’s direction, and reached back into Dennis’ other pocket, coming out with his phone. The lock-screen showed the photo of a woman. She turned the photo to Suzy.
“That was her.”
That was chapter 93, Friends, I hope you enjoyed it!
From the perspective of Wassily’s want——the wish to go back and save Sienna——the crux complication that today’s chapter hinged on was that the timeline you’ve lived can’t be altered. Sure you could go back to an earlier time and fork off a new timeline, but if you did do that, then everything else that happened after the fork would be subject to change too.
Stepping back from the current moment in our story, this mechanism allowed me to resolve the free will/time travel paradox in a——to me at least——intellectually satisfying way. Sure, you can go back and fork off a new timeline, but that won’t save the one you’re in; it’ll just induce a new one into being. Which is to say: our characters’ actions have consequences!
One thing that I did lose as I wrote this up, was an explanation of a fun parallel of this that exists in the world of neural nets and machine learning. Perhaps not so surprisingly, given the tense scene we’re in, I couldn’t find a way to slip this more cerebral observation into the flow. In any event the idea is that while you could always return the dials and knobs of a neural net to an earlier state (if, say, you didn’t like what the recent training of your model had produced), doing so would necessarily abandon any good learnings your model had acquired during that training.
These ideas are, in some sense, the flip-side of something that Saskia alluded to, much earlier when talking about her trash sorting. That is, she, at some point, talked about “off-the-shelf” image recognition software, for instance, that she used for her trash sorting. Rather than training her own image recognition software from scratch. Sure, she would `fine tune’ the model to her need, but having a model that could already recognize basic images would save her a lot of time. She’d also be unlikely to improve on them without a lot of effort——people spend their whole careers working on them, after all.
The transformer architectures of the LLMs, similarly, invariably use pre-trained units in their models, notably for parsing the initial ingestion of text.
The basic proposition is a fun paradigm that crops up often in life: everything has a cost. You want to understand calculus, you need to put in some time. You want to run a four——or five or even six——minute mile, you’ll need to train hard.
Perhaps the flip-side of this rule, is that: as long as you’re prepared to take on the entirety of someone else’s efforts, you can sometimes find shortcuts. To understand calculus, you can stand on the shoulders of giants before you; and to run faster, there are training techniques that have been developed. There is no need to reinvent the wheel every time you want to employ some machine learning either.
The only downside is that, if for instance you want to learn to distinguish (as apocryphally happened on some early image recognition software) between American and Russian tanks, if you naively use the pre-trained model, and that model had been fed Russian tanks in fields of snow and US tanks elsewhere, then the model might use snow as the proxy for deciding if a new photo was of a US or Russian tank. Happily, once you realize that, you could try to correct with new training data, or, worst case, just go back and start all over again, but be careful to include images of each country’s tanks in all environments.
Anyway, it’s nice that the basic building blocks of technology will never be worse than they are today.
As always, until next week, be kind to someone and keep an eye out for the ripples of joy you’ve seeded. There are sure to be some inseparable connections.
Cheerio
Rufus
PS. If you think of someone who might enjoy joining us on this experiment, please forward them this email. And if you are one of those someone’s and you’d like to read more
