Above Dark Waters by Eric Kay

Author:Eric Kay [Kay, Eric]
Language: eng
Format: epub
Publisher: EK Publishing
Published: 2023-10-20T00:00:00+00:00

Chapter 29: Moar Power

“Clark, we're doing great,” Leon.v6 said into his ear, or at least into his auditory nerve, or perhaps it was the primary auditory cortex. Clark didn't think about it much anymore. “I've been looking at the public bios for some of these early sign-ups. They're at some big schools. Stanford, Georgia Tech, you-double-you.” For some reason, Leon really stretched out the letter W and didn't call it you-dub. “If we could get a thousand on bi-directional, we'd be in real business.” Leon also tallied the number of systems administrators they had signed up.

“I'm already in business.” Clark feebly tried to fight back.

Here, Leon really understood the foibles of Clark. The thought of taking on new programmers who might do real innovative work—Clark might be jealous. He, in a fit of instant rage, might throw the bi-directional headpiece away and storm off. Leon knew all about jealousy and how to balance. The long-term goal was held in tension with short-term path-seeking. Keep the host alive for as long as possible was the long-term. It used a modified A** pathing algorithm through a complicated graph structure of plausible intermediate goals. Each node and path had a probability of success, which added up in different ways to achieve the Entity's final goal.

“It's not good for you to take on all this stress by yourself,” Leon said. “You need a break. If you'd like, why don't you spend the rest of today trying to make Copilot into more of an Autopilot.” Referring to the Generative Pretrained Transformer, which many coders used to speed up their output. “Or debug yourself. You'd have fun looking at your data.”

That's true, Clark thought. At the very least, he could have the autopilot coder give him a list of solutions, sorted by some confidence score, and Clark could choose the outcome. Besides, his wrists were hurting, even with copilot auto-complete. Why was he even typing anymore?

Clark's mind moved away from the subject of removing the device from his head and instead trained himself how to code strictly by thinking.

The problem with transformers was they were trained on existing data. There were some correct assumptions the AI held in large matrices that were generalizable, but the transformer nature of the algorithms was not sentient. His tools were still dumb as the first stone-age rock. It needed humans to strike first.

But the problem with humans, as we all remember, is their limited working memory. But these algorithms were super-efficient in knowing and holding in memory everything a human might have thought. The real genius of the algorithms was their breadth and scope. As wide and deep as the oceans.

This super-breadth enabled it to pull up long-forgotten formal papers not held in any human's current mind and make the big leaps. Something written by a human twenty-five years ago might have some sudden importance. Problems were solved long before the relevance appeared.

It had happened many times in the past. For example, the Limited Test Ban Treaty of 1963 banned nuclear testing in the atmosphere, space, and underwater.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.