For 40 advance/early chapters : p atreon.com/AutumnXd
"As everyone knows, the learning and training of an artificial intelligence requires enormous amounts of data. Only through training on vast quantities can an AI summarize patterns and features," Ryan said.
"This is also the reason people say AI learning takes a hundred examples to grasp one thing."
"But the distribution of language training resources is extremely uneven. Over ninety percent of the training data is paired data between the two most common languages, and the training data for other languages is severely lacking."
"And training an excellent algorithm model with only a small amount of resources is extremely difficult for peers at home and abroad alike."
"This is also why the quality and quantity of translation between the two major languages far outstrips that of other languages."
"But all of you should be well aware of where our advantage lies..."
At this point, Ryan looked at the crowd around the conference table, and everyone broke into smiles as if on cue.
Training an excellent model with only a small amount of resources.
Was there anyone better at that than they were?
The Argus Algorithm could teach the competition how it was done in the space of a minute.
Ryan made a simple summary. "What we need to solve at present is the first and second problems. The third problem, on the contrary, is a wall of copper and iron blocking off our competitors."
Next, Ryan pulled out a whiteboard and began writing and sketching on it.
At first the crowd had no idea what he was doing, until Ryan finished the top half and started on the bottom half, at which point they caught the first clue of what was going on.
"This is..."
"Could this be a technical approach for solving the first two problems?"
An AI translation expert Hargrove had brought in pushed his glasses up. "This looks like a special encoder and decoder approach!?"
In the traditional line of thinking for solving this problem, an encoder was used to bidirectionally encode the words and sentences. Roughly speaking, the AI would read from both the front and the back of the text to be translated at the same time and encode it, and finally a decoder would decode the encoding, produce the translated sentence, and deliver the result.
That way, when facing the earlier "book" problem, the AI could tell it was a book being read rather than something else, thereby improving the speed and quality of the translation.
But in the approach Ryan was writing out now, this was a brand-new encoder and decoder structure.
The encoder structure was extraordinarily complex, but it could be seen that it would comprehensively consider the text to be translated, calculate the connections that existed among the pieces, and output the corresponding result.
Ryan wrote at the front, and when one board wasn't enough, he switched to another. Hargrove and the others had already planned to hold a discussion session after the meeting anyway, so there were plenty of whiteboards on hand.
The crowd of experts around the conference table watched the approaches described across the many whiteboards, and the more they watched, the more their hearts pounded.
They felt Ryan was a lighthouse, the content on the whiteboards illuminating an entirely new road for them.
"If this structure can actually be realized, then forget being the strongest in the world, this AI's translation ability would probably already be no different from a real human's in the translation field!"
"Worthy of the country's most famous genius! This approach is astonishing! And it seems to be completely matched to the Argus Algorithm?"
"Right. This is a customized encoder and decoder. Leave the Argus Algorithm behind and this approach probably wouldn't work at all."
The crowd exchanged glances.
Clearly, Ryan's strength far exceeded their imagination.
They'd come here today wanting to discuss how to solve these problems, only for Ryan to lay out the solutions directly, as though he'd had a plan in mind all along.
They weren't wrong. Ryan had long since had a plan for these three problems.
He'd taken time the previous night to study the material in the system, and looking at it now, the effect he'd achieved wasn't half bad.
Once he'd finished writing out all the technical approaches, Ryan tossed the pen in his hand aside and turned to face the crowd.
Clap, clap, clap.
Wave after wave of applause suddenly rang out in the conference room, carrying all the way outside, where even the programmers in the office could hear it.
A small scene like this was nothing to Ryan. He'd been through plenty of them.
He calmly had everyone stop the applause, then looked at them. "Which of you is in charge of this AI translation project?"
Hargrove's earlier strength had been in autonomous driving, so as long as he wasn't secretly moonlighting as a career diplomat, it shouldn't be possible for him to be the one leading the AI translation project.
Sure enough, a person raised his hand at once. "That's me."
Aside from Ryan, this was a rare sight of someone not wearing glasses, and he looked quite young, probably only in his early thirties.
"My name is Owen Castellano, and I'm currently in charge of developing the AI translation system." Castellano stood as he spoke.
Hargrove put in a word for him too. "Dr. Castellano was previously the head of the AI translation team in the machine learning department at a major overseas tech giant. I'd been in communication with him about coming back to work here ever since before Argus's first test."
Castellano looked at Ryan with a smile.
Ryan nodded in acknowledgment of him. To have risen to the position of AI translation team head at a giant like that, he clearly had more than a couple of tricks up his sleeve.
Ever since the Argus beta was released, the headcount at the AI research center had kept climbing, and those joining were senior and even top-tier talent, the total number breaking from a hundred and fifty all the way to four hundred.
And this was the result of his father exercising every effort to keep it in check, selecting the best of the best. Otherwise the salaries of these people alone could have dragged Prism Sciences under.
Ryan's thoughts returned to the whiteboards.
"By my estimate, if the approaches up here are fully realized, Argus won't just be able to translate. It'll even be able to connect context while translating and do a simple piece of reading comprehension along the way."
"Next I'll walk you through the approaches up here in detail, and your task is to rapidly turn these approaches into a usable AI translation software."
The crowd around the conference table all nodded, their faces brimming with confidence.
Ryan began explaining according to the approaches on the whiteboards, laying out every key point in great detail.
And the crowd below listened as if entranced, completely immersed in the world of programming that Ryan painted with words and text.
Every one of them took notes on the spot, terrified they'd forget some technical node once the explanation was over.
Even someone like Hargrove, whose focus was autonomous driving, listened with keen interest.
The meeting ran for four hours, and by the time it neared its end it was already noon.
Everyone felt they'd gained a great deal.
Ryan said one last sentence to the crowd. "As for the approaches just discussed, I hope all of you can turn them into a finished product within a month."
This requirement instantly left everyone feeling immense pressure, and the confidence on their faces vanished by more than half in a moment.
A month...
That timeframe was far too rushed.
Ryan looked toward Castellano.
He propped his chin in both hands and thought for a while, then did a quick calculation on paper before finally saying, "It can be done."
Ryan was very satisfied.
"Very good. Then next, we're going to take the offensive against the outside world."
---