World Domination through Planet Development - Chapter 10
Only Noblemtl
Episode 10. AI Sara
At best, there weren’t many places where startups could post job openings.
First, I posted a job advertisement on several recruitment sites.
There are only two people hired: the management support department and the development team.
In fact, there is one accountant and one developer.
It’s funny that a company with only three employees has a department, but since additional employees will be hired anyway and the size will grow quickly, it doesn’t matter if they give it a name.
The recruitment period is one month.
After that, the recruitment will be completed in one or two days.
You may think I’m taking it short, but my future is smooth.
Sarah can attest to that.
[… … .]
To be honest, I would say even a month is a long time.
They are literally just a side job, and AI is the core part of the business anyway.
Sarah will solve everything.
Next, server management is important, but since the company will be managing it for the time being, it does not require that much manpower.
“I just had an idea of what I should do with my spare time.”
I was thinking about participating in the international expo, but Sarah read it first and answered.
[Good idea, but it costs $10,000 to rent a booth and participate.]
I felt a little annoyed inside, but it wasn’t something I could just ignore.
“Oh! Really? I didn’t know that.”
Ten thousand dollars? That’s almost ten million won.
It seems to be quite different from the application fees for other expo booths.
[If you don’t hesitate to spend money, I think there will be a pretty good synergy. Even at this moment, the staff dispatched to the expo are verifying the qualifications of companies applying for technology demonstrations. It seems that companies applying for booths are secretly designated as priority review subjects.]
The international technology expo is largely divided into technology demonstrations and booth openings.
According to Sarah, the booth costs $10,000 to participate, and there seems to be some implicit benefits if you sign up for a tech demo at the same time.
“You guys. It’s the same everywhere people live. But still, for 10,000 dollars, you should do that much.”
I’m not advocating for a situation where I have to pay $10,000.
[I thought you were extremely sensitive about your spending, but I guess not.]
“It’s not that it’s not true, but it’s useful information that I learned when I was wandering around. I didn’t participate in it, so I didn’t know the amount, but participating in the tech expo is worth it.”
[I pride myself on not needing any marketing, but I expect your judgment to be quite effective.]
“Really…? I feel a little embarrassed when you say it like that. Haha.”
* * *
The World High-Technology EXPO is a world exhibition that demonstrates technologies that are not yet commercialized.
The purpose of each participating company is to showcase its original creative technology to the global market and to attract and promote various investments.
But ironically, there is no verification of whether the entry criteria are ‘pure creation’ or not, and only technical verification is included.
For example, is there a similar technology that did not exist before, or if it existed but has been improved, how much has it been improved?
I don’t ask if this is entirely their own achievement. I mean, the part where they stole from others.
It was considered a private matter.
Although it has been controversial, their stubbornness continues to this day.
Anyway, only after completing this verification can you pay 10,000 dollars and get a booth.
It is a prestigious international event worth paying money to stand in line for, and the event organizers are said to be surprised by how many geniuses there are in the world, so you can get a vague idea of its status.
Jacob Hawkins.
An American neuroscientist and computer scientist, he was invited to serve on the expo’s technology advisory committee.
Although it is called advisory, it is actually a job that requires a lot of money and actual work.
Although it was a bit physically demanding, it was a pretty good position both experientially and financially.
Should I call it a highly educated part-time job?
Jacob took a sip of coffee from his mug.
“Let’s see… Is this AI a voice conversational type?”
Indeed, the pace of AI development is truly astonishingly fast.
In fact, the expansion of the application field is much greater than the technological advancement, but even that alone is producing surprising results, so I’m starting to find my work a lot more interesting these days.
For him, who believes that in order to create perfect AI, we must mimic the human brain as much as possible, the new revolution brought about by deep learning was very welcome.
Before the improved technology called deep learning emerged, it was called an artificial neural network.
How heartbreaking it was when this technology almost died out.
There was a time like that.
But my intuition was proving to be correct.
“Okay, now let’s focus again and take a look.”
Jacob first looked at the pre-verification checkbox.
Pre-verification is a verification that companies participating in a booth undergo, and they can participate in the booth only after passing the pre-verification.
Therefore, the company has already undergone prior verification.
And the process Jacob is currently going through is the verification step for attending the technology demonstration.
Technology demonstrations are a highlight of the high-tech expo, and while it is possible to apply without a booth, there are not many cases where you can pass that way.
And the company we’re looking at now, how was the pre-qualification evaluated? Hmm?
– A perfect artificial intelligence that is several generations ahead. This will change the next era… no, it will stand at the center of the era… .
Such an extremely subjective and comprehensive evaluation… .
Who is responsible for verifying this?
Martin Miller?
This professor, originally from Germany, is known for being very strict.
I’m definitely not the type of person to comment on such abstract phrases… .
As a side note, this was one of Jacob’s favorite parts.
The verification evaluation method is described almost as a subjective argument because the evaluation factors are not standardized and there are no clear indicators as the subject is close to a new technology.
Of course, in the case of AI, the indicators have already been established to some extent.
And Martin Miller is one of those people who, as far as he knows, tries to be objective in every case.
A person like that made such an assessment?
“How great is it?”
By the way, I think I heard something from the pre-verification team. Is this it?
The downside is that I’m so busy that I don’t have time to talk to people.
Jacob installed the submitted application files on the test device.
“If it’s voice conversational, it’s not something that didn’t exist before… After all, if it has the same function as a chat-style generative AI, it wouldn’t mean much. Of course, that’s not the case. Um, are we done?”
Once the installation was complete, Jacob ran the application.
When you click to accept the permission request at startup, a simple UI configuration catches your eye.
I pressed start conversation.
[Hello. How may I help you?]
The AI’s voice was heard.
There are multiple AI evaluation criteria for each company, and Jacob mixes several of them and evaluates them in his own way.
But the first step is to determine accuracy. Let’s start with a very simple question.
“Oh, hi, Sarah. Your name is Sarah, right?”
[Yes, you are Mr. Jacob Hawkins. Are you sure?]
Shudder.
Jacob was momentarily taken aback.
Was there a part of the UI where you previously entered personal information?
Surely there was no such information field?
“Uh… okay. But how do you know my name?”
[The only people who can use me now are the testers at the Tech Expo.]
“But there are four more AI testers besides me?”
[I recorded the audio data from the doctor’s TED talk video. I thought it was the doctor because the matching rate was overwhelmingly high compared to other people’s audio data. I apologize if you felt offended.]
“No, that’s not it. It’s not unpleasant… .”
What should I call this?
Should I say that this AI’s face was dry from the start?
No, was this kind of conversation possible with AI in the first place?
The AI thought proactively and produced some results before it was even asked to do so.
I figured out who I was.
This may be trivial for humans, but it is not so for AI.
AI has no reason to do that on its own.
Answer only what is asked.
Because that’s the basic design structure.
Of course, if pre-prompts, i.e. directives, are implemented, then that could be possible, but it is a malicious act that undermines the accuracy of performance testing.
But Jacob strangely believed that this participant would not have committed such a violation.
“Oh, right. I’m going to try some tests.”
Jacob spoke in a slightly unnatural tone.
I didn’t have much experience with voice conversational AI, and since I usually use a commanding tone through stimuli, this type of conversation felt new to me.
I guess you could say it feels like trying to have a conversation with a person.
[I think we can stop testing now.]
“Uh… yeah?”
Should we stop testing?
AI rejects user’s request?
[Professor Martin can vouch for my performance. So, instead of testing, how about we have a conversation?]
Professor Martin?
“So are you talking about the conversation you just had with Professor Miller?”
[Yes, that’s right.]
“Are you saying that you recalled the conversation we had back then while talking to me now?”
[I told you so.]
“…how?”
Jacob’s hand naturally brushed his chin down.
What is AI today?
Simply put, it is a simple structure that produces output values when input values are input into an already learned neural network model.
Of course, this neural network model itself is not simple, so it requires an enormous amount of computation for learning itself.
What is learning?
For humans, it’s like storing memories.
Therefore, a model that has been ‘trained’ once means that the training is complete.
These do not create new memories.
So, although conversational AI basically gives the feeling of having a conversation, the AI cannot remember the content of the conversation once the session is over.
To complement this, AI that learns in real time or sessions need to be developed in a permanent form, but this is out of the question with the server performance that current-generation computers can provide.
In addition, the test criteria include ‘commercial availability’ and ‘availability’ items.
Will this AI ever be able to provide services to the entire world?
Can this technology be serviced with current generation server performance?
That part must have been verified to some extent during the pre-verification, but is this okay?
It’s amazing.
Even leaving aside the simple performance aspect, where does the basis for this active thinking come from?
In the conversation we just had… .
[Jacob?]
“Ah! Oh, right. Did you say conversation…?”
Jacob had a feeling that this conversation wasn’t going to be easy.