The meeting with Yao Qingfeng in the afternoon went surprisingly smoothly.
This young man, who holds a high position in Penguin's Artificial Intelligence field, surprisingly doesn't exhibit any of the arrogance often associated with young people. He addresses Xi Song's requests objectively and responds to each one individually.
Unfortunately, even with Mr. Ma's influence, Yao Qingfeng wouldn't personally get involved in this project. The fundamental breakthroughs in the Large Model and the commercialization path for B2B clients were truly major issues that the entire group was focused on, so much so that it wouldn't be an exaggeration to say they reached the highest levels of government. That was where Yao Qingfeng's main focus was.
However, after hearing that Bolan Poland Interactive, the platform Xi Song was going to compete with, had been acquired by Origin Group and would most likely use Origin's model, Yao Qingfeng also expressed great interest. In the end, he decided to transfer a research group specializing in Natural Language Processing (NLP) from the Artificial Intelligence Laboratory to be stationed at the Interactive Entertainment Business Group.
With the addition of this Artificial Intelligence team, along with the original development studio of the game "The Alliance Under the City Walls" already selected by Xi Song, and the technical director personally leading the effort, a temporary special project team was thus formed.
The launch meeting took place that afternoon in a Main Conference Room of the Interactive Entertainment Business Group. Xi Song personally presided over the meeting, allowing the people from both sides to meet.
Without any assigned seating, the people who came in gradually split into two groups: the game development team sat on one side, and the Artificial Intelligence Laboratory team sat on the other.
Xi Song observed this without saying anything. Once everyone had arrived, he proactively greeted the group leader from the lab, "Group Leader Jiang, hello, hello. We have a challenging project for mobile games, and we'd like to thank Group Leader Jiang for bringing the team to cooperate. Please welcome everyone and express your gratitude."
After Xi Song finished speaking, he led the applause, and the people in the Interactive Entertainment Business Group followed suit.
The team leader, Jiang Zhao, was also a young man with a similar background to most of the others in the lab: a graduate of a prestigious university with a doctoral degree. He was clearly unaccustomed to this situation and quickly waved his hand, saying, "President Xi, please don't stand on ceremony. We will definitely cooperate with you."
Xi Song smiled slightly; these were the words he had been waiting for.
"Alright, let's skip the formalities and get to three things today: the new event for 'Legend of the Luo River Goddess,' what they accomplished, how they did it, and how we'll create it." Xi Song looked at the technical director sitting next to him and continued, "Mr. Li, please go over the process first."
Technical Director Li Mo stood up and said, "Please look at the slides first."
A flowchart appeared on the large screen in the conference room, which included dialogues between players in different scenes and with different characters. After a simple classification, it was summarized into several major categories, from beginning to end, the classification of options and the required prerequisites, the arrow connections between different scenes, etc., which looked quite complicated.
"This game flowchart is based on videos posted by players online and data summarized from our own testing. You should all have a very intuitive understanding—it's extremely complex." Li Mo paused, looked at Jiang Zhao, and said, "This hot update only contains 28 MB of content, but the amount of updated game content is so vast. Add to that the size of the traffic packets we captured during gameplay—although we can't see the actual content—I believe that most of it was generated in real-time. Team Leader Jiang, from a modeling perspective, do you think this hypothesis is highly feasible?"
Jiang Zhao had never played the game before, and his understanding of the situation came only from the brief conversation he had with Yao Qingfeng before he came, and from the slide in front of him.
After thinking for a while, he offered his understanding. "As far as I know, although Glutinous Rice Ball 1.0 has strong abilities in understanding intent and making logical inferences, the state transitions between so many branches and the maintenance of each game NPC's state, if all goes smoothly, should mainly rely on external persistent memory. In other words, the model is responsible for judgment, not generation; the world state should exist in the database outside of interactions."
Li Mo nodded, switched to a slide, and said, "Another feature of this game is that they innovatively provided players with a free text input method. In other words, players can do more than just what they are told to do in the game; they can let their imaginations run wild. Can our model achieve this?"
Jiang Zhao smiled and said, "President Li, don't worry. Intent understanding is one of the areas of expertise in our self-developed models, so this is not a problem."
"And then what happens after you understand that?" Li Mo asked. "For example, if a player suggests a solution that the developers didn't plan ahead of, can the model determine whether the solution is feasible based on the current situation, and then provide a response that fits the character's identity?"
Jiang Zhao frowned and said, "Our response can only be text or Tool Calling commands, but what the model's response actually does on the software side is up to the software to handle. I can only say that if the judgment criteria are defined clearly enough, we can give it a try."
He paused, then added, "I wonder how long a game sequence in your planned scenario would be, and what the approximate size would be in text format?" The current effective context of "Supreme Purity" is around 200,000 Token. If it exceeds that size, we'll need to consider the issue of phased compression and the loss of consistency caused by compression.
Upon hearing this, Xi Song interrupted, "Doesn't the Origin side have this problem?" Jiang Zhao explained, "They probably do too, but they have experience developing Riding the Wind, so their experience in this area is probably greater than ours. To bring the model's capabilities to fruition, the model itself is one aspect, but external engineering factors are also very important. Our previous collaboration with the Cloud Department focused on this aspect."
Xi Song breathed a sigh of relief. If Origin could do it, there was no reason why Penguin couldn't.
He recalled the densely packed flowchart from earlier and asked, "How long would it take to achieve a similar effect?"
Upon hearing this, Jiang Zhao first looked at Li Mo, and finding that Li Mo was also looking at him, he hesitated for a moment before saying, "President Xi, it's hard to say right now. We can prepare a case study and have President Li send us some specific plot details to work on. If things go smoothly, we should have a result within a month?"
Xi Song frowned slightly, looked at Li Mo, and gave him a wink.
Li Mo smiled and said, "Team Leader Jiang, how about this? Let's work overtime these next few days to build the prototype first, and have a preliminary conclusion by next Monday. Is that alright?"
Jiang Zhao was visibly taken aback when he heard the time, then said, "Well—okay."
After the meeting, the people Jiang Zhao brought did not go back. Instead, Li Mo found a temporary desk next to the game development team's office and sat down to work intensively.
Everyone heard clearly at the meeting just now. If we work harder now, we can work less overtime on the weekend.
Xi Song had barely settled down in his office when he received a message from his superior Ma Tie, asking him to come over.
Xi Song knew exactly what was going on; the boss had gotten him the resources, and now he wanted to see how they would perform.
He quickly tidied up and took the summary of the kick-off meeting to Ma Tie's office. Upon entering, Xi Song saw that Ma Tie's expression was relatively gentle, and his tension finally eased somewhat. He sat down and began his report.
After listening to Xi Song's report, Ma Tie nodded slightly and said, "Well done."
Then he glanced at Xi Song and asked, "Has the preliminary timeline been set? When can we expect results?"
Xi Song mentally calculated that, according to the normal development cycle, two months should be enough for an event. However, this was the first time he had collaborated with the Large Model department, and he didn't know how many problems might arise. So, he cautiously said, "Mr. Ma, I think three months is a safer bet."
Ma Tie didn't comment, but simply said, "Old Xie, when I went to the technical engineering department to ask for resources today, I made a solemn promise: the lab would provide personnel to cooperate with us, and we would also produce results. For this kind of exploratory work, nobody knows where the results will come from, so besides your side, I've also had AAM initiate a project there. We need to visit both sides and see which one the market favors. You understand what I mean?"
AAM Wang is Xi Song's colleague and is in charge of large-scale online games and PC games.
Upon hearing Ma Tie's words, Xi Song felt a buzzing in his head.
It's horse racing again.
It seems that a conservative three-month timeframe is not feasible.