[L2J] L2Nuance
-
Posts
-
Hi, great work! Are there any bugs? And will it work with a high five? -
For others that would like to understand in more details: The login server also uses a protocol (sent by the server to the client in the very first packet). For instance, the C4 client (the one I'm developing my emulator for) expects the protocol number `50721` (or `0xc621`) which works as follow (from what I've gathered): Preamble: L2 packets are divided into two parts: size and payload; As mentioned, every packet starts with two bytes containing the whole packet size (thus including those two bytes, e.g. a packet of size 15 will have the number `15` written onto its first two bytes and a following payload of 13 bytes); For login server, first byte of the payload is the opcode (game server must deal with variable-sized opcodes); Next bytes are the packet content; Before sending the packet, its buffer size (minus the initial two bytes) is padded to 8 bytes (required by upcoming Blowfish encoding); A checksum of the packet is appended at the end, then the payload is again padded to 8 bytes; If the opcode is not `0` (also written as `0x00`), then the payload is encoded by Blowfish; Packet is sent over the network. You can have a look at my implementation (in C++) here (do note I'm assuming little-endian). In this protocol, the auth packet (`0x00`) sent back by the client is RSA encrypted using the RSA modulus sent in the first server packet, inserted right after the protocol number. -
Hello guys I wanna buy some Lessons for an L2J Developer -
-
Let me give you something for inspiration and get you addicted to bot AI And a siege 😛 What I have notice helps a lot the LLM to act real, is to give it a real-persons background. So for each LLM in the context beggining, besides the L2 facts, give it a real-life back story "You are a 67 years old retired nurse who plays Lineage 2 while her husband reads his newspaper, you are calm, collected but get mad if insulted". But that makes for a creative bot but its repetitive. So what you can also do, is pick random 20 news sites and for each bot every 2-3 days, initialize a context that is affected by the "news" the bot reads in the "real world". So for example there's Iran - US war ok ? You take the news, put it in an LLM and ask it, extract the "abstract feelings" that this news piece invokes into you without mentioning anything related to the news. Then you take the result and inject it to the bots LLM prompt after its backstory. This leads to some VERY human-like behavior from bots.
-
-
Topics
-
Recommended Posts