14:40Machine-translated from Azerbaijani by AIRead in Azerbaijani
GPT-6 Astra AI Model Completes WoW Orc Starting Zone Without Dying
OpenAI’s GPT-6 Astra artificial intelligence model completed the orc starting zone in World of Warcraft in 40 minutes without dying. The result was announced by the developer of the agent-wow project. The model did not see the game’s visuals, basing its decisions on data received from the server and quest files. The test was conducted on a private server using 1 prompt written for Codex. The agent started the game as a level 1 orc, completed all the quests in Valley of Trials, and stopped in Sen’jin Village.
agent-wow is open-source software designed to allow artificial intelligence agents to play autonomously. The software does not provide ready-made movement, combat, or other gameplay rules. Instead, the agent builds the necessary modules itself. The system connected via the game protocol to the AzerothCore server software running WoW 3.3.5a, the final version of Wrath of the Lich King. This protocol defines the rules for exchanging data with the server. Live World of Warcraft servers were not used in the test. The agent created a module that stored 28 types of messages received from the server. Based on this data, a Python script generated the current state of the game world. AzerothCore’s SQL files were used to find quests. These were used to identify quest-giving characters, quest turn-in locations, and enemy spawn locations. The model completed the quests in sequence, sold unnecessary items, equipped better gear, and learned new abilities. During the test, a reasoning mode 1 level below the highest setting was selected.
To plot its route, the agent developed a C++ utility that worked with AzerothCore’s navigation maps and the Detour library. The program calculated the path between the starting and ending points and sent the coordinates to the agent. According to the developer, the system took advantage of weaknesses in collision rules and map errors in some areas. The 2 quests in the final cave were also completed in the same run. The model had previously completed Portal in approximately 24 hours using screenshots and information about the player’s position. The new test did not use screenshots, and only the starting zone was completed. The project’s next goal is to test whether the agent can reach level 80 completely autonomously. After that, several agents may be tested to see whether they can team up using the game’s social features and complete Icecrown Citadel on Heroic difficulty.
agent-wow is open-source software designed to allow artificial intelligence agents to play autonomously. The software does not provide ready-made movement, combat, or other gameplay rules. Instead, the agent builds the necessary modules itself. The system connected via the game protocol to the AzerothCore server software running WoW 3.3.5a, the final version of Wrath of the Lich King. This protocol defines the rules for exchanging data with the server. Live World of Warcraft servers were not used in the test. The agent created a module that stored 28 types of messages received from the server. Based on this data, a Python script generated the current state of the game world. AzerothCore’s SQL files were used to find quests. These were used to identify quest-giving characters, quest turn-in locations, and enemy spawn locations. The model completed the quests in sequence, sold unnecessary items, equipped better gear, and learned new abilities. During the test, a reasoning mode 1 level below the highest setting was selected. To plot its route, the agent developed a C++ utility that worked with AzerothCore’s navigation maps and the Detour library. The program calculated the path between the starting and ending points and sent the coordinates to the agent. According to the developer, the system took advantage of weaknesses in collision rules and map errors in some areas. The 2 quests in the final cave were also completed in the same run. The model had previously completed Portal in approximately 24 hours using screenshots and information about the player’s position. The new test did not use screenshots, and only the starting zone was completed. The project’s next goal is to test whether the agent can reach level 80 completely autonomously. After that, several agents may be tested to see whether they can team up using the game’s social features and complete Icecrown Citadel on Heroic difficulty.