Hacking In MLBB Without Getting Banned

Building an AI for Optimal Hero Picks

Initial Overview and Goals

• The speaker introduces the project of building an AI to optimize hero picks in games, referencing a previous video on the topic.

• Acknowledges that no updates were made since the last video but plans to implement changes during this session.

Testing Current Model

• The current model was tested by reaching mythical glory with 167 matches and a 66% win rate, despite experiencing some losing streaks.

• Identified flaws in the existing model, including the need for better focus on lane encounters and team composition balancing based on damage types.

Implementing Changes

Lane Counter Feature

• Developed a toggleable lane counter feature that predicts enemy hero lanes and adjusts suggestions accordingly, though it has limitations due to unpredictable role assignments.

• An example is given where the AI suggested an unexpected hero pick (Kaja mid lane) that led to a successful game outcome.

Balancing Draft Suggestions

• Discusses the necessity of balancing draft suggestions based on team needs (e.g., suggesting damage heroes if there are already two tanks).

• Proposes implementing a build suggestion system that provides general directions rather than specific builds, which would enhance reliability.

Challenges Faced

• Despite improvements, frustrations arise from uncooperative teammates in solo queue games, leading to thoughts about creating a feature that highlights poor hero choices instead of optimal ones as a joke.

• Encountered difficulties connecting the program with owned heroes; however, a simple solution was found by adding buttons for scrolling through best hero options.

Final Testing and Observations

Performance Insights

• The AI suggested playing Hannabi again due to her effectiveness against meta heroes with crowd control abilities; this choice proved successful in gameplay.

• While using the lane counter mode, challenges arose from unusual enemy picks affecting predictions; nonetheless, personal intuition helped navigate these situations successfully. Double kills were achieved despite initial concerns about enemy tankiness.

Conclusion of Testing Phase

• Conducted five games across different roles; while some suggestions worked well early on (like Kadita), late-game performance varied due to uncontrollable team dynamics in solo queue settings. Future testing may involve coordinated play with viewers for more reliable results.

Game Analysis and Reflections

Damage Output Concerns

• The speaker reflects on the late game, noting a lack of damage output to compete effectively against opponents. They mention that their team’s Hannabi was underperforming, which contributed to this issue.

• The model suggested using Ammon as a counter but the speaker did not own the hero, indicating a missed opportunity for better gameplay.

• Acknowledges forgetting to utilize lane counter modes, leading to negative consequences in the match.

Hero Selection and Gameplay Experience

• The speaker was advised to play Kaja after an enemy pick of Fanny, which they found logical. Despite challenges faced during the game, they felt their performance was satisfactory.

• Discusses being affected by the "dark system," suggesting that some games are inherently unwinnable regardless of player skill.

Understanding the Dark System

• Introduces the concept of the dark system mode, which recommends heroes based on low performance metrics rather than optimal choices.

• In this match, Valentina was recommended due to her limited utility against the enemy lineup; however, despite these limitations, the speaker performed better than teammates unexpectedly.