MiscellaneousStuff

Oct 14, 2024

Exploration: Hackathons, AI Agencies, Startups and What Gets Monitored Gets Managed

10 months since last post - what a wild year! Shifted from consulting and AI research to hackathons, startups, and a crash course in business and relationships.

Dec 23, 2023

TLoL: Human level in League of Legends using Deep Learning (Part 7 - Dataset Acquisition, Transformation and Model Training)

This post covers the acquisition of a large and diverse dataset of replays, the transformation of these replays into a format suitable for training and the training of basic models.

Sep 17, 2023

MechInterp: TinyStories-1Layer-21M Model Embed, Attention and MLP Analysis (Part 1 - Experiment Plan, Basic Attention Analysis)

This post begins the exploration of mechanistic interpretability for large language models.

Nov 19, 2021

TLoL: Human level in League of Legends using Deep Learning (Part 6 - Dataset Generation)

Following a long hiatus, this section will explain the data generation procedure for the TLoL project and how the datasets generated will be used to create the first agent which can play League of Legends.

Sep 8, 2021

TLoL: Human level in League of Legends using Deep Learning (Part 5 - Download Scraping)

In summary, we are now able to find a list of relevant replay files using u.gg. Next we will need to use the League Client Update to download the *.rofl files automatically.

Sep 4, 2021

TLoL: Human level in League of Legends using Deep Learning (Part 4 - Exploring the Literature)

This post reviews successful MOBA playing AI agents within the literature, what they done well, what could have done better and then concludes with what can be used in creating a human-level League of Legends AI.

Sep 3, 2021

TLoL: Human level in League of Legends using Deep Learning (Part 3 - Initial Ideas)

This post explores how to acquire replay data in League of Legends, the League of Legends rofl replay format, and how to extract granular information from a replay file using a method which alleviates the encryption of the replay files in a way which is robust from patch to patch, as the game is updated once every two weeks.

Sep 2, 2021

TLoL: Human level in League of Legends using Deep Learning (Part 2 - Problem Analysis)

This post reviews existing approaches to creating game playing AIs which have evolved in complexity over time as the complexity of the games which have been tackled has increased. The approaches are than analysed and the key strengths and weaknesses, as they apply to creating a human-level League of Legends AI, are outlined. Then previous approaches are analysed, with regards to how they can inform the development of a League of Legends AI.

Sep 1, 2021

TLoL: Human level in League of Legends using Deep Learning (Part 1 - Existing Solutions)

This post reviews existing League of Legends AI systems from different games which have been released by Riot Games themselves, to bots and scripts which have been ever present since the initial release of the game and finally to recent academic approaches which have used cutting-edge AI techniques to create AI systems which can play League of Legends.