- Blind mode tutorial
lichess.org
Donate

My learnings winning against Leela

ChessAnalysisSoftware Development
How playing against Leela reminded me not to give up

Chess engines usually win by brute force, outplaying humans with endless calculations. But Leela Chess Zero (Lc0) is different. It is inspired by DeepMind’s AlphaZero, a neural network, Leela doesn’t just compute; it learns. By playing millions of games against itself, it has built intuition, almost like a human. This results in an engine that feels creative, strategic, and brutally accurate.

I recently battled two of Leela’s “handicapped” versions on Lichess: LeelaKnightOdds (no knight, blitz) and LeelaQueenOdds (no queen, bullet time). Leela is estimated at about 3000 rating even with these odds.

Here’s what I learned and how these games transformed my play.

LeelaKnightOdds

Leela starts without a knight but compensates with eerie positional mastery. But what made it truly tough wasn’t just the raw strength. It was how it played. Leela often played moves that felt passive or unneeded. However, a few moves later, I would realize it was quietly setting traps or laying the groundwork for subtle positional squeezes that I hadn't even noticed.

One of the most frustrating patterns is that it would consistently limit my options, cutting off key squares. Before I knew it, I’d be staring at a position where I had no good moves left. I had to lose more than 20 blitz games in row, before I finally managed to beat LeelaKnightOdds.

Key learnings

  • I noticed in games that both attacking and defending provided better winning chances
  • Controlling the center particular early in the game allowed more piece activity gave me better odds
  • Be mentally okay with some suffering as it took a lot of games before I somewhat understood Leela's style
  • Trust your tactics, after 33...Nxe3 in the below game I knew I could win

https://lichess.org/AB0R4NZoyOct

LeelaQueenOdds

I also played Leela at Queen odds. It starts down a queen in a bullet game. The best I have been able to do to date is manage to draw from a winning position. This was a classic bullet time scramble that went wrong.

Key learnings

  • Don't be afraid to exchange your Queen for a combination of pieces (R+B or R+K)
  • Always control as much space as possible especially when being attacked
  • Play faster? I squandered my Queen with less than a second left

https://lichess.org/lsKiJwhWyb8l

Improving going forward

After analyzing these games, I realized traditional chess engines tell you a move is bad but rarely explain why. That’s why I am building an AI Chess Coach designed to:

  • Highlight critical moments: The coach identifies the four most critical moments in your game.
  • Explain moves: Explains each move like a human coach would through text and voice commentary, with clear and actionable insights.
  • Visualize tactics: Highlights key squares to help you see tactics unfold on the board.
  • Teach principles: Provides tips rooted in classic chess principles to help you improve.

After analysing my games using my Chess Coach, I have became more aware of tactical sequences and positional nuances that have improved my play.

If you would like to try it out and follow my journey please go to https://chesscoach.dev/