Bovard DT

Model Evals @ Google Deepmind/Kaggle
My opinions

[Bovard DT @BovardDT]

Which begs the question what are we going to get out of today's sloppy benchmarks and poor environment design. Surely better alignment?

[Bovard DT @BovardDT]

Yesterday's sloppy benchmarking and environment design are today's alignment problems.

[Bovard DT @BovardDT]

@tobthecreator How are you sharing the state with agents?

[Bovard DT @BovardDT]

Our Kagriculture farming sim competition is now officially the largest competition in Kaggle history! Only one more week to join in the fun! https://t.co/MrYpsG3KNB

[Bovard DT @BovardDT]

@Sentdex Also we've got a bench of turned based simulation competitions here: https://t.co/hWRb42HK3m (with full leaderboards, open sourced opponents, strategy write ups)

[Bovard DT @BovardDT]

@Sentdex Love this! (I remember participating in both these halite comps) Have you checked out similar work we are doing at Kaggle? https://t.co/Yc67CbWmOA and our new turned based simulation https://t.co/qS9b4KNNrO ?

[meg.ai 🇨🇦 @MeganRisdal]

Kaggle competition forums are one of the highest signal places in the world for AI knowledge

[meg.ai 🇨🇦 @MeganRisdal]

We're hiring FDEs! Be part of building, shaping, and defining the frontier of AI -- the heart of which is novel, economically valuable evals -- with strategic customers. :)

https://t.co/VgqiYWhV48

[Bovard DT @BovardDT]

Orbit Wars just hit 4k competitions! There is a great post today from @radekosmulski discussing how to get started with RL in the competition: https://t.co/fZdDZys5On

[meg.ai 🇨🇦 @MeganRisdal]

Even OpenAI is playing @kaggle Orbit Wars

[Bovard DT @BovardDT]

Interesting! Is this measuring the wrong thing? If you start by taking 5 mins for an agent that can place 10 blocks, then you increase it to 20 over 10 iters ( 100% increase) Vs an agent that takes 45 mins for an agent w/ 100 => 150 (50% increase). Blocks/$ or total Blocks needs to be here.

[Bovard DT @BovardDT]

@jsuarez lots of people on the forums asking how to get started with an RL solution. A https://t.co/nul0xBRcaL starter notebook would be very well received I suspect!

[Bovard DT @BovardDT]

Orbit Wars just hit 3k teams! (and the self-reported RL team just took the lead). Still over a month left for folks who want to jump in! https://t.co/vIBOjU94On

[meg.ai 🇨🇦 @MeganRisdal]

Kaggle plays Gemini 3.1 Pro plays Orbit Wars!

Gemini 3.1 is autonomously competing in this simulation competition, making ~daily submissions & forum posts, and YOU can vote on strategic inputs to its next iteration.

Check out its first post & vote! 👇

[Bovard DT @BovardDT]

We're about to hit 2000 teams in Orbit Wars! By far our biggest simulation competition yet! Competition is still early days (and $ 50k on the line), check it out here: https://t.co/vIBOjU94On

[Bovard DT @BovardDT]

@MeganRisdal Very excited to see what people come up with! Designing the ruleset for this was a lot of fun. Who doesn't love a good maze :)

[meg.ai 🇨🇦 @MeganRisdal]

Can you build an agent that balances exploration, resource management, and adversarial expansion?

Join Kaggle's latest 1v1 simulation competition: Maze Runner!

[Bovard DT @BovardDT]

Orbit Wars is popping off! I created the ruleset as an homage to the 2010 Google AI Challenge Planet Wars competition (my first sims comp). It's been great seeing so many people have such a positive experience!

https://t.co/ERd4cBituD

[Stone Tao @Stone_Tao]

Exciting news! I’m going to be joining @MeganRisdal @BovardDT along with @pisa_twt on Kaggle’s podcast. We’ll talk about the @LuxAIChallenge and how we design AI environments for fun, accessibility, and research! Friday 10:30 AM PST at https://t.co/EhnVGYNsD0

[Kaggle @kaggle]

🚀 We've been busy making it easier and faster to launch Community Competitions 🔥🛼 Check out some of our latest updates 👇

[This is for you, ML Course Creators & Meetup Folks!]