Recreating Minecraft Is Not a Benchmark
The author argues that current AI benchmarks are flawed because they rely on predictable, visual tasks that labs can easily optimize for. This creates a cycle of 'demo-benchmarks' that measure preparation rather than true model capability.
Why it matters
It highlights a growing crisis in AI evaluation, suggesting that current metrics may be misleading stakeholders about the actual progress of artificial intelligence.
GPT Astra released a couple of days ago and, inevitably, within the hour my entire feed was the same five things: recreating Minecraft in one prompt, painting themselves in MS Paint, the pelican riding a bicycle as an SVG, a ball bouncing in a rotating box with believable gravity, and an SVG game controller.
On paper these look like harder, more visual problems for a model to solve, there’s a reason they’re as big as they are. I’ve started calling them demo-benchmarks, visual and understandable enough for everyone to get but finite enough for the next model to be “perfect” on.
That’s the problem, these tests can’t tell you how good a model is anymore because it’s trivial for labs to optimise for exactly these tests by the next release.
Get smarter about the news
Sign up free for a feed built around what you actually care about, Dive Deeper research on any story, and the full text of every article.
Create free accountAlready have an account? Sign in