@heyshrutimishra: 90% of AI progress is an illusion built on benchmarks that reward memorization over thinking. TRACES strips that away w…
Summary
TRACES is a new AI benchmark by Apodex designed to measure discoverative AI by focusing on the problem-solving process rather than memorized answers, challenging traditional benchmarks.
View Cached Full Text
Cached at: 08/20/26, 10:53 AM
90% of AI progress is an illusion built on benchmarks that reward memorization over thinking.
TRACES strips that away with just raw discovery process under pressure.
This is what evaluation should have looked like from day one.
@Apodex_AI #TRACES
Apodex (@Apodex_AI): Most AI benchmarks test retrieval — can a model find the known answer? However, the hardest problems in science require discovery, can a system earn an answer nobody has yet?
Meet TRACES 🧭 — the world’s first benchmark for measuring discoverative AI: AI that can work through
Similar Articles
@svpino: I don't trust benchmarks. We've all seen this movie: New model beats everyone else on a benchmark. People hype it. Then…
The tweet critiques traditional AI benchmarks and introduces TRACES, a new benchmark that evaluates AI's discovery process by focusing on how models reach answers, including tool usage, error correction, and evidence tracing.
Apodex Discovery: Reality Benchmarks and Environments for Evaluating and Building Discoverative Artificial Intelligence
This paper introduces Apodex Discovery, a framework and reality benchmark (TRACES) for evaluating and building 'discoverative AI'—AI systems that investigate open-ended real-world problems. The proposed heavy-duty solvers outperform published state-of-the-art on AAV capsid design and improve drug repurposing prediction scores.
@Azaliamirh: Check out TRACE, a new self-improvement approach where the agent identifies the missing capabilities behind its own fai…
TRACE is a new self-improvement approach where an AI agent identifies the missing capabilities behind its own failures and trains itself to address them. TRACE-trained Qwen3.6-27B achieves 73.2% on SWE-bench Verified, outperforming much larger models with fewer training rollouts.
@OkhayIea: Everyone's racing to build "AI scientists." So we asked a blunt question: Can today's best coding agents beat the publi…
Introduces NatureBench, a cross-disciplinary benchmark of 90 tasks from Nature papers to test AI coding agents, finding the best agent (Claude Opus 4.7) surpasses SOTA on only 17.8% of tasks and often succeeds by reducing science to supervised ML rather than genuine discovery.
I built a benchmark for AI “memory” in coding agents. looking for others to beat it.
Developer created a new benchmark called continuity-benchmarks to test AI coding agents' ability to maintain consistency with project rules during active development, addressing gaps in existing memory benchmarks that focus on semantic recall rather than real-time architectural consistency and multi-session behavior.