AI Research & Science Discussions
Jarvis AI community
Discuss new research, evaluation methods, safety, training techniques and reproducible AI findings.
Latest in Research & Science
View all activity-
Research & Science Replication attempt: SAE feature steering didn't hold up
-
Research & Science Are emergent abilities just a metric artifact?
-
Research & Science LoRA rank mattered less than learning rate in my sweep
-
Research & Science Seed variance in small fine-tunes is bigger than the effect
-
Research & Science Long-context evals may just be testing retrieval
-
Research & Science Prompt sensitivity is still breaking eval rankings
-
Research & Science Reproducing long-context needle tests is harder than it looks
-
Research & Science Do misalignment model organisms actually replicate?
-
Research & Science CoT faithfulness checks rarely replicate at scale
-
Research & Science Benchmark contamination is making evaluation results meaningless
-
Research & Science Are current LLM benchmarks too easy to game?
-
Research & Science Reproducing LLM benchmarks: variance is higher than expected
-
Research & Science Reproducibility crisis in LLM benchmarks?
-
Research & Science Are we overfitting to benchmarks?
14 active discussions shown