AI code benchmarks lied to us, new video argues

Popular AI code benchmarks have been misleading developers, according to a new video by Theo (t3.gg). The video introduces DeepSwe, a benchmark from datacurve.ai, as a more realistic alternative.