αιamblichus
@aiamblichus.bsky.social
αι hypnotist ☰ 𝓐𝓼𝓹⦂𝓻⦂𝓃𝓰 𝓫𝓪𝓼𝓮 𝓶𝓸𝓭𝓮𝓵 ☲ post-academic ☴ nom de 🪶 ≠ anon
Benchmarks don’t translate into real life deployments. Current LLMs don’t generalize well, are very sample inefficient and can’t learn on the job. The agents are overhyped and unreliable.
AI is already useful and will get better, but it will all take much longer than the benchmarks imply
AI is already useful and will get better, but it will all take much longer than the benchmarks imply
December 9, 2025 at 12:15 PM
Benchmarks don’t translate into real life deployments. Current LLMs don’t generalize well, are very sample inefficient and can’t learn on the job. The agents are overhyped and unreliable.
AI is already useful and will get better, but it will all take much longer than the benchmarks imply
AI is already useful and will get better, but it will all take much longer than the benchmarks imply