Comments by "" (@grokitall) on "Top 4 AI Tools for Engineers that are ACTUALLY useful" video.

  1. while it seems that systems which constrain, analyse and debug your system seem to have some value, the current systems have a fundamental flaw. they are black box systems, which means it is entirely up to you to figure out if their hallucinations make any sense. because most black box ai is statistical in nature, it has a basic flaw which makes it unsuitable for a lot of workloads. it has absolutely no idea what it is talking about, and only produces statistically plausible guesses that it can't explain. even worse, it is susceptible to poisoning the output, as most providers do not bother to filter their training data for quality. this is why code assistants produce such bad code, which means that the people who can get the best use out of them are the ones who need them the least. in fact, due to how they are trained, and how they work, most code assistants are actually automated copyright infringement machines, which is why most open source projects are actively considering a total ban on all ai based contributions. for a lot of use cases, where creativity or liability are involved, black box systems are completely wrong for the job, as they cannot tell you why they suggested something. unfortunately white box systems are currently fairly primitive, rendering them unsuitable as well.
    1