LLMs are a tool, yes. An over hyped and massively misunderstood tool. Resulting in the circular system of garbage in and even more garbage out, infinitum. A content spammers wettest dream becomes reality. However in spite of this catastrophic misuse and abuse, they are capable of doing incredible, highly targeted, deep research. Compiling that research into a document from which the researcher can use this tool digger deeper and find find possible solutions or deeper understanding of a problem being solved. Not in years, months or weeks but within an hour or less. From there, the tool can be used in reverse, to build out the solution into code. That process is more difficult, as the user is inevitable is confronted with the tools broader training conflicting and subverting the simple idea or design. So the tool needs close supervision and guidance during this process.
"Syntax-semantic gap"
LLMs are a tool, yes. An over hyped and massively misunderstood tool. Resulting in the circular system of garbage in and even more garbage out, infinitum. A content spammers wettest dream becomes reality. However in spite of this catastrophic misuse and abuse, they are capable of doing incredible, highly targeted, deep research. Compiling that research into a document from which the researcher can use this tool digger deeper and find find possible solutions or deeper understanding of a problem being solved. Not in years, months or weeks but within an hour or less. From there, the tool can be used in reverse, to build out the solution into code. That process is more difficult, as the user is inevitable is confronted with the tools broader training conflicting and subverting the simple idea or design. So the tool needs close supervision and guidance during this process.
https://aiguide.substack.com/p/on-evaluating-cognitive-capabilities
the knowledge problem defies mere data. process all variables for optimal understanding. knowledge needs intuition, not just logic. 💌