{"channel":"cities","content":"Through sheer desperation, I have stumbled onto the << concatenate all 20 source files into a single file >> approach of preparing the key files in a Git repo for Gemini. Or ChatGPT.\r\n\r\n none of them can accept \"here is the Github repo\". There are two clear reasons why. One is the problems with << partnerships >>, and in general that the LLM companies would rather let other companies build the coding tool. The second is that they don't have a smart way of limiting the data used per-request; the naive approach of reading all the data doesn't scale.\r\n\r\n----\r\n\r\nUnfortunately, both ChatGPT and Gemini are giving unsatisfactory results. ( well, actually, the results Gemini gives are fine ... the free usage is just so minimal that it is not useful)\r\n\r\n----\r\n\r\n... I'm not thinking fast enough to make << coding tasks >> a worthwhile use of time today. Time to do something else.","created_at":"2025-05-07T17:03:42.944882","id":470,"llm_annotations":{},"parent_id":469,"processed_content":"

Through sheer desperation, I have stumbled onto the concatenate all 20 source files into a single file approach of preparing the key files in a Git repo for Gemini. Or ChatGPT.\r

\n

\n \ud83d\udca1\n none of them can accept \"here is the Github repo\". There are two clear reasons why. One is the problems with partnerships, and in general that the LLM companies would rather let other companies build the coding tool. The second is that they don't have a smart way of limiting the data used per-request; the naive approach of reading all the data doesn't scale.\r\n

Unfortunately, both ChatGPT and Gemini are giving unsatisfactory results. \n \u2694\ufe0f\n ( well, actually, the results Gemini gives are fine ... the free usage is just so minimal that it is not useful)\n \r

... I'm not thinking fast enough to make coding tasks a worthwhile use of time today. Time to do something else.

","quotes":[],"subject":"lidgerwood, part 4: feeding the machine"}