There is still impressive progress in large language models themselves every week. (Just to mention some of the past month: Claude-3.5-Sonnet, Gemma2, Nemetron

There is still impressive progress in large language models themselves every week. (Just to mention some of the past month: Claude-3.5-Sonnet, Gemma2, Nemetron and many more) What is, in general, strange is that noone has really figured out how to do the plumbing. We have the LLM and they can perform almost any text related task, summarize search results or provide complex code snippets based on limited specification. But somehow, the integration into work flows remains cumbersome. - Search somehow seems to be burdened by the inability of the search providers to process entire webpages. Google, despite their search advantage, seems to only be able to process the search snippets instead of summarizing the entire content of websites. Most likely a copyright issue... - Github Copilot is still basically autocomplete or a chat interface where I manually have to copy&paste results. Prompting for changes across multiple files is not really solved. (I know there is cursor, but my experience was quite mixed). - All the hailed agents seem to create a lot of fluff but little actual code beyond what I would get with zero shot prompting. (Just tried a new tool today, which consumed $2.00 in API credits on the first task and left me with a broken codebase). - Nothing that properly addresses slide generation yet? Anthropics new workflow with Artifacts and Projects seems very promising and is a great leap forward. But it cannot natively process diffs or work with multiple soruce file and is therefore limited in total codelength. As other people in this thread already remarked, maybe this is early stage technology that is pushed to commercialization too soon.