The incentives are not aligned.
Nobody gets promoted for fixing these small issues. You’ll get rejected for “lack technical complexity”
The way it used to work was that you launch a big project and move onto another team to work on a new shiny project so you can get promoted. Nobody gets promoted for doing the polish, bug fix, or cleanup work.
@GergelyOrosz This reminds me of 2010 when our bonuses were based on integrating every product into Google+. It resulted in a lot of half ass monstrosities. I would not be surprised if there was a similar mandate to Gemini everything.
@rexsalisbury A lot of this is Claude/AI Agents over-engineering the initial draft. You put together an initial draft that is your voice and ask Claude to polish and it overthinks and keeps adding more.
Eventually, you end with this monstrosity.
Hey @OpenAI , I think you have data leaking across different peoples sessions. Seems like someone’s picture of a Mercedes got leaked into my chat session. 😬
Wow… I’m going down a rabbit hole. All I wanted to do was learn how to solve a Rubik’s cube!
If anyone has any suggestions on prompts, I’ll try it out.
Hey @OpenAI , I think you have data leaking across different peoples sessions. Seems like someone’s picture of a Mercedes got leaked into my chat session. 😬
@sidbid This has got to be the challenge at Anthropic. Because Anthropic engineers don’t feel the cost of these tools, they can build things that burn tokens.
Meanwhile us plebs are seeing our token budget evaporate by the day.
@PriyankaPhatak@clairevo Besides the code start, where do you see opportunities to speed up the response?
35s for first response and 10-15 for subsequent responses with light reads is just way too long.