643 karma · joined October 6, 2023
Further the SpaceX IPO was on revenues from AI data centers. This might be a problem.
What? GPT-3.5 was the moment, not a “parlor[sic] trick that would never amount to much”.
GPT-3.5 was a massive step forward. LLMs were suddenly good and produced meaningful content.
I was extremely early to the agentic stage and had it looping and was dealing with the 4k context issues.
The problem right now is the barrier to create automation and use AI has reduced so much people are not deliberately making decisions about how these tools should work. This means the tools primarily focus on happy paths, and only get refined to the point that they do a “good enough job” and then are shipped. The other issue is people who don’t understand what they are building are shipping brittle and fragile solutions but don’t understand how that is an issue because the AI “just works”, however they get upset when it doesn’t scale.
The solution is to embrace AI and Automation, but slow down planning, architecture, error handling and testing.
This means we still reap the efficiencies but don’t ignore reality.
Code happy, Dale
Right now you could run Laravel new app (replace with any new framework) and Claude/Cursor/Codex git commit will claim to of created the code.
LLMs are not great at preserving cultural uniqueness and diversity. Take how “delve” has reentered the lexicon because the human assessors for pre training dialect of English uses “delve” a lot.
There is a lot of benefits to training specifically for a unique culture with unique norms to preserve the culture as we increasingly rely on LLMs.
https://www.scientificamerican.com/article/chatgpt-is-changi...
The issue is people mix up complexity, novelty, repeatability and scale.
Well documented complex problems can easily be solved by LLMs.
Doing the same thing over and over again is easy for an LLM.
Novelty and scale is very hard for an LLM.
Even small novel problems confuse LLMs.
When you start a new code base the LLM smashes through the boilerplate work. Then when it gets to scale it struggles with context rot plus novelty.
I would agree too that as the codebase grows the LLM struggles more and more with generating code. It is probably misaligned incentives, it wants to complete the isolated task without too much context consumed, at the POC it can consume most of the app, by about 30K lines of code it is quite complex code base to navigate.
Basically I used a loop to keep the agent running.
I have seen a few people have the same experience as me where some people are using AI inefficiently in the workplace and creating significantly more more for others instead of using it to reduce the cognitive load on everyone.
D