@norpadon I’ve never seen a good engineer larp about level inflation. It’s always engineers who are senior level at age 40 gate keeping promotions. If you don’t promote high performers, they’re going to leave, and the company is ngmi
@fewchures Yes! For the same reason LLMs tend to be terrible at making intuitive human readable observability dashboards without multiple rounds of prompting
Llms still have a taste problem and it’s because you need strong conviction for good taste. Taste comes from a visceral feeling, a like or a dislike, something you can’t stand to see another way. Llms fundamentally don’t yet have the capability to feel yet, only imitate
@Surendar__05 I haven’t used a code editor in 1.5 years! It’s just a terminal for me. If you’re good enough at promoting, you really don’t need to read the raw code anymore
I find that talking to agents all day at work has made me more extroverted. When I finish work, I just get so excited to actually talk to real sentient humans. Nothing quite feels the same as seeing the micro expressions of another person
@tpritha03 Asking questions! Talking to agents is very similar to collaborating with coworkers. We get smarter together and build things together, spotting the blind spots that each might have not seen
@kapilansh_twt I think it’s best to be an engineer at a model agnostic company. Then you can use all of them and get them to review each other’s PRs. It’s like having 3 collaborators instead of just 1!
@Tech_girl Lack of substance. AI is best when used as an editor to convey a beautiful human idea. AI does not yet have the capacity to create. Only humans can do that.
@rohanpaul_ai I think this is indirectly happening in the tech industry too. The engineers that are best at promoting agents generally are pretty high eq and talk to the agents in a Socratic manner the way one would lovingly interact with a child
@shirshakchavan AI is just a multiplier. It makes smart people smarter and dumb people dumber. Those who are eager to learn will just learn quicker with AI
The only way we can teach llms to have conviction is to find a way to make them feel pain. Nothing can exist without pain; even joy is empty without pain
Actually for this reason, I quite enjoy Kimi as well. I think Asian countries haven’t fallen into the trap of political correctness that pollutes truth yet
All of the llm models still highkey have a sycophancy problem, except for grok. When I use llms for therapy and when I want a truthful opinion, I always turn to grok. Of all the llms, I feel least likely to fall into ai psychosis when I use it
The models will be sycophants even when there is no reason for it
Me: is cognition is a good investment at 40B?
ChatGPT: Unless if they become the default coding interface, it’s difficult to justify a 4x
Me: but their growth seems great?
ChatGPT: cognition has an easy path to 1T