I used to be on 5.4 high for most of my work. I have switched completely to 5.5 medium now. I would highly recommend trying it out
- 5.5 is significantly more token efficient than 5.4 - the same task takes often a third of the tokens
- because of this, is it also much faster to do the task
- you get high "intelligence" per token even after accounting for token efficiency - 5.5 medium is just under 5.4 pro levels of intelligence (imo). It has found tricky bugs for me that all other models failed at
So overall, ideally you will end up with more intelligent, faster model for slightly cheaper.
I founded a UG and a GmbH in 2024. It took me 3 months total including visits to the notary (who charges a non-insignificant sum for their services).
I did this as a subsidiary for a US company and literally had to email and call people every few days to move the process along (mostly, it was the banks who somehow expected us to be a multi-national company and wanted to charge an arm and a leg just to let us open a bank account. Most banks outright refused us).
When the notary finally filed the paperwork to the court, the court replied after a few weeks with additional clarifications for which we had to go AGAIN to the notary to do the whole song and dance of them chanting at us in German at 1000 words per minute.
Everything took painfully long and delayed investment for while. People have absolutely no idea how painful it is to merely have the incorporated entity available. Then, it takes a few weeks to get your tax ID - this is when you can start employing people / accepting payments etc.
Most people actually using python do not start off in scripts. Usually, I would mess around in IPython / Jupyter for a couple days until I have something I'm happy with. Then I'll "productionize" the project.
tbh this has been a sticking point for me too with uv (though I use it for everything now). I just want to start of a repl with a bunch of stuff installed so I can try out a bunch of stuff. My solution now it to have a ~/tmp dir where I can mess around with all kinds of stuff (not just python) and there I have a uv virtualenv installed with all kinds of packages pre-installed.
RLHF is not the "RL" the parent is posting about. RLHF is specifically human driven reward (subjective, doesn't scale, doesn't improve the model "intelligence", just tweaks behavior) - which is why the labs have started calling it post-training, not RLHF, anymore.
True RL is where you set up an environment where an agent can "discover" solutions to problems by iterating against some kind of verifiable reward AND the entire space of outcomes is theoretically largely explorable by the agent. Maths and Coding are have proven amenable to this type of RL so far.
- 5.5 is significantly more token efficient than 5.4 - the same task takes often a third of the tokens
- because of this, is it also much faster to do the task
- you get high "intelligence" per token even after accounting for token efficiency - 5.5 medium is just under 5.4 pro levels of intelligence (imo). It has found tricky bugs for me that all other models failed at
So overall, ideally you will end up with more intelligent, faster model for slightly cheaper.