Dynamix

[Dynamix @wreckitdynamix]

@aldiandyath @thsottiaux possibly, but pro users use 10-20 times as much as plus users, and maybe even more than that

[Dynamix @wreckitdynamix]

@thsottiaux So smoothening the load on your compute is more important than not annoying the hell out of customers by stopping their work mid-task?

[Dynamix @wreckitdynamix]

@thsottiaux the real question is.. what do you mean by "tomorrow"? 23-08 or 24-08?

[Dynamix @wreckitdynamix]

@thsottiaux The goals/tasks it's working on and how far along it is, should be much more clear. Now it keeps shifting from 2/7 completed to 3/7 completed and then suddenly back to 1/5 completed etc. If you're not a programmer, you have no idea how far along code is.

[Dynamix @wreckitdynamix]

@thsottiaux I have a company with ~1700 employees and the IT-department is using Claude. They're saying that Codex is much slower than Claude is right now, so that's why they prefer Claude. Time is of the essence.

[Dynamix @wreckitdynamix]

@thsottiaux @ajambrosino me no understand

[Dynamix @wreckitdynamix]

@thsottiaux - The Chatgpt-app (I use Windows) is the best app for any coding a.i. in the market right now. The remote feature is also useful.
- No 5-hour limit anymore.
- Cheaper than Claude.

Improvements to make:
- the remote feature often doesn't work.
- one button to improve prompts

[Dynamix @wreckitdynamix]

@thsottiaux It needs tool batching. I'm using that in codex via agents.md and it saves me 45% in total costs and it makes codex 27% faster.

[Dynamix @wreckitdynamix]

@thsottiaux There should be a button in every chat window so that users can have their prompt automatically made much better by the a.i., also based on the context of the conversation, with a short explanation of what the a.i has improved in the prompt and why.

[Dynamix @wreckitdynamix]

@thsottiaux There should be just one model and the thinking effort should be determined by the a.i./harness itself for every task. Sub-agent use should always be maximized when it is more token-efficient than not doing so. This saves time and frustration when picking the wrong model/effort.

[Dynamix @wreckitdynamix]

@thsottiaux @ilanbigio @sandersted Okay, nice, but wasn't it obvious too you guys that 5.6 Sol wasn't remembering anything it did before? This seems a pretty crucial basic thing to know.

[Dynamix @wreckitdynamix]

@elonmusk Maybe explain what SFT & RL means. Anyway, an update to 4.6/4.7 is really welcome because my work with 4.5 needs to be corrected by Chatgpt very often.

[Dynamix @wreckitdynamix]

@RobHeinze @elonmusk Bro, there's been a current promotion going on for a while where Supergrok Hevay is $99 a month for 3 months..

[Dynamix @wreckitdynamix]

@thsottiaux challenge accepted!

[Dynamix @wreckitdynamix]

@FurkanGozukara @thsottiaux I use Sol ultra and it only uses 1% per hour roughly, so perhaps you're doing something wrong. Although I have optimized for token usage, 4% per hour is too much.

[Dynamix @wreckitdynamix]

@elonmusk @xai Please develop your own powerful AI-GPUs for this and then release them so that people can play video games with a GPU that's reasonably prices and more powerful than the GPUs that now get 24 fps in Cyberpunk without any upscaling

[Dynamix @wreckitdynamix]

@SAldwais @MKBHD That’s why Tesla share went up by 50% after Trump won the election

[Dynamix @wreckitdynamix]

@therabbithole Sad but good, because this is nature giving us feedback that we are doing something very wrong and need to change our ways asap. If not, we will perish as a species and it will be due to our own doing