John Collins

Pneumonia really slows down progress. Ouch. You never really realize how much energy everything takes, till you have none left to give.

It is just a matter of time before AI mixes and masters music better than I can. Because I am going to train them to.

What if I was building a product that would enhance AI capabilities and didn't even realize?

I'm thankful to see so much positivity about AI advancement on my feed today. All it took was for Elon and Sam to agree with one another. I'd like to think it was coordinated reverse psychology. Good job. Good job.

Remember the dinosaurs? Bet they wish they had AI and defensive measures.

Astra tho. Grok. What a time to be a human.

Astra is learning and QAing pretty fast.

Astra got the DaW usable by AI. Still has some rough edges clearly. Ran out of usage and only had 13 hours to wait (decided not to reset) so wondered if Grok could steer what Astra had built. The good stuff starts around 1:30. Clearly there are still gaps in what exist in the DaW but finding them now? 100X easier.

Astra is learning to use a fully functioning DaW. By the end of this, we will have pro level audio editing, pro level video editing and a fully functional stream service all in one app. And controllable by AI. it is taking me longer than I wanted. But it is approaching.

Astra with 2 resets? Audit everything I will.

The best workflow I have found so far. Everything is Sol Max, and somehow the usage has been fantastic. An overview chat that I start with. I can talk to it like a collaborator. The overview creates research prompts that I hand to GPT pro, and then I bring back to the overview. We save that. The overview then makes a plan based on the research to implement what I am wanting. I like to create a to do list in phases that the overview can follow as we go so it doesn't drift. We then create an implementer chat that codes based on the direction of the overview. When the implementer returns I hand it to the overview and ask the overview to determine if they finished the phase. If yes then we create a tester chat and run the phase through synthetic and headless tests to confirm its working live. Hand the test result to the overview and ask him to hand me a prompt to hand the next chat for the next step. Once we get this workflow going. I tell the overview to set a ten minute timer to wake up and see if the chat working is complete or not. If the chat has completed its task it needs to continue with the workflow but rather than handing me a prompt to hand the next chat it needs to just give the prompt to the next chat himself. Until the entire plan is finished. He is to maintain the workflow as if I was there, besides now he is acting as the in-between for all chats now. Once that is done I create a monitor chat. And tell it not to interfere with the workflow that is going on. But that I want him to create a percentage based return of what work is actually being done. Whether it's planning, testing, compiling... Anything that is being done make a list of it and put a percentage of the time that is being used. Set a timer for 30 minutes and audit and make our percentage list. And then me and this chat will talk about how we can optimize the workflow that exists. What we can cut out. How we can prompt the overview to speed up the process without losing quality. This is the current workflow that seems to work autonomously. Stays on track the best. And allows me to have numbers on what is being done so that I can make small adjustments as we go.

Another snag that could be fixed by OpenAI. When codex is asked to create a new project he was either unable or failed. He named them properly and pinned them instead. If you could give codex chats the ability to create projects and create chats inside the appropriate project that would be nice.

First run through of the new workflow. Research. This is the first area that cannot be automated. Sol Max uses too much usage for me to want to use as my researcher. There is no way for our overview to trigger GPT pro. The work around is that I'll have our overview hand us Separate prompts for each piece of research that needs to be done and the user will have to give that to a GPT pro chat. This saves usage and actually get's better research results from my experience. The fix for OpenAI. Allow codex chats to use GPT chats without the GPT eating using. Just make the back and forth connection.

I wanna talk about workflow. All I hear is usage, usage, usage.

If your idea can be accomplished in a week, your idea is not near big enough. Aim higher.

The smarter AI gets the more likely it is to tell you how you are doing wrong. Isn’t that the exact same with human intelligence.

When codex feels stronger than it has ever been but everyone is complaining. Hmm. I’ve steady worked on my workflow tho. The more intelligence grows the harder it seems to lead. But once you find the correct collaboration style, man can things get done. Audit your workflow.