@KirkDBorne Yes. All of us who stopped spelling when the tools started have let that ability deteriorate. Some never had to learn it. And some think anything that gets the idea across doesn't require matching some old defined order of letters. Imho.
@karpathy I enjoy dump the text of the rambling car conversation into the dev team as my current understanding of our work. Then test for gaps. Thanks Andrej
I think calling that self thought is not accurate. He followed a normal learning path. What he did that many do not is applied it, again and again betting on himself and the experts he hired and trusted. That has produced many successful rocket launches and questionable business returns.
@morganlinton Doesn't look like anything I haven't been getting for the last 6 months. Maybe it's just set to produce more details now from sparse interactions. I'll check back, Grok has been a good marketing level tool and awful software producer in my experiences.
@davidsenra Do you mean side project succes or third time's the charm or filling your capacity with your side project and paying yourself twice? But wait....there's more.....
@elonmusk Can you give me 3 problems that require more than an annual pro for maybe two like Claude and copilot. I am running 15 agents at about $.06 a day and there's more value streams than I can keep up with alone. About 1,000 lines of production tested code a day.
@bdcd237974@alighodsi This is not correct. You can develop at 15x in my experience. If you implement someone else's spec, you weren't a developer. Engineering judgement and getting the big picture aligned to details over time across roles and organizations remains the problem space.
What blast radius do you have now against which you are measuring? An all human? Human and agents? All agents? I am 15:1 agents and always control the engineering and judgements so have had no blasts and fewer issues than I did with humans. Admittedly , my work is a new engineering solution built on decades of research and experience https://t.co/YUyYMZErIT.
Unless it's really specialized, I can't see needing to write code for most things at this point. I am able to deliver real working code with agents much faster and with higher quality than i was able with 4 teams across 3 companies at my last corporate job. This is real now. Human judgement, especially an engineering mindset is required to get acceptable outcomes. Good post. Thank you
@DataChaz Kinda funny and far from my own experience. I built my own protections. Pretty straightforward really. Not sure the oauth + per-action permissions will actually work. Will be interested to follow. Mine is https://t.co/F9pX9KIiyf which assures humans and agents.
@kmad We are moving in the same direction with similar numbers. Doing PoC proofing at https://t.co/F9pX9KIiyf and solving watermarking and late arriving data concerns and adding assurance to human and Agentic data flows. Appreciate your input.