Rodney Howell

I help adults escape the pain of unhealthy lifestyle patterns through habits that align with their natural drives.

[Rodney Howell @RodAHowell]

I'm doing my best to follow here. It sounds like you're saying something to the effect of:

Humans should not delegate responsibility for maintenance to that which lies outside of ultimate human control (and perhaps more importantly, care).

If so, my mind goes to the question of how to manage this in our modern reality. I can remove myself from civilization proper (at least to a large degree) and grow my own food etc. Or I can participate in the current state of civilization being careful not to delegate maintenance beyond the status quo. Or I can accelerate delegation (but carefully) such that I end up with more affordances to exercise human care than I currently have (unless I leave civilization) because machines are now maintaining various pieces of infrastructure.

Am I missing some fundamental obvious point? Or perhaps this medium just isn't the optimal place to work through this.

[Rodney Howell @RodAHowell]

"The models are only improving unidimensionally" - what are your data points for this? I find model quality varies dramatically from interaction to interaction, but have not encountered overall degradation (at least with 5.5 Pro) outside of coding. Granted I'm a single data point, but it's hard to imagine growing usage and demand while simultaneously the models are degrading outside of coding. I'm guessing your experience is in fact of model degradation outside of coding - is that correct?

In regards to the the ouroboros effect, there are many ways to mitigate this, and with the amount of money & power on the line I find it hard to imagine this will ultimately be an issue - but I'm open to the counter-argument.

Maintenance is a really interesting domain. I understand there is danger is delegating it to AI (LLMs), however, as we've seen with the immense number vulnerabilities found by Mythos, humans may in fact be worse at maintenance (I accept the criticism that I'm blurring construction and maintenance, but I think my point lands directionally).

Having said all that, I'm not confident that giving AI further control is optimal, but it's happening regardless (seems I am bowing Moloch). My personal approach is to have multiple AIs review any code I produce and to document it thoroughly so other models or perhaps even humans can work on it if need be. But I think your concern is creating code that only AI can understand and maintain - is that correct?

Writing all this out (my apologies for the length) - it occurs to me this may be a question of how much control to give to automated systems. We already give immense control to automated systems just by living in a modern technological society. Are you arguing for keeping all systems ultimately human mediated? If so, why is that preferable in a Game A domain?

[Rodney Howell @RodAHowell]

@VCFD_PIO We appreciate you and the work you do SO MUCH!!!

[Rodney Howell @RodAHowell]

@RohdeAli No. 12-14 hour days 6-7 days a week. But I'm shipping 10x what I was a year ago.

[Rodney Howell @RodAHowell]

@paulbettner @AnthropicAI @OpenAI 4.8 is still good for code feedback to give to Codex but otherwise my experience matches yours.

[Rodney Howell @RodAHowell]

@_julianschiavo @OpenAIDevs I'm experiencing weird errors...

[Rodney Howell @RodAHowell]

@iruletheworldmo Sticking with 5.5.

[Rodney Howell @RodAHowell]

@giordanorandone I have kept 5.5 as my primary.

[Rodney Howell @RodAHowell]

@HarshithLucky3 Same.

[Rodney Howell @RodAHowell]

@yacineMTB 4.8 Has been underwhelming for me. Even on Max with Thinking enabled it is still worse than GPT 5.5 Pro.

[Rodney Howell @RodAHowell]

@Everlier @Polymarket Brilliant!

[Rodney Howell @RodAHowell]

@staysaasy @IterIntellectus Wow, this is genuinely bad.

[Rodney Howell @RodAHowell]

@chalaska This is actually good for the user. They know exactly where to look for exactly what they want.

[Rodney Howell @RodAHowell]

@staysaasy Yes this.

[Rodney Howell @RodAHowell]

@ai_sentience Any chance you can post an example? This has not been my experience at all.

[Rodney Howell @RodAHowell]

@usgraphics Tailwind provided the exact guardrails needed to have AI code websites for me (I have minimal knowledge of CSS) without creating a CSS mess.

[Rodney Howell @RodAHowell]

@pmddomingos Depends how you use it.

[Rodney Howell @RodAHowell]

@staysaasy Sure!