Shaun Prince

Constantly trying to improve the things I am responsible for, and mentoring or assisting those that I share goals with.

[Shaun Prince @Suparious]

@TTrimoreau @grok how many times has this exact question been posted on X? Is this a bot account? Also what are the rules about content farming on X.

[Shaun Prince @Suparious]

@KaiXCreator install homebrew, install a better terminal software, like Tabby or Ghostty, remove all the shitty built-in apple apps from your shortcut bar. That's a clean start.

[Shaun Prince @Suparious]

@Kappaemme1926 Low reasoning gives shallow surface results. Consider gpt-5-mini instead.

[Shaun Prince @Suparious]

@naturedotcom No, that is what I use Codex for. It is really good at auditing slop from other AI.

[Shaun Prince @Suparious]

@KaiXCreator It's important to use both, as they are meant for different purposes. Gemini has it's place too.

[Shaun Prince @Suparious]

@0xSero I have to audit the results with Codex, but now I don't have to waste my Codex usage for all the heavy lifting.

[Cybernetic Lover @lucasteske]

Every website should default to dark theme.

[Shaun Prince @Suparious]

@haider1 Not only the harness, but the configuration of your harness needs to be optimized for the model you are using.

[Shaun Prince @Suparious]

@TheAhmadOsman You are the reason why nobody can afford any of this hardware anymore.

[Shaun Prince @Suparious]

@antirez 6bit quants minimum, otherwise the solution is garbage.

[Shaun Prince @Suparious]

@kobyjconrad Configure your settings. It will be more productive than complaining about it.

[Shaun Prince @Suparious]

@nicdunz Codex connects fine from the ChatGPT app on Android. Absolutely no issues at all with the current setup.

[Shaun Prince @Suparious]

@zekramu @ThePrimeagen Only for the free GitHub Pages hosting for static content. Just one enterprise seat gets you unlimited website hosting. Everything else I use a local Gitea instance.

[Shaun Prince @Suparious]

@DavidOndrej1 Tmux has no mouse support and requires you to use a codex to use the keyboard shortcuts. It's better to use a terminal software like Tabby or Ghostty, and launch your agents using the screen command. You still get the multi-windows, but now you also get to use the mouse.

[Shaun Prince @Suparious]

@jedisct1 This is a garbage 2bit quant that you ran on 128GB of Apple Silicon unified memory. It is not possible for this to perform better than deepseek-v4-flash, and even if it does, the cost of your apple hardware will never ROI compared to the API pricing. Also slow as fuck.

[Shaun Prince @Suparious]

@thdxr Privacy should not be a reason either. Most flagship API's will explicitly state in their terms of service that they will not use your data for training.
So if not cost or privacy, what is the obsession of running weak local models?

[Shaun Prince @Suparious]

@Teknium @_vmlops I was gonna say. Hermes Agent can already do all of this already, you just have to ask it. If it fails, you iterate and improve. All it takes is patience, perseverance and / or tenacity.

[Shaun Prince @Suparious]

@antirez You should use ValKey, not Redis.

[Shaun Prince @Suparious]

@bridgemindai @bridgebench Seems better than any sort of local LLM that people can host, and R.O.I. from using the API vs. buying a cluster of NVIDIA Sparks, is phenominal.

[Shaun Prince @Suparious]

@sunnypause @bridgebench What model is it using? Cursor runs other peoples models under the hood, so that's why we don't benchmark it.