PhreeStyle

Fix the money, fix the world: ₿
Former software engineer, aspiring investor, IA fam.

Opinions are my own and tend to change over time.

[PhreeStyle @PhreeStyleBTC]

I've spent months researching AI usage, and teaching others how to use these tools in large enterprises that are otherwise drowning in AI hype with zero substance.

This tiny anon acct lets me speak freely about my experiences in these massive companies.

[PhreeStyle @PhreeStyleBTC]

I'm urging Anthropic to STFU.

[unusual_whales @unusual_whales]

BREAKING: Anthropic has urged for a global pause in AI development as artificial-intelligence models are nearing capability to improve without human intervention, per WSJ

[PhreeStyle @PhreeStyleBTC]

My early professional experience came from: "this should be easy", followed by "wow, I had no idea".

AI accelerated this process for me. Simply reading its output taught me better practices and terminology that I now use in my projects.

[ThePrimeagen @ThePrimeagen]

this is a perfect example of understanding the very basics of a problem and having no idea what it takes to create the real thing

little piece of advice: every time i go "oh this should be easy" i have been wrong. Did this for 20 years now. You are wrong, just too ignorant to understand why.

That is ok! Its what keeps people learning and becoming better is the "this should be easy" -> "holy shit, i learned alot"

I would ask questions as oppose to insult then ask for a job, its really not the move you think it is

[PhreeStyle @PhreeStyleBTC]

@billyuchenlin Haven't seen the reset yet, but really looking forward to Grok Build improvements.

[PhreeStyle @PhreeStyleBTC]

Codex usage limits inspired me to really push some of the cheaper alternatives. Observation:

MiMo V2.5 Pro is winning at tasks per dollar, beating:
- MiniMax M3
- Qwen3.7
- Composer 2.5
- Grok Build 0.1
- Kimi K2.6
- GLM-5.1
- DeepSeek V4 Pro

[PhreeStyle @PhreeStyleBTC]

Comparing Qwen3.7 Plus vs MiMo V2.5 Pro:
- Similar pricing and benchmarking scores
- Qwen has a speed advantage

Simple task: write a bash script to download and install an RPM in distrobox. Results:
- Qwen: placeholders everywhere, unusable
- MiMo: flawless but slower

🏆 MiMo

[PhreeStyle @PhreeStyleBTC]

To anyone from the future reading this, yes I am totally shitposting between prompts.

[PhreeStyle @PhreeStyleBTC]

To everyone only just now freaking out over the GitHub pricing changes: you might be part of the problem.

[PhreeStyle @PhreeStyleBTC]

We knew this was coming, but it still hurts.

[PhreeStyle @PhreeStyleBTC]

Composer 2.5 actively cheats its way to finishing tasks, whereas MiniMax M3 seems to respect the intent behind them.

[PhreeStyle @PhreeStyleBTC]

@VisualCap Numbers are way too high for New England

[PhreeStyle @PhreeStyleBTC]

The following CLI harnesses are great:
- Codex
- Grok
- OpenCode

These aren't so great:
- Cursor

[PhreeStyle @PhreeStyleBTC]

@theo If I'm not mistaken, $60 is the limit you get from the $10/mo plan from OpenCode Go.

[MHD @mhdcode]

quick way to spot a BS LLM benchmark:

if gemini is anywhere near the top 5 positions

[PhreeStyle @PhreeStyleBTC]

Backend language preferences for AI appears to be, in order of project complexity:
- Bash
- Python
- Rust

Unless you steer it to something else, these appear to be the ones they choose more confidently.

[PhreeStyle @PhreeStyleBTC]

@skcd42 @_imdawon Had a similar experience, a 20 minute task took nearly 50% of my usage for the month.

[PhreeStyle @PhreeStyleBTC]

Composer 2.5 isn’t happy until your Rust files are >50k lines long. Seriously, why am I having to enforce monolithic file gates into the build process?

[PhreeStyle @PhreeStyleBTC]

Anyone else notice the sudden proliferation of TSV files over CSV files?

[PhreeStyle @PhreeStyleBTC]

Here's my Cursor / Grok Build feedback for the day:

/loop != /goal

I'm actually not sure what /loop is for, other than to throttle usage? The same prompt takes minutes vs hours when using /loop. It doesn't respect the interval, and doesn't even loop half the time.

[Matthew Berman @MatthewBerman]

Show me DeepSWE

[PhreeStyle @PhreeStyleBTC]

This is probably the 3rd feature I've asked for, and they delivered again! Loving this pace! Now about the subscription tiers... 😉

[xAI @xai]

Composer 2.5 is now available inside Grok Build.

Composer 2.5 is a fast, highly intelligent model that excels on long-running tasks and following complex instructions.

[PhreeStyle @PhreeStyleBTC]

People say talking to their coding harnesses is the next great efficiency unlock. Were none of y'all around when Apple launched Siri? It was a gimmick.

I have trouble streaming thoughts to spoken words, so learning to type faster was my efficiency unlock.