@DrSouthern@olsenbdnr@elonmusk Wtf do you want him to say? "Bullshit me, give shallow answers you claimed to have expertise in, and communicate poorly."
@scaling01 Minimax M2.7 is by far the worst reward hacker I've ever used. I have no clue how it scored so well here. It literally deleted tests and replaced them with comments about the scenario they were supposed to cover before.
@misraetel Probably, but when you are ahead it's advantageous to freeze everything so you remain there. Anthropic sucks. They're a bunch of rent seeking moochers.
In the TUI I have two main issues (I love tui for coding, IDK why):
1. No diff on edits is shown. Just the "Edit" command.
2. 3.5 very frequently just blows past "stop" points in workflow where it's supposed to wait for human approval.
3. Anthropic models doing seem to know they can set Schedule to wait for things and will just keep going chattering on and doing random shit while commands run.
@thegenioo X has a really strong UI design aesthetic. Even compared to other apps, it's significantly better IMO. It feels more tailored and unique, while others feel like generic, corporate, cookie-cutter designs. The Grok Imagine UI & the little agent orbs are good examples of this.
I hope that AI accelerates quickly enough that it takes over everything so that we're no longer beholden to emotional people that can be easily convinced of OPEC style cartels for AI
@PrismML Any chance for a model optimized for Vulkan? I converted this to gguf to run with stable-diffusion.cpp but it melted my razr fold when I offloaded to GPU ๐
@scaling01 Lisan, Anthropic has IPO valuation, regulatory, and reputational incentives that are served by publishing dramatic articles about how powerful Claude is becoming. wdyt?
@billyuchenlin TUI kicks ass. It's the best one I have used.
I have used: Claude Code, Codex, Cursor, Factory Droid, ForgeCode, Antigravity, and Gemini CLI.
Very excited about the new models X+Cursor are cooking up.