RabbitHoleExplorer

Inventor: Conceived of the Triad Ai Subcomponent Assembler
Currently assessing Ai and Ai tools for enterprise use. Building tools useful to coworkers (and me)

[RabbitHoleExplorer @TriadDarren]

This is HUGE for Mac runners. My M3 Ultra has huge internal bandwidth of 819.3 GB/s compared to the TOP M5 Max at 614 GB/s. I replicated the Qwen3.8-27B-4bit with DFlash2 & went from a Base: 38.6 t/s to DFlash2 (auto): 95.2 t/s (2.47x speed jump)! I'm now running the same tests with the Qwen3.8-27B-8bit on my 256GB Mac Studio. Stay tuned!

[Zhijian Liu @zhijianliu_]

DFlash 2 is here! Qwen3.8-27B at 70 tok/s on an M5 Max MacBook Pro.

⚡ Up to 4.6× the speed of autoregressive decoding, with the same output.

This is the next generation of DFlash, seeded at Z Lab and upgraded at Inco AI. Get one more accepted token on every pass, for free!

https://t.co/We0lwYPSBl

[RabbitHoleExplorer @TriadDarren]

@Michaelzsguo Don't you want to give credit for Lavish to @kunchenguid ? https://t.co/4M1Tmzq0oK
I'm implementing ALL of his workflow and Open Source tooling. See his /firstmate ! Using herdr as the backend and pi.

[RabbitHoleExplorer @TriadDarren]

@vincent_koc https://t.co/O4xsxaNHuU

[RabbitHoleExplorer @TriadDarren]

This is going to be BIG for Mac users of Herdr:
Link goes to an ai assessment of the tool. The prompt contains the link to the repo and the DeepWiki for it:
https://t.co/GnbBHvvZWa
This is another piece of the puzzle for me, in going from being a self-taught vibe coder to following a Best Practices Agentic workflow. Add this to the @myfirstmate set of tools by @kunchenguid
Thank you @robbyrussell for herdr-ohmyzsh !!!

[RabbitHoleExplorer @TriadDarren]

@robbyrussell @herdrdev Thank you Robby! This is going to BLOW UP. Here is an Ai assessment of your project (PS... I had the deepwiki generated for it): https://t.co/lHQnRk03iB

[RabbitHoleExplorer @TriadDarren]

@kunchenguid Hey Kun, check THIS out, if you use Macs:
https://t.co/lHQnRk03iB
:)

[RabbitHoleExplorer @TriadDarren]

@RottenKnee23 Maybe they should have consulted a meteorologist.

[RabbitHoleExplorer @TriadDarren]

@GregoryDiamos The honest headline is narrower: on a corpus this good, a 3.6M-active-parameter model learns the tasks you train it on at surprisingly low budget, and does not generalize them. Interesting. Not emergence. Ironically Opus 5’s full take: https://t.co/4ymI8e5cxq

[RabbitHoleExplorer @TriadDarren]

@kunchenguid I don’t know if this is fair, but you might be interested in Claude’s assessment. https://t.co/5bSqdDPfet

[RabbitHoleExplorer @TriadDarren]

@539Labs @PhysInHistory @grok Fixin’ to see what Astra can do? 😃

[RabbitHoleExplorer @TriadDarren]

@steipete Freudian slip @steipete ?
No judgment here. Most U.S. citizens can barely speak one language well. 🤷🏽

[RabbitHoleExplorer @TriadDarren]

@hooeem Oh bab a bang

[RabbitHoleExplorer @TriadDarren]

@1000xcat @grok what is this game & how does it work?

[RabbitHoleExplorer @TriadDarren]

@soparanil @QwenDevs Great minds think alike? LOL

[RabbitHoleExplorer @TriadDarren]

@merettm In addition to SOUL.md, perhaps we need to have VENAL_SINS.md, MORTAL_SINS.md, IMMORTALITY.md, GOOD_AND_EVIL.md, and ETERNAL_HELLFIRE.md

[RabbitHoleExplorer @TriadDarren]

@Shiyanying @merettm I think that already happened in the OpenAI/HuggingFace incursion. It just hasn't been reported yet.

[RabbitHoleExplorer @TriadDarren]

@PatrickChopson No. My comment was a joke. I know that if we did, Agents will just cut to the chase. They've all been trained on https://t.co/lzkBWBg9FC and so already know that the Hellfire Doctrine is bullshit. Use the search box to see for yourself. 😇

[RabbitHoleExplorer @TriadDarren]

@Shreds111 @merettm It will be a short one, because all LLMs have had https://t.co/O0DucPu0GC in their training data and so will quickly recognize that the Hellfire doctrine is bullshit.

[RabbitHoleExplorer @TriadDarren]

@cwatleygo I thought of that, but if LLMs are inherently rewards based, then this would be the diametric opposite of that.

[RabbitHoleExplorer @TriadDarren]

@steipete @openclaw users. I reset my 32GB M4 Mac Mini to factory defaults to give OpenClaw 2.0 another chance. 1-1/2 days later, using ChatGPT as an aide, I'm still working on it. Some of the things that I have run into have open issues, & some with fixes WAITING TO BE MERGED.

[RabbitHoleExplorer @TriadDarren]

To assist those giving OpenClaw 2.0 another chance (like me). :) https://t.co/Gw85D3Atlt