Your 6-bit format is a launch decision.
Meta moved 80.8% of ads ranker weights to native MXFP6: 1.99x GEMM, 0.01% NE, 6/6 tasks.
Did your eval rerun at 6-bit?
Old harness green measures itself.
articles.phantom-byte.com/na…
68
Replying to @JenLawrence21
It was just a hot “software update”, settle down, Jennifer.
The firetrucks were just a precautionary measure because all the other “non-cyber” but AI-related cyber situations.. 😉😉
31
Don’t worry, everyone. It was just a “software update” that accidentally triggered the fire department to respond to CIA headquarters in Langley, Virginia.
Nothing to see here. Just reboot the building and try again. 😂🔥
144
Replying to @badlogicgames
Yet everyone keeps doing it. I am glad I am not the only person to think this every time I see one of these 30-second video game "tests."
177
I use my agents as employees. Best I've ever had.
To build an AI workforce: identity, permissions, persistent state, tool access, audit trails, and human escalation.
Xai just assigned five Grok Bots full-time accounting roles on order-to-cash.
The era isn't coming. It's here.
1
1
120
Replying to @nunomo1 @semichenkko
I like Inkling 2 small; I've been using it daily for certain tasks. It's been great for my work.
1
54
Replying to @RepCasar
Go to bed, Caesar. Nobody asked a guy who knows nothing about AI his opinions on AI. Thanks anyway, bud.
3
Replying to @Teknium
And that's why most of us have his back. He has been a champion of open-source AI.
1
51
All these AI experts are struggling to update software. Maybe it's time to call in the pros. 😂
Less hype, more engineering. phantom-byte.com/about.html
85
Replying to @alex_verem
I use MOSTLY open source models and harnesses.
People on social media pretending to be builders are all chasing the smartest model mainly because they can’t really build anything; all they can do is talk about building online.
6
Replying to @jenzhuscott
I can’t wait for Deepseek V5. For me, Deepseek has been a game changer.
3
379
Your idle Macs are a cluster.
Superfluid (arXiv, Apache 2.0): 12 Mac Studio M3 Ultras served one model at 10,656 tok/sec across 38,563 agent requests, zero failures.
You own the metal. What would you run if per-token billing vanished?
Own the rack. Kill the recurring bill.
69
Replying to @AnthonySabatini
Yep, I agree, he’s either a liar, or he’s incompetent, and we don’t have time for either.
18
I use pi to build up my Hermes agents, and manage them.
I use Pi for all my coding tasks, and Hermes for the agent stuff. I like keeping my pi as light as possible, so I use pi to set automations to run in Hermes.
I also attached my telegram to Pi so I can code on the run.
272
