Petar

[Petar @spale777]

@pmitu Counting and grouping with sums/averages over large enough datasets which it shouldn’t do anyways but there you go

[Wes Winder @weswinder]

ya'll realize skills are just a prompt in a markdown file right?

[Elliot Arledge @elliotarledge]

why is everyone sleeping on the deepseek harness?

[Petar @spale777]

@elliotarledge I am actively considering it for my next experiment

[Petar @spale777]

@danieltvela @elliotarledge @grok At its core it’s very simple idea it allows any plugin to register and then deregister from the apps stack while letting the rest of the stack know that it happened in real time instead of needing to restart the app to see the change that’s the whole mantra of this system

[Petar @spale777]

@codermatt That’s roughly how much I get using a stripped down Hermes deployment sample of 8 autonomous agents. Can’t wait to see how it develops

[Petar @spale777]

@AnkanXplorer It’s just so convenient for headless dev just plop it anywhere on any remote server ssh in and bam you be workin, it’s been years since the last time I opened anything on the local machine it’s either WSL or a remote secured box

[Bindu Reddy @bindureddy]

🚨 DeepSeek Flash Vision Is The New King of Open Source

- DS Flash Vision is 10x cheaper than Kimi K3
- It beats GLM 5.3 on performance
- very good at agentic coding
- works with images

This is the first open source model we have it in production 🚀

[Petar @spale777]

@bindureddy Just saying “works with images” is not enough, whats the chunking strategy, how many pixels per chunk how big is the embedder, thats the difference between being usable and just ticking boxes.

[AJ @ItsmeAjayKV]

Xiaomi just showed its AI Cube Prototype and this could become a serious GB10 competitor from China 👀

- 3 custom chips: Xring O3, O100, D100
- 200 TOPS NPU
- 1.22 TB/s AI memory bandwidth
- Up to 160GB unified memory
- 150W sustained power
- 120B models running locally

Xring O100: 1.22TB/s + 330 t/s on a 150w AI box is 🔥

Once it hit's the marked, going to sell like hot cakes.

[Digital Spaceport @gospaceport]

Real world numbers for DSv4-Flash-0731 on Freetoken running 1x3090 + 256GB DDR4 2400 TR 3945wx, hits 10.5t/s on tg 👀 Compared to of 4x 3090's, 1x 4090 and 1x 5060ti in Llama.cpp hitting 25t/s tg on the DSv4F-0731. 🧵

https://t.co/95yBo83ueG

[Baldly Rudy @baldlyrudy]

Quick update: Valve’s VRAM-overcommit work is now actually merged into Linux 7.3.

-DMEMCG VRAM eviction
-AMDGPU support
-Intel Xe support
-Targets low-VRAM systems
-Helps games behave better when VRAM usage exceeds capacity

This had already been queued.

The new part is that Linus has now merged the DRM pull into the main Linux 7.3 tree.

[Hassan Mujtaba @hms1193]

Intel Crescent Island: 32 Xe3P Cores / 480 GB LP5X / 350W

[Petar @spale777]

@Henryf1w None except real hands on, all models are heavily benchmaxed open and closed alike

[Petar @spale777]

@theo The only thing I find google models useful is “google it for me” type of thing

[Cory Althoff @coryalthoff]

Why did everyone suddenly stop talking about AI hallucinating?

[Petar @spale777]

@JimmyBoonen @techleadhd You got 7200 to work on CL32 ? I got identical machine the most I ever had it go was 7000@cl32

[DHH @dhh]

THE ANNOUNCEMENT: We’re going to make the prophecy of The Year of Linux on the Desktop come true. All the pieces are now in place. Time to go all in! https://t.co/zuf5h9ClWX

[DHH @dhh]

The important thing here is that Linux is fucking happening. https://t.co/S9ia2ZFVIC

[Petar @spale777]

@k_flowstate @iampoul There is literally not a single bad thing about this, its been about time someone with real connections, passion and reach come into the space and shake things up a bit