Mark

GPU Middle Class | Investor | Playlist King

[h100envy @h100envy]

Ex-NVIDIA engineer who built Unsloth explained RL, kernels, reasoning, quantization, and agents in 2 hours 42 minutes - better than $5000 fine-tuning bootcamps.

pick the base model -> write triton kernels for 2x faster fine-tune -> quantize to 4-bit -> run GRPO/DPO -> ship a reasoning model on your single GPU.

That loop is why Unsloth is the default way to fine-tune Llama, Qwen, Gemma, and Phi on hardware you already own.

Unsloth + Triton kernels + 4-bit quantization + GRPO/DPO + single-GPU fine-tuning - that's the stack.

Watch and save it, then fine-tune your first model tonight.

[Alexandr Wang @alexandr_wang]

1/ muse spark 1.1 is an industry-competitive agentic and coding model. across many agentic evals it rivals gpt-5.5 and opus-4.8.

available now through the new meta model api and in meta ai. 🧵

[Tulip King 🌷 @tulipking]

local 1T parameter models are now only $94,000. I don't think we've ever seen a cost curve get pushed to zero as quickly as AI. We live in a prosperity era

[will brown @willccbb]

we raised $130M @ $1B for our series A

>$100M run rate

we’re just getting started

[Porg @PorgimusPrime]

This guy is the fucking grim reaper for ML engineers

[QumulusAI @QumulusAI]

Our fleet of 2,500+ GPUs is live across Kansas City, Philadelphia, and Atlanta. It's a distributed U.S. footprint, deployed and running today, putting real capacity close to the teams that depend on it. And there's even more coming online soon.

[ℏεsam @Hesamation]

guys, DoorDash has its own AI lab.

[Google AI Studio @GoogleAIStudio]

https://t.co/ohyGvAO3w8

[Suhail @Suhail]

Looking for 64 B200s. DM if you know a vendor.

[Amjad Masad @amasad]

Atlanta-based real estate company saved $100k by replacing Salesforce with their Replit-built CRM.

[Sarah @araseb_]

Employee: 2k, 2k, 2k, 2k, 2k

Freelancer: 800, 3k, 0, 6k, 1k, 4k

Founder: 0, 0, 0, 0, 0, 0, 300k

Open Source Maintainer: 0, 0, 0, 0, thanks bro

[Techmeme @Techmeme]

Sources: ElevenLabs held early talks with investors on a secondary share sale for staff that would value the startup at ~$22B; it was valued at $11B in February (Bloomberg)

(Visit Techmeme dot com for the link and full context!)

[Gizem Akdag @gizakdag]

Midjourney 8.2 skin textures!

Fireeee

[will brown @willccbb]

love @dylan522p podcasts bc he just leaks lab secrets by accident as supporting evidence for arguments he’s making. like apparently gpt-4o was around 600b parameters, and openai models are now much sparser than anthropic models

[Philip Kiely @philipkiely]

Local and datacenter inference engineers face different constraints, but we have a lot to learn from one another.

Thanks for the conversation Sero!

[Save Your Cinema @SaveYourCinema]

The Odyssey 4D experience?

[Lukas Ziegler @lukas_m_ziegler]

Finally, someone automating the boring infrastructure stuff! 🫟

Road markings is tough. Nobody cares about them until they're gone. Then suddenly everyone's mad.

But someone has to paint those lines, in traffic, in bad weather, repetitively.

So 10LINES built a robot that just... does it.

Workers don't have to stand in live traffic anymore.

Digital mapping. Data-driven operations. The robot knows what it painted and where.

Quite productive I would say :)

This is the stuff nobody talks about in robotics.

Not sexy like other embodiements, but a robot handling infrastructure work that's dangerous and tedious.

Roads, parking lots, tunnels, all need this.

~~

♻️ Join the weekly robotics newsletter, and never miss any news → https://t.co/GoA3ZuwoPB

[Mark @JacksonHPC]

@TheAhmadOsman Cool. I’m at AIE. Where will this be? Moscone?

[Craig Weiss @craigweiss]

your startup only exists because the ai labs have shown you mercy

[clem 🤗 @ClementDelangue]

A study from @Stanford showed that 71.3% of chatgpt queries could be accurately answered by a local model. I suspect a major part of enterprise AI workloads could be run locally too for free (compared to the massive costs of frontier API cost).

Also, it reduces the risk of these workloads being taken away from you because you own the models instead of renting them - which sounds like a good idea these days haha.

That's why we're introducing the ability for everyone to filter AI models on @huggingface based on your local hardware.

For me, there are 800k+ public models that fit on my M5 24GB and that I can use easily thanks to llamacpp.

Let's go local AI!