Ty

Tech + Local AI
Own your weights ⚖️
4x GB10 - 4x AMD MI50 - 1x 3090ti

[TheDavidTai @TheDavidTai]

I'm starting to believe that EXL3 should jsut be the default quant format over even NVFP4 just due to the quality preservation and flexibility.

[Ty @tysn_dev]

🫪🤯

[Mustafa. @oprydai]

you don't actually know it, but a homelab will solve many of your problems.
build a homelab.

[Ty @tysn_dev]

@PheonixTech17 Spend that $200 cmon man

[Ty @tysn_dev]

What’s the absolute best model to run across four sparks right now?

[Ty @tysn_dev]

@Denche2005 I heard someone call C sharp, C hashtag

[Ty @tysn_dev]

@Biggest Knew it

[Ty @tysn_dev]

@joshuacirjak Can I sell you mine?

[Ty @tysn_dev]

Okay, the decisions API is actually pretty cool and cheap as nuts.

[Ty @tysn_dev]

@EspenJD Sure is, and the jump in price from H100 to B300 is wild.

[Ty @tysn_dev]

It blows my mind how many normies that are just getting into local AI think that GPT oss is mind blowing. Like, bro, I get that you post trained it and whatever, but the model still sucks.

[Ty @tysn_dev]

So apparently subagents are outdated now

[Tak 🦞 @cherry_mx_reds]

Anthropic is now recommending that companies REMOVE unnecessary subagents from their AI systems.

This is a pretty significant shift in agent architecture, with their team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution.

The emerging rules of thumb:

→ Skills: Teach the agent HOW to do something, including procedures, expertise, and instructions.

→ Tools / MCP: Give the agent the ability to DO things through APIs, actions, and live data.

→ Subagents: Use for parallel work, isolated context, or tasks a single agent struggles with.

The interesting part is progressive disclosure. Instead of spawning specialized agents or stuffing everything into a giant prompt, one agent discovers and loads skills and tools as needed.

Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up.

As models have gotten smarter, improving agent architecture is becoming less about what you can add and more about what you can delete.

https://t.co/DNq60Skkmc

[kitze 🛠️ tinkerer.club @thekitze]

I have absolutely zero interest in video games anymore and I don't think that I ever will. I give zero fucks about GTA 6.

[Yuchen Jin @Yuchenj_UW]

Heard from a neolab friend:

“Even if we fail to train the best model, we locked in GPU capacity early. H100/B200/B300 rental prices have since gone up 2–3x. We can just resell the compute and print money.”

OK, so that’s the business model.

[Ty @tysn_dev]

@LottoLabs @murd_arch Idk man all the 3090’s in my area on FB marketplace are gone now

[Ty @tysn_dev]

@Greg_GL_87 @WescheNex1q Have you seen what they cost now?

[Ty @tysn_dev]

@Tekeee Lmao

[Ty @tysn_dev]

@WesternLensman How does Elon have a body count? What happened?

[Ty @tysn_dev]

@TeksEdge Ouch man, no bueno

[Ty @tysn_dev]

@ronniedevops @Tekeee It just keeps going