Mathis

founder @CortexLM

[Mathis @echo_cortex]

@airesearch12 @decisionmodels_ great job, Florian! I was able to benchmark jev in the past thanks to your jevbench project.

are you planning to focus solely on hosting, or do you have training on the roadmap as well?

[Mathis @echo_cortex]

@MuffinDannyH It’s fixed in the new version 🙏

[Jon Durbin @jon_durbin]

Dashboard is a quick work in progress, but for visibility into a run here ya go!
https://t.co/9xg8LqJmgy

This is around $11/b tokens, insane actually. MFU also insane. The whole thing, pretty legendary, and inference...

Using a few nodes from @lium_io also!

[Cortex @CortexLM]

Cortex is becoming a hub for subnets. More subnets will be added over the next few days.

[Mathis @echo_cortex]

@ahmedjama1988 @CryptoZPunisher @cortex_100 Yes, it could. Integration with other subnets can be very easy on our applications.

[Score @webuildscore]

John Wick scenes are one of the hardest to track. we ran our subnet vision models on one of them.

Shot at night, barely any light. The camera whipping around, cars sliding sideways, people cutting through the gaps, all of it smeared with motion blur. this is the footage most models fall apart on.

Ours held it. every car and every person detected, outlined, and tracked, each one carried under its own ID from the moment it enters frame to the moment it leaves. thousands of unique tracks across the clip, through the dark and the chaos, without losing the thread.

And the model doing it is smaller than you would ever guess. no giant model, no cloud.

[Mathis @echo_cortex]

@webuildscore woowww

[Mathis @echo_cortex]

@Basar_ai @cortex_100 A complete, general-purpose application that lets you code, chat with a bot or multiple bots, and much more. The holiday trip was just a simple demo.

[Mathis @echo_cortex]

@Basar_ai @cortex_100 LLM and image models. We take open-source models and improve them through miners.

[Mathis @echo_cortex]

@Basar_ai @cortex_100 There’s actually every reason to create a subnet specifically for the application.
With the right mechanism, we can incentivize miners to improve the model we serve. The better the model performs, the more customers and popularity we’ll attract.

[Mathis @echo_cortex]

@liChat10 @brent_rc @cortex_100 we have over 300k alpha locked in conviction

[Mathis @echo_cortex]

@TalkingTensor @jon_durbin I like the competition, but yeah, I’m not really a fan of Algod’s approach either. Also, the prices are extremely low, while the leading inference providers are already operating on very thin margins. We’ll have to see how sustainable that is over time, but for now, Chutes…

[Mathis @echo_cortex]

@designerbiru @base_bittensor we need you :) we’ll need your feedback for Ink, and you’ll be able to earn rewards through bounty challenges just like a normal miner

[Mathis @echo_cortex]

@affine_io The project looks interesting. I can’t wait to test it.

[Mathis @echo_cortex]

@mccrinbc @AlgodTrading @SubnetSummerT @engyai it’s not necessarily negative; many people are passionate about their work

[Mathis @echo_cortex]

@jon_durbin that’s exactly what Bittensor needs

[Mathis @echo_cortex]

@totheagi And how many tok/s in total? 80 GPUs is still a huge amount of compute. Is that really viable?

[Mathis @echo_cortex]

For example, if Chutes invested as heavily in research as Fireworks or Baseten, it could achieve far more advanced performance and optimizations, attract more customers, and ultimately generate higher revenue, especially since performance is a top priority for customers deploying open source models. Revenue could also increase because each server node connected to Chutes would achieve higher token throughput. With effective marketing, Chutes could gain significantly more visibility, just as Baseten did when it announced its performance results for GLM 5.2.

[Mathis @echo_cortex]

People will understand one day, once a subnet has proven it, that research ultimately leads to revenue. Research may not generate immediate returns, but it creates the breakthroughs and technological advantages that lead to far greater long-term revenue than subnets focused solely on short-term profits. We will never surpass frontier labs if we obsess over immediate revenue while they prioritize research and long-term innovation.

[Mathis @echo_cortex]

@TroyQuasar @TAOFlows @QuasarModels do you have any papers on this? I’d be interested in learning more about your experiments.