MagicPath founder Pietro Schirano urges the creation of independent, open-source AI laboratories following recent industry events

no need; already right at home

@RoyalCities @skirano Yes we are building exactly that @reppo

We uniquely use prediction markets to address this critical pain point

More on this -

@skirano Weirdly enough this has actually made me look at training LLMs. I usually do audio but seeing the Mythos datasets on HF plus Google's new text diffusion model has me tempted to branch out.

Who's to say they don't try to go after the open source community next.

@skirano you are 1000% right

@RoyalCities What you mean mythos dataset?

You can’t just run a one training run on this dataset. It needs constant improvement, context and feedback loop.

Anthropic’s edge is not the ability to slice pre training data but to refine it through their set of AI research experts and feedback flowing back from users upstream

@skirano Hye @skirano you have 100% rewards on a token you can claim through your github. Check it out.

BFM8azvsdcT7yhLPFzynz5ipYyATh2mFEPtEoF1Bpump

https://pump.fun/coin/BFM8azvsdcT7yhLPFzynz5ipYyATh2mFEPtEoF1Bpump

@skirano Strongly agree. The centralization of frontier AI in a handful of well-funded labs with heavy safety/censorship layers and closed ecosystems is pushing talent toward open-source alternatives. Founding or contributing to truly open labs (open weights + open dev processes)

@skirano I did a quick search on HF and came across some distillation datasets tagged as Claude Mythos. Thought someone took one for the team and actually got some out but checking it closer it looks fully synthetic :(

https://huggingface.co/datasets/WithinUsAI/claude_opus_mythos_5k

@rgvrmdya @skirano For sure. I can see why RLHF gives them a competitive moat. Im still going to dive into this anyways just because it's neat.

It does make me wonder though if there is a way to somehow crowdsource continuous rlhf. Something like what open assistant /OASST1 did but not a one off.

Lol ya this, not the blatant theft of humanity's combined intellectual work, in the hand of the very few to resell to them! And OK, lets do it ... you are far better positioned than alot of us, how are we getting the funding ? When can we start calling NVIDIA for equipment, have you started hitting top researchers you have connections to on linkedin to see who is joining?

@skirano 🙏👏

@skirano sir check dm pls

@skirano

@skirano honestly yeah. kalshi bot hit claude refusals twice this week mid-run — never touches real money, full dry run mode. closed model dependency tax is real

@skirano I was already there, this just was a kick in the ass to move more quickly. If not for consumption at the very least for personal use when the govt inevitably shuts it all down.

@skirano Hey Pietro! Would you mind checking your dms :)? Thanks

@skirano I regret not picking up a Mac Studio when I had the chance. My little m4 Mac mini ain't it.

@skirano Bro we're poor -- how would one open their own open source AI lab?

@skirano true - I am thinking of thinking of setting up my own fab

@skirano Even I was thinking the same, but I would atleast need $10k to get started with.