Oro is the first startup with a live crypto token to ever get accepted into YC. And it’s on Bittensor. A few months ago, co-founders Shardul and Seth were probably at their lowest point. They’d just been hacked, things were looking rough, and they decided to apply to YC as basically a Hail Mary. To their surprise, Bittensor, their subnet, and what they were trying to build caught some attention and they got accepted. The guys have built one of the most innovative subnet designs on Bittensor, using incentives to create a machine for collecting diverse, high-quality agent traces at scale to build better agentic commerce systems. Now YC can give them the other half of what they need - support with turning all of that into a killer product and company. We’re at a point where basically every investor has spent the last year spitting on crypto tokens. For valid reasons tho because most tokens add absolutely nothing. Oro is a bright spot for the opposite case, with the token being an essential part of the core mission and system. Overall, a huge signal for Bittensor

The first piece I wrote on Bittensor was published at Messari in March 2023. @Nick_Hotz was the only person who reached out after reading it as he already had eyes on the network and wanted to learn more. We've stacked a lot of lore together since then. And the lore stacking shall continue with him now joining Unsupervised Capital as our Head of Risk! Nick understands Bittensor deeply. His background investing at crypto firms taught him how to think through game theory, value capture mechanics, and protocol design in ways that most people can't. Running this fund has felt a lot like building a clan in an MMORPG. You want people around you who've been in the trenches, that you enjoy being around, and who can bring a new skill. Grateful to have Nick joining the clan to help the fund and Bittensor win!

Isabella just launched Connito, a decentralized training network built on a different premise than systems like IOTA and Pluralis. Instead of splitting one training run across workers, Connito splits the model itself into individual experts. Workers bring their own expertise and techniques and compete to improve specific experts, and the winning updates get integrated back into the shared model. The team at Unsupervised has obsessed over decentralized training for the past year, and Connito is one of the most interesting new approaches we've seen. We're investors, and we're excited to watch the subnet's progress.

Two things: - Quasar is a scam - stay away. They’ve been lying since day 1. I was one of the first bittensor investors they pitched and after some diligence, we found out they were lying about their past experience. I told everyone who’d asked me for my opinion afterwards to stay away. - Do your own damn research and stop trusting people Here’s a tweet our team put together on Friday after the novelty search that we were saving for Monday: —————————- There are a lot of big claims Quasar is making that are hard to verify with the current public information available. Given the incredibly difficult work that goes into decentralized model training, we’re hoping the Quasar team can provide more clarity around the research and work they’re doing. Below are some of the areas that require further clarification. First, the 5M context window is a big claim and one of the core differentiators with their model, but none of the benchmarks released by the team evaluate the model on long context abilities. Quasar’s architecture substantially compresses context, meaning that, although the model can hypothetically process 5M tokens, the proof that it can stay coherent requires benchmarking on standard long context evals like RULER and/or NIAH. We’re awaiting the results from these benchmarks before we evaluate the subnet’s progress on this claim. Second, the training process itself is unclear. True decentralized training means using untrusted and non-collocated compute. Was the 120B model that was trained on 7T tokens trained in a centralized setting or in a decentralized setting using the subnet? If it was trained on the subnet, is there any data showing that the training process is fault tolerant and can handle interruptible nodes? Which techniques are the team using to compress the data (e.g., gradients, activations, or both) shared across miners? How many independent compute contributors participated or are participating in the training run? What are the hardware requirements for compute contributors?

Meta building a cloud business to resell excess compute validates that compute owners won't be able to maintain full utilization across the lifecycle of their infra. But vast majority of compute owners dont have the resources to casually build a cloud buiz to monetize excess capacity. Meaning bullish crypto compute marketplaces that can support liquid compute, and decentralized training which is fueled by idle compute