Bittensor Interview with Jacob Steeves (Const): OUTsider INsights

When the founder of a $3 billion decentralized AI network sits down for a text interview, every sentence carries weight. Jacob Steeves, co-founder of Bittensor, did exactly that in a recent conversation with Tao Outsider for the 10th edition of the OUTsider INsights series. The result is a compact, unfiltered look at where Bittensor stands today and where the founder’s mind is focused next.

This article is based on Tao Outsider’s OUTsider INsights #10. We expand on the topics Jacob Steeves touched on and add ecosystem context for readers who want the full picture. Full credit for the original interview goes to Tao Outsider.

Templar Changed Everything

The first question Tao Outsider asked was simple: why is everyone suddenly watching Bittensor? Const pointed straight at Templar (SN3).

I think that the Templar run played a massive role in opening up the eyeballs and unleashing the self-reinforcing flywheel of attention

He said, giving full credit to Sam Dair and the Covenant team.

That run is Covenant-72B, a large language model with 72 billion parameters trained on roughly 1.1 trillion tokens. What makes it historic is how it was built. Over 70 independent nodes contributed to the training through Templar (SN3) on the Bittensor network. No centralized compute cluster was involved. The model scored 67.1 on the MMLU benchmark, outperforming Meta’s Llama-2-70B (65.7) under identical zero-shot evaluation conditions.

The market noticed. Templar’s alpha token surged more than 430% in 30 days following the announcement on March 10, 2026. NVIDIA CEO Jensen Huang referenced the achievement on the All-In Podcast, calling Bittensor “a modern version of Folding@home.” Anthropic co-founder Jack Clark also highlighted Covenant-72B in Issue 449 of ImportAI, his weekly newsletter, describing it as an important technology to track.

Bittensor’s founder was right to call it a self-reinforcing flywheel. The Templar breakthrough triggered a wave of attention that lifted the entire ecosystem. As a result, Bittensor’s top 10 subnets reached a combined valuation of approximately $550 million at the time, and TAO itself surged above $330.

How to Start Exploring Subnets

Tao Outsider raised an important data point during the interview. Around 18% of the available TAO supply is currently allocated across the network’s 128 active subnets. The rest sits in Root or on centralized exchanges. So how should newcomers take their first step? Const Bittensor’s founder gave a characteristically direct answer:

Try to mine one with an agent. See if it sucks or if the agent is smarter than the subnet.

That response reflects the hands-on philosophy behind the entire network. Bittensor is designed as a system where miners compete to produce valuable AI outputs and validators evaluate their quality. The best way to understand whether a subnet delivers real value is to interact with it directly.

For those who prefer a less technical starting point, platforms like SimplyTao let users buy, swap, and explore subnet alpha tokens without needing to set up a wallet.

From Google to the Frontlines of Decentralized AI

One of the most compelling threads in the interview touches on Jacob’s personal history. Before Bittensor existed, he was a software engineer at Google in 2017 when the now-legendary paper “Attention Is All You Need” introduced the transformer architecture that would ignite the modern AI race.

“And I met a lot of the people writing that paper too! Geniuses,” Const recalled during the interview.

He left anyway. And people around him thought it was unrealistic. “I never used to shut up about it and they thought I was crazy,” he admitted.

Jacob Steeves studied Mathematics and Computer Science at Simon Fraser University before starting his career as a machine learning researcher at Knowm Inc. He joined Google in late 2016 and left by April 2018. He co-founded Bittensor in 2016, overlapping with his time at Google, and eventually relocated to Peru, where he spent years building the network before the world started paying attention.

When Tao Outsider asked whether there was a moment when the rational choice would have been to stop, Const pushed back on the premise entirely.

Naive rationalism would have stopped me from quitting the job

Then he added something that resonated far beyond crypto:

It is irrational to focus your life’s effort on something that doesn’t compel you. Your state of mind is what makes your life successful. To give up on what compels you is to lose already.

Aligned Incentives Over Raw Scale

The conversation also explored the tension between centralized AI labs and decentralized alternatives. Today, companies like Microsoft, Google, Amazon, and Meta collectively invested over $200 billion in AI infrastructure in 2025 alone. Bittensor’s entire annual incentive budget is a fraction of that figure. So has Bittensor actually started changing the trajectory?

Despite that gap, Bittensor’s founder was not concerned about scale.

I believe it is all about aligned incentives. When we get the tech properly aligned things will move exponentially fast. So I am not concerned about the scale difference here. I am concerned about making Bittensor’s subnets perfectly aligned and working.

That philosophy already has a proof point in Covenant-72B. The model was trained without a centralized cluster, without whitelist approval, and without corporate backing. It relied entirely on Bittensor’s incentive structure to coordinate dozens of independent participants across the globe. SparseLoCo, the gradient compression algorithm behind the training run, achieved a compression ratio exceeding 146x, which solved the bandwidth bottleneck that had previously limited decentralized training at scale.

Jacob’s vision extends beyond a single model. Covenant Labs now operates three interconnected subnets: Templar (SN3) for pre-training, Basilica (SN39) for compute services, and Grail (SN81) for reinforcement learning and post-training. Together, they form a vertically integrated stack that covers the full lifecycle of large language model development within a decentralized framework.

The Bigger Picture

Tao Outsider and Jacob also discovered a shared love for Brazil, the Amazon, and the Yawanawa indigenous people before diving into the interview. It is a small detail, but it speaks to something larger. The people building and covering this ecosystem care about sovereignty, access, and systems that distribute power rather than concentrate it.

Const built his network on that principle. The interview with Tao Outsider captures a founder who remains focused on one thing: getting the incentives right.

Read the full original interview by Tao Outsider: OUTsider INsights #10 on X

Want to explore Bittensor subnets? Start on SimplyTao.