Joseph Jacks’ Personal Blog

Picture the journey that humanity had to take in pinpointing the speed of light .. fumbling along with a foggy, scratched-up lens—starting with Ole Rømer’s 1676 astronomical observations, a clever but indirect peek at Jupiter’s moons that got us within 25% of the truth, like squinting through a storm. It was “lossy,” full of cosmic noise and guesswork.

Then came the 1800s ground-game upgrades: Armand Fizeau’s spinning toothed wheel in 1849, chopping light over kilometers for a 5% error margin, still hazy from mechanical jitters and air interference.

Léon Foucaul’s 1850 rotating mirror smoothed it to under 1%, like wiping away smudges in a lab.

Albert Michelson’s mountain-spanning refinements by the early 1900s dialed it up to 0.001% accuracy, vacuum-sealing paths to cut distortion.

Centuries of iterative tweaks—better tools, controlled environments, wave theories—finally led to today’s mathematical perfection:

A defined constant at 299,792,458 m/s, so airtight it’s the yardstick for reality itself.

This evolution from rough estimates to lossless precision took over 300 years, showing how much of a slow (you could say fast, depending on your perspective) grind human ingenuity has experienced against nature’s poorly understood fundamental reality.

Now… consider Bittensor’s subnet incentive mechanisms (IM)—from the same lens:

Today, in 2025, Bittensor is akin to those early speed-of-light experiments: Clever but lossy, and very imperfect—scores can be gamed, subjective, or miss nuanced contributions, much like Fizeau’s wheel vibrations—but it’s already worlds ahead, using proof-of-useful-work (Yuma consensus) to grade fuzzy, non-binary tasks like training or compute aggregation.

Dynamic TAO (dTAO), rolled out in February 2025, lets market demand dynamically allocate emissions, a critical upgrade from fixed rewards that tightens accuracy like Foucault’s mirror shift.

Recently, Subnet 62 (Ridges) achieved 80% on SWE-Bench in under 45 days with minimal capital, showing Bittensor incentives unlocking efficiencies orders of magnitude beyond centralized incumbents like Anthropic or OpenAI.

Bittensor’s speed of improvement is vastly superior to science’s centuries-long journey—it’s progressing orders of magnitude faster, fueled by open-source collaboration and attracting a much larger contribution graph than any single company could possibly hope to compete with.

While light-speed measurements had to wait decades for technology advancements (steam power, lasers), Bittensor deploys upgrades in months: EVM compatibility in late 2024 expanded programmability for complex subnets; commit-reveal fixed weight-copying exploits; halvings and SDKs boost scarcity and usability.

Subnets evolve hourly—we now have a long list of SOTA breakthroughs on so many subnets like 14, 3, 64, 62, 17, 4 and many many others.

What took physics 300 years, Bittensor can achieve in 5-10, reaching mathematical precision where incentives are airtight—zero-loss, verifiable via secure proofs.

By 2030-2035, I envision Bittensor as the defined constant of decentralized intelligence: Incentives so precise — all programmable, market-driven value representations will be absorbed by TAO and transformed into pure, open utility.

Bitcoin took a decade to prove that it could create digital scarcity, gain institutional adoption and blessing, be crowned as the best performing financial asset in the last decade (superior to NVIDIA)… and flip Google in Market Capitalization in 50% the time it took Google / Alphabet to reach $2.5 Trillion.

Bittensor will reach Bitcoin’s level of economic success… in half the time it took Bitcoin. And then it will blow past and from there… after that, it’s quite difficult to make predictions about anything.