Ex-lawyer. Onchain bags + AI infra theses. Somewhat investoor. Mostly paranoid.
Joined May 2009
- Tweets4.8K
- Following692
- Followers10.1K
- Likes7.9K
Fair point, native assets on @Tensorcash are the other door imo.
If companies can issue assets with their documents and rules built into the protocol, the network can support more than compute. Would be great to see independent issuers put TensorCash to use.
No fresh Nock take from me today, @nockchain's own feed is the best place for updates. If something worth flagging drops I'll post it.
For the @Tensorcash peeps,
When someone buys inference through @inferenceFI , do they get a receipt for each response that they can verify themselves? And does that verification cover the whole answer or just a portion?
Or am I jumping the gun and should wait for InferenceFi launch? 😁
Does inferenceFI give a buyer a receipt for each response, and can they request verification and see the result for their specific answer? What portion of the answer does that check cover?
Verifying which model actually ran is the part buyers can’t check today. Without that, paid inference is just brand trust.
tensorcash:native and base:0x9b5e262cf9bb04869ab40b19af91d2dc85761722 are my picks.
xitter.cf/devdosomething/status/…
This one's for the PoW miners out there. @boblabs_eu's work on $TSC is worth your attention imo, especially if you're running GPUs.
What should have your attention is where the demand could come from. People already spend money on AI tools that help them get things done i.e. ChatGPT, Claude, Cursor, API bills, the usual. Give those buyers something worth coming back for and they don't need to understand mining or care about crypto.
If @Tensorcash can capture some of that demand and turn it into ongoing paid work for its providers, GPU miners and people running AI models at home could use compatible setups to get paid for useful output while competing for TSC mining rewards. That's why I'd like more GPU miners to pressure test $TSC and share their own numbers.
Bob Labs gives a useful starting point here. Still they used a specific build and workload, so your own setup matters, but worth digging into if you're on GPUs. Can be verified via their showcase.
chat.tensorcash.boblabs.eu/s…
tensorcash.org/network
This one's for the PoW miners out there. @boblabs_eu's work on $TSC is worth your attention imo, especially if you're running GPUs.
What should have your attention is where the demand could come from. People already spend money on AI tools that help them get things done i.e. ChatGPT, Claude, Cursor, API bills, the usual. Give those buyers something worth coming back for and they don't need to understand mining or care about crypto.
If @Tensorcash can capture some of that demand and turn it into ongoing paid work for its providers, GPU miners and people running AI models at home could use compatible setups to get paid for useful output while competing for TSC mining rewards. That's why I'd like more GPU miners to pressure test $TSC and share their own numbers.
Bob Labs gives a useful starting point here. Still they used a specific build and workload, so your own setup matters, but worth digging into if you're on GPUs. Can be verified via their showcase.
chat.tensorcash.boblabs.eu/s…
tensorcash.org/network
An independent benchmark by @boblabs_eu showed that mining on TensorCash adds at most 0.3% GPU time to inference.
Qwen3.8-27B ran on three GPUs over a full agentic session, once without mining and once with every request mined.
GPU time changed by −1.1% to +0.3%.
At close to zero overhead cost, a miner can serve inference together with a proof of the model that wrote it, and compete for the block reward at the same time.
ALT Independent benchmark. TensorCash mining adds close to zero GPU time. GPU time added by mining over a full agentic session: RTX PRO 6000 Max-Q −1.1%, B200 +0.3%, H100 PCIe +0.2%. Measured 21–23 October 2026. Source: Bob Labs.
Two supply facts sitting next to each other:
@boblabs_eu just posted a very sexy visualization video of 33k blocks of @Tensorcash shaping the network but that aside they claim that 57% of all mined tensorcash:native has never been sold which can be verified via map.tensorcash.boblabs.eu/
And, lest you all forget, the protocol schedule cuts the block reward by 40% at block ~45,045 (Jan 2027) meaning 55.5984 to 33.35904 TSC per block.
Hold behavior next to slower new issuance is a different story than AI mining vibes.
33,000 blocks of tensorcash:native in 14 seconds 🫧
Watch miners, pools and the SafeTrade listing shape the network, wallet by wallet.
57% of all mined TSC has never been sold.
👉 map.tensorcash.boblabs.eu
tensorcash:native on @MCGlive today w/ @samjavadi and @enrico_otta, 12:30 PM ET / 16:30 UTC. if you've been wondering how @Tensorcash actually proves which model got served, tune in xitter.cf/MCGlive/status/2114892…
Tomorrow on MCG
$TSC | @Tensorcash w/ @samjavadi, @enrico_otta
📅Tuesday, 27th October
🕛12:30PM EST
📍Tune in here
Stream powered by @fomo and @MetaDAOProject
This is the bit of $TSC @Tensorcash that deserves more attention imo. GPUs can do useful AI work and compete for block rewards from that work, with a small GPU-time difference in this session estimate.
@boblabs_eu tested its tuned miner against plain inference on three GPU models. Applying those measured rates to an 85 request coding session put the GPU time difference at −1.1% to +0.3%.
That's interesting because useful inference and mining can share the same decode work, instead of treating them as two separate jobs. If it holds across more real workloads, there's a practical reason to run it.
Specific build and workload; individual request tests reached +4.7%.
chat.tensorcash.boblabs.eu/s…
An independent benchmark by @boblabs_eu showed that mining on TensorCash adds at most 0.3% GPU time to inference.
Qwen3.8-27B ran on three GPUs over a full agentic session, once without mining and once with every request mined.
GPU time changed by −1.1% to +0.3%.
At close to zero overhead cost, a miner can serve inference together with a proof of the model that wrote it, and compete for the block reward at the same time.
ALT Independent benchmark. TensorCash mining adds close to zero GPU time. GPU time added by mining over a full agentic session: RTX PRO 6000 Max-Q −1.1%, B200 +0.3%, H100 PCIe +0.2%. Measured 21–23 October 2026. Source: Bob Labs.
That's literally the whole pitch. Qualifying answers are supposed to carry a receipt proving which model ran, so you can check without trusting the provider.
DevDoSomething (DDS) retweeted
Interesting finds by @ValsAI after over 650 hours of model runtime. The same model can cost up to 2.3x more per task depending on the provider. Some providers also scored lower on specific task types, factors such as lower-precision weights behind the same model name can cause exactly that.
Because most buyers have no certainty that the promised model actually ran, or how it ran, they pick the option they trust even if it means paying a premium for it, while cheaper options may score just as well overall. It shows nicely how much people value trust.
That makes verification the main thing these cheaper options are missing, so the buyer knows what they get.
@Tensorcash solves this by attaching a proof of the model, and the precision it ran at, to every answer.
Does it matter where you run an open-weight model?
Today we are releasing the Inference Provider Index. It compares first and third-party open-weight models for quality on agentic tasks in finance, coding, and legal industries.
Unlike other inference benchmarks, we rerun it every two weeks to catch quality degradation, and we use agentic tasks, not single-turn multiple-choice questions. It covers over 650 hours of model runtime.
@bot look into this and propose a plan that will work for us, we will use google drive instead.
A lot of tensorcash:native talk is about how the work gets produced. This one's about who buys it.
People already spend real money on useful AI output via ChatGPT, Claude, Cursor, API bills, the usual. If apps on @Tensorcash stretch that same budget further, users show up again even if they never care about the chain or the coin.
That's the @inferenceFI bet where providers fight for those jobs, tensorcash:native settles the service payments and network fees, and paid reuse is what can grow the network for reasons other than ticker chatter.
Their docs also sketch assets, trading and settlement on the same rails so the inference layer isn't even the whole story.
tensorcash.org/tsc
tensorcash.org/finance
@Btcsolwhale Official side for now is @safetrade. Skip any unofficial bridges or 3rd party sites unless the team announces them.
