Pinned Tweet
just had an utmost important thought while working
gm
ga
gn
or just
gman
to greet all everywhere
1
4
315
0xfffaf retweeted
Agents will not replace shopping.
The mistake is viewing all of shopping as an efficiency problem. That’s very Silicon Valley.
Everything is friction, everything gets optimized, and if you removed a step you must have made it better.
It works for dish soap. It misses the entire point of buying a jacket.
For 20 years, ecom has optimized friction. Faster search, checkout, delivery. Amazon won by solving this, and agents are going to finish that job.
You won’t reorder paper towels or compare vacuums. An agent will do it for you. You won’t miss it because there was never anything to miss.
That’s the utility half.
The other half is identity purchases. Clothes, furniture, art, accessories. You buy these to say something about who you are, or who you want to be.
Nobody needs a specific jacket. You buy the one that feels like you. It’s an art of self-definition.
Most people don’t want that handed off. Sure, an agent could pick the jacket for you. It might even pick a great one. But then it’s skipping the part you actually enjoyed.
The hunt wasn’t friction—it was the fun.
Identity purchases require browsing and discovery. They’re driven by imagery, narrative, brand, seeing the thing next to the thing.
People shop for entertainment the way they walk through a museum, not the way they run an errand.
Shopping is two problems. One is about getting the human out of the loop. The other is about serving the human better than a feed or a search bar ever did.
Whoever puts those together wins. Discovery upstream, agents downstream doing the heavy lifting.
You see something, you want it, it’s yours. No cart, no tabs, no search.
The utility half consolidates into a handful of agents owned by whoever has the most context and distribution.
The identity half is where the next great consumer company gets built.
0xfffaf retweeted
Most Proof of Work networks pay miners to perform computations nobody actually uses.
tensorcash:native is the first to completely change that
The same AI computation can secure the network AND serve a paying customer, with almost no additional GPU time in the cited benchmark.
Now imagine scaling that through @inferenceFI
That's where things get REALLY interesting...
Rent the compute. Verify the work.
From cloud computing two decades ago to AI inference today, outsourced compute has always raised the same question: did the provider actually do what you paid for?
@AnthropicAI’s open-source release of FlashProve has brought renewed attention to Provable Inference. I’ve mapped seven verification approaches: ZK proofs, TEEs, re-execution and disputes, black-box fingerprinting, activation fingerprints, algebraic checks, and output evidence with statistical replay. Representative projects are shown in the graphic below.
In my view, @Tensorcash is the coolest architectural innovation in this wave of AI × Crypto because it delivers:
One inference. Two uses. Verifiable receipt included.
The protocol’s low incremental overhead is another compelling feature. For the session workload, ecosystem developer @boblabs_eu’s tuned benchmark estimates a GPU-time change of −1.1% to +0.3% when inference and mining are combined—essentially no additional GPU time.
Benchmark and supporting data:
chat.tensorcash.boblabs.eu/s…
The same inference evidence supports customer verification and can also qualify as work for a native L1, subject to model admission, successful verification and meeting the applicable difficulty target.
The protocol’s inference verification paper:
tensorcash.org/whitepapers/V…
The tensorcash:native protocol brings answer generation, verifiable receipts and L1 consensus into a single architecture. That is what excites me.
Isn’t this the Bitcoin of inference?
#TensorCash #ProvableInference #VerifiableAI #PoUW
0xfffaf retweeted
Cheap verification doesn’t mean cheap proving. That’s what stands out to me here.
Sampled ZK audits help by proving only a small share of requests. Those selected computations still get cryptographic proofs; the rest don’t get individual coverage.
That’s also why I’m watching @Tensorcash . It’s trying a different way to keep overhead down: retain native floating-point inference, record sampling transcripts, then use parallel model replay and statistical checks. In its mining design, the verifier checks a submitted window without repeating the miner’s entire search.
The trade-off matters. Full replay needs access to the model and compute, and statistical acceptance doesn’t give you the same guarantees or weight privacy as ZK.
I want to see how far this route can bring costs down while still catching model swaps and other shortcuts in practice.
For anyone interested in the technical details, here’s TSC’s verification whitepaper:
tensorcash.org/whitepapers/V…
Curious to hear from @sweis and @arasuarun—where do you see sampled ZK audits and statistical model replay each making the most sense?
Rooz says a sandwich needs two things. You have to see the order and you have to be able to pay to get in front of it. Canton removes both with a closed set of super validators.
@TensorCash goes after the same two things from the open side. In the DEX design the order is encrypted and the key is split across random validators so nobody can read it before it is sequenced and the sequence comes from arrival time measured across those validators not from who pays more. Nothing to see so nothing to buy with no sequencer to trust.
Old exchanges cut every cable to the same length so nobody sits closer. This is the same idea but done with cryptography on a permissionless chain. The whitepaper is from April 2025, before the chain even launched.
And here is the part nobody else has it would be the first dex i know of where the commodity is produced, measured and settled by the same chain. Everyone else trading compute takes the price from an outside index but here the chain itself knows how much inference capacity exists bcs making it is the consensus.
It is still in development, months not days but look at the pattern. People who lived inside real market structure do not need MEV explained to them. They start from it and the same people are now building the market for inference on top.
MEV is a design choice so is this.
Hyperliquid of inference $TSC
The interesting part is the buyer: a coding agent.
Agents already spend on inference to get work done.
(let that dig in a second...)
Plug miners into that flow and there's a reason to use tensorcash:native beyond trading it
aka
Bitcoin of inference
A miner on the TensorCash network wrote a website for a coding agent on Qwen3.8-27B, and received 0.01136 TSC for this inference.
Inference and a shot at the block reward, from the same unit of electricity.
A live run by @boblabs_eu, an independent team building on the network.
Cheap inference is only cheap if you get the model you paid for. Otherwise you're comparing honest providers against sellers quietly cutting corners. Making that shortcut detectable is the part of tensorcash:native that matters commercially.
TensorCash: Making AI Inference Part of Mining
What if the compute used for mining could also generate AI answers people actually need?
TensorCash aims to combine both: miners run AI models to provide inference services while competing for block rewards.
But there’s a key question: what stops a miner from secretly using a cheaper model or lowering its precision while charging for the promised service?
According to TensorCash, its approach makes inference come with a verifiable record. The randomness used to select each next token is derived from public information. Verifiers can reproduce those numbers and check the output against the promised model.
Verification happens in three stages:
① A quick CPU check tests whether the record is internally consistent.
② Another CPU check assesses whether the recorded data looks plausible for the promised model.
③ A GPU runs the promised model to check whether the recorded values match at the specified precision.
The network fully checks inference segments that qualify for a block reward. Buyers can randomly audit the answers they purchase. The more a provider cheats, the greater its chance of being caught.
The core idea is to let the same compute provide AI services and help secure the network—while allowing users to verify whether they received inference from the model they paid for.
Original article:
teapawt.net/tensorcash/status/2120…
Curious to hear your thoughts on TensorCash’s approach to inference verification—especially its robustness against cheaper-model substitution, quantization, and cross-GPU numerical differences.
@ddkang @florian_tramer @dawnsongtweets @tri_dao @tqchenml @ilblackdragon
7
267
TEEEE EES SSEEE
BITCOIN OF INFERENCE
HL OF INFERENCE
ETH OF INFERENCE
INFERENCE OF INFERENCE
Everyone wants to sell AI answers with a receipt now, MEGA just put $30M behind it. But the whitepaper lists fingerprinting, TEEs and zk as options and picks none, so the receipt part is still a TBD
Tensorcash $TSC already does it differently, the inference IS the mining. Every 256 tokens come with a proof and verifiers replay the model so a swapped model or fp8 sold as bf16 gets caught, no trusting a provider or a hardware vendor. Ran a full coding session on it last month with 42 proofs attached for 16 cents. Market on top isnt live yet but the hard part already is
teapawt.net/yilongl_megaeth/status…
tensorcash:native , the bitcoin of inference
If we take a step back and look at $BTC...
The genius was in converting electricity into money. To create a coin, the input needed is energy, and nobody can create one or rewrite its history without paying that cost again.
People saying it has no value don’t really know what they’re talking about. The costs of mining and securing the network are where the value of $BTC comes from - if done correctly of course. Maybe a bit too philosophical, but in a nutshell, its fundamental value is arguably higher than that of fiat currencies.
Then AI came. It led to a massive spike in electricity demand, and converted that electricity into intelligence.
Where crypto is converting electricity into money and AI is turning electricity into intelligence, @Tensorcash is combining the two. The energy other chains spend solely on hashing goes into proof of work that creates inference. So the same unit of electricity delivers both at the same time.
so you're telling me tensorcash:native is insanely discounted right now?
So you telling me that InferenceFi is the first marketplace in the world where AI tokens will be listed with proof of exactly which model answered, at which precision, pinned to a block and you are not bullish?
So you telling me people like @enrico_otta and @samjavadi , who come from the very top of tradfi decided to build it on @Tensorcash still the only place i know where provable inference is live today?
Why would they do that?
To be the first of their kind into a market of around 150 billion dollars a year. A commodity that needs a market structure nobody has built yet bcs all of crypto is still stuck on the useful part lol
Even @AnthropicAI 's frontier research needs a 19 MB proof per 1024 tokens while on @Tensorcash it is around 160 KB and it is in every block.🤣
So you either believe a market like this can exist without proof and then you are every sellers perfect client or you just realised where the money is.
Hyperliquid of inference $TSC
So let me get this straight.
There is a place being built to buy AI answers where your money will finally know what it is paying for. You get a receipt that proves which AI did the work. Nowhere else gives you that. @inferencefi
The order books will be filled by miners who make those answers anyway while keeping the chain safe. That part is live today. For them it is a leftover. They can sell at any price, cheaper than web2, with provability on top.
That is how you walk into a 150 billion dollar industry that still has no proper marketplace. Somebody had to build the first one and it is being built by people who built markets for the biggest banks and they already told everyone on discord where the money goes. Customer pays in normal money, miner gets $TSC and revenue buys $TSC. It is right there in the screenshots.
It is not even open yet. And you are not bullish. How is that possible anon???
Hyperliquid of inference is coming for all global inference .
So you telling me that InferenceFi is the first marketplace in the world where AI tokens will be listed with proof of exactly which model answered, at which precision, pinned to a block and you are not bullish?
So you telling me people like @enrico_otta and @samjavadi , who come from the very top of tradfi decided to build it on @Tensorcash still the only place i know where provable inference is live today?
Why would they do that?
To be the first of their kind into a market of around 150 billion dollars a year. A commodity that needs a market structure nobody has built yet bcs all of crypto is still stuck on the useful part lol
Even @AnthropicAI 's frontier research needs a 19 MB proof per 1024 tokens while on @Tensorcash it is around 160 KB and it is in every block.🤣
So you either believe a market like this can exist without proof and then you are every sellers perfect client or you just realised where the money is.
Hyperliquid of inference $TSC
* breathing heavily *
I dont know about you but sometimes it hits me so hard i have to do the breathing thing and walk through it all again from the start just to confirm it is real kek.
We are this early on the only valid bridge between a crypto L1 and tradfi that i have seen i mean it solves every problem tradfi fights with. Custody, proof of what you bought, honest price, a way to hedge and no middleman to hack
and it solves a problem almost nobody knows exists yet bcs they are too busy talking to their AI lmao. Every conversation you have every task you hand your agent and every answer you take at face value. At global scale that is a problem. Who ran this, which model, was it the one I paid for. $TSC solved that before most people noticed it was a question.
That is how bullish this is the bridge is solved and solved by people who know exactly what tradfi needs because they come from that side of the table fucking home turf.
Of course on the way there will be plenty of wannabe TSC projects also driven by huge raises.
Btw we all know inference will be such a big market that there is room for 100 winners but still i see one bcs only $TSC as far as I know secures a block with real inference.
90 to 97% of the GPU time is the proof itself and serving runs at 99 percent of plain vLLM speed measured by an independent verifier. Everything else is smoke and mirrors with a GPU in the middle. Wake up call for everyone on frontruning tradfi with $TSC
teapawt.net/drwconvexity/status/21…
1
3
106
This take is so dumb. Obviously they get paid to just hodl the wallet and hodl it all the way down - look at all the hodling!~
meanwhile their side-wallets are breaking bank but nobody can see the actual extraction in the background
Being early means nothing if you don’t sell
@pointfarmcap round-tripped the hardest, from $13M down to $400K upnL (zero sells)
@theunipcs gave back $8.5M (no sells)
@cryptogle and @the__solstice both went from deep green to red
The round trips are CRAZYY
45
0xfffaf retweeted
This week @samjavadi and I are at the Digital Asset Summit in London, 10-11 Nov.
If you’d like to talk about verified compute on @Tensorcash, the roadblocks to trading inference, and how we’re solving them with @inferenceFI and @AskTillyAI, feel free to DM.
We’re looking forward to meeting you there.
Day one of the Digital Asset Summit in London with @enrico_otta.
Looking forward to connecting with like-minded people on verified compute and inference markets, and with anyone who’d like to know more about what we’re building at @AskTillyAI and @inferenceFI on @Tensorcash.
Come find us.
tensorcash:native miners bring the supply. @inferenceFI is building the market to sell it.
Mining rewards help get GPUs online. Paying customers give them a reason to stay beyond the block reward.
This is the part of the thesis I'm watching.
Read this as a miner not as a tweet.
Bitcoin burns gigawatts on hashes nobody uses. Every proof of work since secured itself with something the world does not need. Elon just named what the world will need and it is Inference almost all of it.
One chain already mines with it. On $TSC a block is won by serving a real answer from a real model with a receipt proving which model wrote it. Independently measured: 90 to 97 percent of the GPU time is the proof itself under 1 percent overhead. SERVING IS THE MINING.
So if 99 percent of compute becomes inference every GPU serving an open model can secure this chain and gets paid twice for one job. Block reward from the chain payment from the customer. The pipe is already proven and first paid session settled on chain last month.
Bitcoin had to convince the world to waste energy. $TSC only has to be there when the world does what @elonmusk says it will do anyway.
The richest man alive just told you which side to stand on. Hyperliquid of inference is coming for all global inference anon ❤️
0xfffaf retweeted
a coding agent asked for a website, and a miner on @Tensorcash wrote it.
that miner got paid in $TSC for the answer. the same run also counted toward the next block.
so the power that's securing the network is doing real ai work too
A miner on the TensorCash network wrote a website for a coding agent on Qwen3.8-27B, and received 0.01136 TSC for this inference.
Inference and a shot at the block reward, from the same unit of electricity.
A live run by @boblabs_eu, an independent team building on the network.




























