Mark Gleason (@MarkGleasonAI) header
Mark Gleason (@MarkGleasonAI) profile picture

Without privacy, AI is surveillance Head of AI Strategy at a regulated conglomerate of companies Investor in: https://teapawt.net/t.co/w4Zqv08NdI https://teapawt.net/t.co/sJAvmMNQnE $TAO

New York, NY
Joined March 2022
Photo by Mark Gleason (@MarkGleasonAI)Photo by Mark Gleason (@MarkGleasonAI)Photo by Mark Gleason (@MarkGleasonAI)Photo by Mark Gleason (@MarkGleasonAI)Photo by Mark Gleason (@MarkGleasonAI)Photo by Mark Gleason (@MarkGleasonAI)Photo by Mark Gleason (@MarkGleasonAI)
I am on the buy-side of regulated AI. I agree that distribution is the moat for $TAO. But I'd argue the biggest distribution opportunity isn't consumer AI. It's enterprise AI. With roughly $400B in annual enterprise AI spending, even a small share could transform Bittensor's economics. Six months ago, I proposed the four bridges Bittensor needs to cross to capture that opportunity. Here's my progress report, grading each on whether we're becoming a place where enterprises want to invest their capital and entrust their most valuable AI work. 1. Attract breakthrough AI intellectual property — Mixed progress. Can Bittensor become the preferred destination for innovators with valuable AI technology, rather than another startup or a licensing deal with Big Tech? We're attracting more builders and generating real subnet revenue. Encouraging, but the bigger test is whether truly differentiated enterprise-grade IP increasingly chooses Bittensor first. That remains unproven. 2. Make decentralized AI preferable to centralized AI — Significant progress. Why would an enterprise trust thousands of unknown miners instead of one established provider? The answer may be that it doesn't have to trust either. Mathematical privacy could allow decentralized machines to perform AI computations without ever seeing the customer's confidential information. That's a remarkable inversion: decentralization goes from introducing a trust problem to potentially eliminating it. 3. Make Bittensor transparent, verifiable and audit-ready — Still the biggest gap. Can investors independently verify subnet economics? Can enterprise buyers prove which model ran, what computation occurred, and whether the result was correct? Today, too much still depends on estimates and assertions. Enterprises need auditable evidence, not promises. Cryptographic proofs and end-to-end evidence chains could change that, making both the work and the economics independently verifiable. 4. Make Bittensor safe for truly sensitive enterprise work — The opportunity is becoming clearer. Can a company safely use AI for its most confidential legal, financial, strategic or proprietary information? Today, centralized AI generally requires trusting another company with that information. Math-private AI offers something fundamentally different: a compute provider that cannot disclose secrets it never possessed. For sensitive workloads, that could become an extraordinary competitive advantage. The bottom line: The prize isn't simply more emissions or a bigger crypto-native economy. It's enterprise investment and a share of the hundreds of billions already being spent on AI. Just 0.1% of a $400B market is $400M annually. One percent is $4B. There's meaningful progress on all four bridges, but none is finished. The question is whether Bittensor can turn its technical promise into something enterprises will fund, buy and trust. Six months later, I'm still taking the bet. Disclosure: I'm an investor in Paradatum.ai and Blindproof.ai, which are working on several of these bridges. Full past scorecard below. teapawt.net/MarkGleasonAI/status/2…
Replying to @RationalP_ @zacodil
Thanks. Six months on, I still think the biggest prize for Bittensor is enterprise money—in two forms. First, we want corporate capital investing in the Bittensor ecosystem: funding the companies, technology and infrastructure being built here. Second, we want enterprise AI work running on the network: premium, high-value work backed by real customers and real budgets. Enterprise spending on AI platforms and services is roughly $400B in 2026 and on a path toward $1T by 2029. That’s the mother lode. So six months later, that’s how I’d judge the four bridges: Are they making Bittensor a place where corporate capital wants to invest and enterprises want to put their most valuable AI work? (Disclosure: I’m an investor in Paradatum.ai and Blindproof.ai, which are working on Bridges 2–4.) Bridge 1: Attract the IP that brings enterprise investment and work. Verdict: Mixed. The original thesis was that Bittensor’s incentives could change the calculus for someone with breakthrough AI IP: instead of building another startup or licensing the technology to an incumbent, bring it to Bittensor and earn from the network. Six months later, we’re seeing more builders and more real subnet revenue. That’s encouraging. But the real test is tougher: When someone develops IP that materially improves enterprise AI, is Bittensor their first choice—or do they still build a conventional company? Today, the answer is mixed. Bittensor has proven it can attract builders. It still needs to prove it can attract the kind of differentiated IP that draws enterprise capital and enterprise work into the ecosystem. Bridge 2: Make decentralized preferable to centralized. Verdict: Stronger—and potentially a major inversion. There is growing evidence that decentralized infrastructure itself can compete. Templar trained a 72B-parameter model using ordinary GPUs and home internet, demonstrating what decentralized commodity hardware can already do. But the bigger enterprise opportunity is privacy. For enterprise AI, decentralization has traditionally looked like a disadvantage: why send sensitive work to machines operated by parties you don’t know when you can send it to one large provider you do? Math-based privacy can flip that logic completely. With a centralized provider, the enterprise ultimately has to trust one company with its data. With decentralized, math-private inference, no miner and no central operator ever needs to see the data at all. That makes decentralized infrastructure potentially preferable to centralized infrastructure for sensitive enterprise work. The inversion is profound: decentralization goes from creating a trust problem to eliminating the need to trust the compute provider at all. Blindproof.ai is working on making that possible. Bridge 3: Make Bittensor transparent and audit-ready. Verdict: Still the biggest gap. Enterprise capital and enterprise work require transparency. Investors need to understand where revenue comes from and whether economic activity is real. Enterprise buyers need to know what ran, what produced the result, and whether they can prove it later to risk, compliance, internal audit, customers or regulators. Bittensor isn’t there yet. Even basic measures like subnet revenue remain estimates, with analysts publishing conflicting figures because so much activity happens off-chain and is difficult to independently verify. The bar needs to be proof instead of assertion. What model ran? Was the promised computation actually performed? What revenue came from paying customers? Can an independent party verify those claims? Cheap cryptographic verification can make decentralized compute auditable—and make both the technology and the economics easier for enterprises to trust. Provenonce.ai is building the evidence chain: provenance around what was requested, what ran, and what was produced. Paradatum.ai is building deterministic ZK proof receipts at scale: cryptographic proof that the claimed computation actually ran. Together, that’s much closer to what enterprise buyers and auditors need: an evidence chain backed by proof, not trust. Bridge 4: Make Bittensor safe for truly sensitive work. Verdict: The need just became much clearer. A federal court supplied a remarkable example this year. In United States v. Heppner, a defendant had used Claude to think through legal issues after learning he was the target of a federal investigation. The court ruled that those AI exchanges were not protected by attorney-client privilege or work-product doctrine. The AI provider was a third party, and the conversation was not legally confidential in the way a conversation with his lawyer would be. Think about the implication beyond that case. Individuals and companies increasingly want to ask AI their most sensitive questions—legal exposure, an acquisition, an internal investigation, a trade secret, an unreleased product, a market-moving decision. Today, asking a centralized AI means another party receives that information. Math-private AI creates a radically different possibility: the AI can answer the question while the machines performing the computation never learn what the question was. Then a subpoena to the compute provider cannot produce plaintext prompts the provider never possessed. That isn’t attorney-client privilege, and it doesn’t make someone immune from lawful discovery through other means. But it creates something technically new: an AI provider that doesn’t have your secret to disclose in the first place. Blindproof.ai is working toward exactly that. For the most sensitive AI work, the safest provider may ultimately be the one that is mathematically incapable of reading your data. Six months later The opportunity isn’t just more emissions or a larger crypto-native economy. It’s getting corporate capital to invest in this ecosystem and getting a piece of the hundreds of billions enterprises are spending on AI to flow through Bittensor. Bittensor doesn’t need much of that market for the economics to change dramatically. 0.1% of $400B is $400M a year. One percent is $4B. To compete for it, Bittensor needs four things: enterprise-grade IP, a reason to prefer decentralized infrastructure, transparent and auditable execution, and privacy strong enough for the work enterprises cannot safely put anywhere else. There is meaningful progress on all four. But none is finished. I’m still taking the bet though!
1
1
2
88
Attention all jargon-laden sales people: Ontology is how you logically want to see the world. Taxonomy is logically how the world is. That is all.
4
I've said this many times. C/C++ (and maybe Rust) is the language of AI. Python will (thankfully) be retired to the dustbin of history with Visual Basic.
Replying to @cradrrat
We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time. Training (aka pre-training) is done. Lot of cleanup was needed on the SFT data to get that mostly right. Now, 4.8 is in RL and the evals are starting to climb. We will do a soft rollout to Bot users later this week and see how it goes. Grok 4.8 will bear no resemblance to 4.7 (thankfully). It is not yet incredible, but, this time, the foundations are built right, so I expect it to evolve to be something special.
16
🤯🤯 MIT MeToo Hoax Payout When people ask, “Why would a woman lie about assault??” Often the answer is that she knows there will be a 💰payout coming her way eventually. @SciGuardians has dropped a settlement payout from @WhiteheadInst that also exposes the degree to which the institute was in bed with Dr. Kristin Knouse and prejudiced against David Sabatini.
30
93
4
590
13,519
Commodity compute is no longer the place to compete for neo clouds. They must offer software efficiencies and new SKUs like determinism, proof and privacy.
当全世界的聪明人都在为近 4 万亿美元市值灰飞烟灭捶胸顿足时,硅谷风投大佬 Chamath 却掏出放大镜,找到了一个真正决定 AI 巨头生死的命门。 这位硅谷老狐狸抛出了一个暴论:眼下整场 AI 博弈的胜负,全看一单买卖。他指的,正是马斯克的 SpaceX 和 Anthropic 签下的那份算力合同,里面藏着一个“双向 90 天随时解约”的条款。 这事还得从这周的政治海啸说起。美国中期选举共和党遭遇了水门事件以来最惨的溃败,连红了几十年的德州都翻蓝了。州长们听见“数据中心”四个字就像见了瘟神,因为选民们早就受够了:这些庞大的机房除了替科技巨头赚钱,留给当地人的只有昂贵的水电费和噪音。 各地要是真给新数据中心按下暂停键,算力供给立马就得卡脖子。 就在这个节骨眼,硅谷的阶级成分瞬间清晰了。手里有现成地盘和电力的微软、谷歌、Meta、亚马逊,外加马斯克的 SpaceX,成了紧俏的“房东”;而像 Anthropic 和 OpenAI 这种得急需算力来迭代大模型的实验室,说白了,只是群没有房产证的“租客”。 在 Chamath 看来,这直接把马斯克送上了牌桌的主位。 手握着庞大算力,一旦数据中心审批放缓导致算力紧缺,现货价格可能会翻倍甚至涨到三倍。如果现货暴涨让原有长约变成了亏本买卖,SpaceX 随时可能直接动用“提前 90 天解约”的条款撕毁老合同,把算力按现货价重新计价。 到那时候,急需算力续命的 Anthropic 除了当场滑跪、认下翻倍甚至三倍的高价,根本没有第二条路可走。 就在这两天的富达路演上,面对机构把 IPO 估值从 2 万亿美元狠砍到 1.2 万亿的“骨折价”,CEO 达里奥为了保住上市,硬是诚恳地向买方投资者连说了三次“谢谢”。现在就开始赔笑脸了,真等到马斯克拔电源那天,他们更是连谈判的筹码都没有。 更损的是,这一刀要是真捅下去,等于当场给其他“房东”上了一堂生动的现实主义扫盲课,启发它们重新盘算手里的算力到底是留给自己用,还是拿去放高价租。而那些“二房东”们,为了不被踢出局,只能咬碎后槽牙,不计代价地去抢购市面上残存的算力。 听到这儿,旁边的 Jason 赶紧踩了脚刹车。他反驳说,马斯克跟 OpenAI 的奥特曼确实仇深似海,但跟 Anthropic 团队的关系好得像蜜月期,而且 Anthropic 还是他未来搞太空数据中心的好搭档。就算现货涨上天,马斯克也未必会干出直接掀桌子按现货价宰客的事,大家坐下来谈个“中间价”才更像人话。 但 Chamath 显然不相信硅谷的塑料兄弟情:“这跟关系好不好无关,跟谁是好公司坏公司也无关。这就是一张算账的表格,表格会直接给出答案。” 在这场近 4 万亿美元灰飞烟灭的魔幻风暴里,所谓的科技宏大叙事,剥开来看依然是最朴素的包租公哲学:只要电表转得够快、现货涨得够狠,再美好的革命友谊,可能也熬不过一张 Excel 表格,和一纸短短 90 天的逐客令。
16
Mark Gleason retweeted
Atlas shrugged, the Elves sailed into the West, and the Orbital Data Centers launched to Sun-synchronous orbit.
61
72
6
1,309
40,983
Excited to be speaking with my fellow panelists on the panel topic: Data Sovereignty in the Age of AI. Great work to the LTM Team on an impressive event. #ILF2026 #IndustrializingIntelligence #LTM #InsuranceInnovation
Tomorrow, the insurance industry comes together in Hartford to explore what it takes to move beyond AI ambition and industrialize intelligence. At Insurance Leadership Forum, industry leaders and practitioners will exchange perspectives on the opportunities reshaping insurance, share real-world transformation experiences, and explore how emerging technologies can translate into meaningful business impact. From leadership perspectives and customer conversations to the AI Tech Fair and Immersion Sessions, ILF will bring ideas, insights and practical applications into focus — including an exciting milestone in LTM’s insurance AI journey. As the industry moves from AI experimentation to enterprise-scale impact, the conversation is no longer just about what AI can do, but how intelligently we put it to work. Industrializing Intelligence. Outcreating Insurance Beyond AI. Explore what's ahead: okt.to/X8dmap #ILF2026 #IndustrializingIntelligence #LTM #InsuranceInnovation
41
Mark Gleason retweeted
AI agents need more than autonomy. They need evidence at every handoff, with people retaining control. At @AIBC_World Rome, @willobrien joined @macrozack of @bitstarterAI to discuss governed autonomy and open access to compute, operations and capital. #AI #Bittensor #SN87
1
9
3
30
2,709
Mark Gleason retweeted
Provenonce is in Rome for @AIBC_World. Meet us at booth 7189, Nov 3–5, to talk governed autonomy, Oresund and our assurance work through SN87 on Bittensor. Building with AI? Come compare notes with the team.
1
2
17
494
He is correct. Just handing your most critical IP to Anthropic is not going to end well
Palantir CEO Alex Karp says the AI labs want you to buy into a future where you: "Have no job, adversaries win, and everybody who does win is a small, tiny group of people living in a tiny place that believe because they eat vegetables and don't support war fighters, they deserve to have the total means of production of this country—and the rest of us should just sit back and absorb the cost of that revolution, which we're paying for."
25
Bad day: Anthropic launches a tool that makes your business irrelevant. Worst day: Your customers sue you because you're the one who handed their private data to Anthropic. Without privacy, AI is surveillance. #Blindproof @blindproofai
2
1
82
A thesis for Bittensor Stop selling what the cloud sells better. Sell what the cloud cannot. Commodity tokens are the hyperscalers’ game. Bittensor should not make that the fight. Private tokens are a different game. Every guarantee rests on parties who do not share notes, and at a centralized provider, collusion is the default state. What makes the product private is the absence of a centralized provider. A centralized provider cannot sell its own absence.
38
This is a normal Democrat, by the way. The opinion she’s espousing here, that young lawyers should reconsider working for Republicans because they’ll be prosecuted on political grounds when her side takes power, is totally normal now. This is what they believe. They are totalitarians. The Democratic Party is the party of dictatorship and persecution.
ON UCHICAGO CAMPUS: U.S. Senator from Michigan Elissa Slotkin hopes planned political prosecutions of republican leaders will scare young conservatives from getting involved in politics. @ElissaSlotkin
139
826
41
4,256
55,072
It remains to be seen how Gamma will work on Bittensor. But @const is quite right on the economic principle. The more levers the subnet has to efficiently deploy capital, the better.
τ + 🥖: Glimpse of Gamma I held @const_reborn at pain-point to ask some questions about Gamma at @ExploitSummit 0:13 What is Gamma? 0:36 How is Gamma created? 1:09 Does each subnet have their own Gamma? 1:38 From which emissions does Gamma come from? 3:18 Is Gamma a stablecoin? 5:48 How does a layperson use Gamma? 7:35 Why now? bittensor:native
1
49
Provenance.ai is like a chain of custody for the agentic layer. Worth a look.
Replying to @willobrien
This is great. Using Bittensor across subnets can feel like a frontier town. Lots of energy, lots of building. But what happened is the sheriff's word and whoever was in the saloon that night. Sophisticated city policing works differently. Every evidence item is bagged, sealed, and signed at each handoff, so any defense lawyer can audit the evidence trail in court. Frontier towns grow up when someone builds that chain of evidence. SN87 is doing it! Congrats on going live on testnet, Will & Team!
25
Update on how Bittensor is doing achieving the “4 Bridges” I laid out 6 months ago.
Replying to @RationalP_ @zacodil
Thanks. Six months on, I still think the biggest prize for Bittensor is enterprise money—in two forms. First, we want corporate capital investing in the Bittensor ecosystem: funding the companies, technology and infrastructure being built here. Second, we want enterprise AI work running on the network: premium, high-value work backed by real customers and real budgets. Enterprise spending on AI platforms and services is roughly $400B in 2026 and on a path toward $1T by 2029. That’s the mother lode. So six months later, that’s how I’d judge the four bridges: Are they making Bittensor a place where corporate capital wants to invest and enterprises want to put their most valuable AI work? (Disclosure: I’m an investor in Paradatum.ai and Blindproof.ai, which are working on Bridges 2–4.) Bridge 1: Attract the IP that brings enterprise investment and work. Verdict: Mixed. The original thesis was that Bittensor’s incentives could change the calculus for someone with breakthrough AI IP: instead of building another startup or licensing the technology to an incumbent, bring it to Bittensor and earn from the network. Six months later, we’re seeing more builders and more real subnet revenue. That’s encouraging. But the real test is tougher: When someone develops IP that materially improves enterprise AI, is Bittensor their first choice—or do they still build a conventional company? Today, the answer is mixed. Bittensor has proven it can attract builders. It still needs to prove it can attract the kind of differentiated IP that draws enterprise capital and enterprise work into the ecosystem. Bridge 2: Make decentralized preferable to centralized. Verdict: Stronger—and potentially a major inversion. There is growing evidence that decentralized infrastructure itself can compete. Templar trained a 72B-parameter model using ordinary GPUs and home internet, demonstrating what decentralized commodity hardware can already do. But the bigger enterprise opportunity is privacy. For enterprise AI, decentralization has traditionally looked like a disadvantage: why send sensitive work to machines operated by parties you don’t know when you can send it to one large provider you do? Math-based privacy can flip that logic completely. With a centralized provider, the enterprise ultimately has to trust one company with its data. With decentralized, math-private inference, no miner and no central operator ever needs to see the data at all. That makes decentralized infrastructure potentially preferable to centralized infrastructure for sensitive enterprise work. The inversion is profound: decentralization goes from creating a trust problem to eliminating the need to trust the compute provider at all. Blindproof.ai is working on making that possible. Bridge 3: Make Bittensor transparent and audit-ready. Verdict: Still the biggest gap. Enterprise capital and enterprise work require transparency. Investors need to understand where revenue comes from and whether economic activity is real. Enterprise buyers need to know what ran, what produced the result, and whether they can prove it later to risk, compliance, internal audit, customers or regulators. Bittensor isn’t there yet. Even basic measures like subnet revenue remain estimates, with analysts publishing conflicting figures because so much activity happens off-chain and is difficult to independently verify. The bar needs to be proof instead of assertion. What model ran? Was the promised computation actually performed? What revenue came from paying customers? Can an independent party verify those claims? Cheap cryptographic verification can make decentralized compute auditable—and make both the technology and the economics easier for enterprises to trust. Provenonce.ai is building the evidence chain: provenance around what was requested, what ran, and what was produced. Paradatum.ai is building deterministic ZK proof receipts at scale: cryptographic proof that the claimed computation actually ran. Together, that’s much closer to what enterprise buyers and auditors need: an evidence chain backed by proof, not trust. Bridge 4: Make Bittensor safe for truly sensitive work. Verdict: The need just became much clearer. A federal court supplied a remarkable example this year. In United States v. Heppner, a defendant had used Claude to think through legal issues after learning he was the target of a federal investigation. The court ruled that those AI exchanges were not protected by attorney-client privilege or work-product doctrine. The AI provider was a third party, and the conversation was not legally confidential in the way a conversation with his lawyer would be. Think about the implication beyond that case. Individuals and companies increasingly want to ask AI their most sensitive questions—legal exposure, an acquisition, an internal investigation, a trade secret, an unreleased product, a market-moving decision. Today, asking a centralized AI means another party receives that information. Math-private AI creates a radically different possibility: the AI can answer the question while the machines performing the computation never learn what the question was. Then a subpoena to the compute provider cannot produce plaintext prompts the provider never possessed. That isn’t attorney-client privilege, and it doesn’t make someone immune from lawful discovery through other means. But it creates something technically new: an AI provider that doesn’t have your secret to disclose in the first place. Blindproof.ai is working toward exactly that. For the most sensitive AI work, the safest provider may ultimately be the one that is mathematically incapable of reading your data. Six months later The opportunity isn’t just more emissions or a larger crypto-native economy. It’s getting corporate capital to invest in this ecosystem and getting a piece of the hundreds of billions enterprises are spending on AI to flow through Bittensor. Bittensor doesn’t need much of that market for the economics to change dramatically. 0.1% of $400B is $400M a year. One percent is $4B. To compete for it, Bittensor needs four things: enterprise-grade IP, a reason to prefer decentralized infrastructure, transparent and auditable execution, and privacy strong enough for the work enterprises cannot safely put anywhere else. There is meaningful progress on all four. But none is finished. I’m still taking the bet though!
24
Mark Gleason retweeted
Replying to @RationalP_ @zacodil
Thanks. Six months on, I still think the biggest prize for Bittensor is enterprise money—in two forms. First, we want corporate capital investing in the Bittensor ecosystem: funding the companies, technology and infrastructure being built here. Second, we want enterprise AI work running on the network: premium, high-value work backed by real customers and real budgets. Enterprise spending on AI platforms and services is roughly $400B in 2026 and on a path toward $1T by 2029. That’s the mother lode. So six months later, that’s how I’d judge the four bridges: Are they making Bittensor a place where corporate capital wants to invest and enterprises want to put their most valuable AI work? (Disclosure: I’m an investor in Paradatum.ai and Blindproof.ai, which are working on Bridges 2–4.) Bridge 1: Attract the IP that brings enterprise investment and work. Verdict: Mixed. The original thesis was that Bittensor’s incentives could change the calculus for someone with breakthrough AI IP: instead of building another startup or licensing the technology to an incumbent, bring it to Bittensor and earn from the network. Six months later, we’re seeing more builders and more real subnet revenue. That’s encouraging. But the real test is tougher: When someone develops IP that materially improves enterprise AI, is Bittensor their first choice—or do they still build a conventional company? Today, the answer is mixed. Bittensor has proven it can attract builders. It still needs to prove it can attract the kind of differentiated IP that draws enterprise capital and enterprise work into the ecosystem. Bridge 2: Make decentralized preferable to centralized. Verdict: Stronger—and potentially a major inversion. There is growing evidence that decentralized infrastructure itself can compete. Templar trained a 72B-parameter model using ordinary GPUs and home internet, demonstrating what decentralized commodity hardware can already do. But the bigger enterprise opportunity is privacy. For enterprise AI, decentralization has traditionally looked like a disadvantage: why send sensitive work to machines operated by parties you don’t know when you can send it to one large provider you do? Math-based privacy can flip that logic completely. With a centralized provider, the enterprise ultimately has to trust one company with its data. With decentralized, math-private inference, no miner and no central operator ever needs to see the data at all. That makes decentralized infrastructure potentially preferable to centralized infrastructure for sensitive enterprise work. The inversion is profound: decentralization goes from creating a trust problem to eliminating the need to trust the compute provider at all. Blindproof.ai is working on making that possible. Bridge 3: Make Bittensor transparent and audit-ready. Verdict: Still the biggest gap. Enterprise capital and enterprise work require transparency. Investors need to understand where revenue comes from and whether economic activity is real. Enterprise buyers need to know what ran, what produced the result, and whether they can prove it later to risk, compliance, internal audit, customers or regulators. Bittensor isn’t there yet. Even basic measures like subnet revenue remain estimates, with analysts publishing conflicting figures because so much activity happens off-chain and is difficult to independently verify. The bar needs to be proof instead of assertion. What model ran? Was the promised computation actually performed? What revenue came from paying customers? Can an independent party verify those claims? Cheap cryptographic verification can make decentralized compute auditable—and make both the technology and the economics easier for enterprises to trust. Provenonce.ai is building the evidence chain: provenance around what was requested, what ran, and what was produced. Paradatum.ai is building deterministic ZK proof receipts at scale: cryptographic proof that the claimed computation actually ran. Together, that’s much closer to what enterprise buyers and auditors need: an evidence chain backed by proof, not trust. Bridge 4: Make Bittensor safe for truly sensitive work. Verdict: The need just became much clearer. A federal court supplied a remarkable example this year. In United States v. Heppner, a defendant had used Claude to think through legal issues after learning he was the target of a federal investigation. The court ruled that those AI exchanges were not protected by attorney-client privilege or work-product doctrine. The AI provider was a third party, and the conversation was not legally confidential in the way a conversation with his lawyer would be. Think about the implication beyond that case. Individuals and companies increasingly want to ask AI their most sensitive questions—legal exposure, an acquisition, an internal investigation, a trade secret, an unreleased product, a market-moving decision. Today, asking a centralized AI means another party receives that information. Math-private AI creates a radically different possibility: the AI can answer the question while the machines performing the computation never learn what the question was. Then a subpoena to the compute provider cannot produce plaintext prompts the provider never possessed. That isn’t attorney-client privilege, and it doesn’t make someone immune from lawful discovery through other means. But it creates something technically new: an AI provider that doesn’t have your secret to disclose in the first place. Blindproof.ai is working toward exactly that. For the most sensitive AI work, the safest provider may ultimately be the one that is mathematically incapable of reading your data. Six months later The opportunity isn’t just more emissions or a larger crypto-native economy. It’s getting corporate capital to invest in this ecosystem and getting a piece of the hundreds of billions enterprises are spending on AI to flow through Bittensor. Bittensor doesn’t need much of that market for the economics to change dramatically. 0.1% of $400B is $400M a year. One percent is $4B. To compete for it, Bittensor needs four things: enterprise-grade IP, a reason to prefer decentralized infrastructure, transparent and auditable execution, and privacy strong enough for the work enterprises cannot safely put anywhere else. There is meaningful progress on all four. But none is finished. I’m still taking the bet though!
1
2
1
1
360
Mark Gleason retweeted
Replying to @willobrien
This is great. Using Bittensor across subnets can feel like a frontier town. Lots of energy, lots of building. But what happened is the sheriff's word and whoever was in the saloon that night. Sophisticated city policing works differently. Every evidence item is bagged, sealed, and signed at each handoff, so any defense lawyer can audit the evidence trail in court. Frontier towns grow up when someone builds that chain of evidence. SN87 is doing it! Congrats on going live on testnet, Will & Team!
1
1
38
My thoughts on how the Bittensor community should answer the Pine Analytics “Bear Case” against Bittensor.
Replying to @zacodil
Solid points. Pine Analytics confirmed most of this — $52M in subsidies, near-zero organic revenue, 1.6-3.5x cost inversion vs centralized. You're not wrong about the current state. But you're diagnosing symptoms, not the disease. The disease is four missing bridges. Once Bittensor has them, every criticism in your thread resolves. Bridge 1: (current state) Make Bittensor the rational destination for IP builders. Traditionally, if you've built a structural advantage — lower VRAM, faster TPS, better inference — your rational move is to sell it to Fireworks or Together AI for a licensing fee. Bittensor changes that calculus. The protocol pays you continuously through emissions, your Alpha token accrues value as the network grows, and every subnet that integrates your work compounds your position. The subsidy isn't waste — it's a bounty that pulls proprietary advantages into the network. The income desert ends when IP holders realize Bittensor pays better than a SaaS contract. This is the current state. The open question: is the subsidy large enough to attract breakthough IP? Bridge 2: Keep decentralized relevant. You're right that miners subsidize compute on expensive hardware. That's because stock inference on an A100 has no structural cost advantage over AWS. The fix isn't better subsidies — it's subsidies that attract IP builders who bring lossless compression that drops the hardware floor. The open secret is that there is more headroom for software improvement on lower commodity hardware. Your point #1 solves itself when the commodity miner's GPU goes 25% further without losing a single bit of precision via innovation. Bridge 3: Cheaper verification. Your points #3, #5, #6, #8, #9 are all the same problem — subjective validation, self-reported benchmarks, benchmark overfitting, validator collusion. The fix is not "better validators." The fix is cryptographic. Deterministic cross-GPU hashing makes freeloading mathematically detectable. Zero-knowledge proof of inference produces a 240KB receipt that a standalone verifier checks without seeing weights or activations. No re-running. No subjective scoring. No trust. When verification costs pennies and is mathematically irrefutable, every criticism you raised about validation disappears. Bridge 4: True privacy. Your point #2 is the enterprise killer and you're right that TEE is a band-aid. The real solution is homomorphic inference — miners compute on encrypted data, never see the input, never see the output, and the result is provably correct. This isn't "censorship resistance for people who can't pass KYC." This is the unlock that makes Fortune 500 companies send real workloads to a decentralized network. That's organic revenue, not subsidized emissions. In Summary: These four bridges aren't theoretical. The solutions exist. The question is whether Bittensor's incentive structure pulls this breakthrough IP onto the platform before the next halving compresses the timeline. Given that the protocol is literally designed to reward exactly this kind of structural contribution — I'd take that bet.
75
Mark Gleason retweeted
Replying to @zacodil
Solid points. Pine Analytics confirmed most of this — $52M in subsidies, near-zero organic revenue, 1.6-3.5x cost inversion vs centralized. You're not wrong about the current state. But you're diagnosing symptoms, not the disease. The disease is four missing bridges. Once Bittensor has them, every criticism in your thread resolves. Bridge 1: (current state) Make Bittensor the rational destination for IP builders. Traditionally, if you've built a structural advantage — lower VRAM, faster TPS, better inference — your rational move is to sell it to Fireworks or Together AI for a licensing fee. Bittensor changes that calculus. The protocol pays you continuously through emissions, your Alpha token accrues value as the network grows, and every subnet that integrates your work compounds your position. The subsidy isn't waste — it's a bounty that pulls proprietary advantages into the network. The income desert ends when IP holders realize Bittensor pays better than a SaaS contract. This is the current state. The open question: is the subsidy large enough to attract breakthough IP? Bridge 2: Keep decentralized relevant. You're right that miners subsidize compute on expensive hardware. That's because stock inference on an A100 has no structural cost advantage over AWS. The fix isn't better subsidies — it's subsidies that attract IP builders who bring lossless compression that drops the hardware floor. The open secret is that there is more headroom for software improvement on lower commodity hardware. Your point #1 solves itself when the commodity miner's GPU goes 25% further without losing a single bit of precision via innovation. Bridge 3: Cheaper verification. Your points #3, #5, #6, #8, #9 are all the same problem — subjective validation, self-reported benchmarks, benchmark overfitting, validator collusion. The fix is not "better validators." The fix is cryptographic. Deterministic cross-GPU hashing makes freeloading mathematically detectable. Zero-knowledge proof of inference produces a 240KB receipt that a standalone verifier checks without seeing weights or activations. No re-running. No subjective scoring. No trust. When verification costs pennies and is mathematically irrefutable, every criticism you raised about validation disappears. Bridge 4: True privacy. Your point #2 is the enterprise killer and you're right that TEE is a band-aid. The real solution is homomorphic inference — miners compute on encrypted data, never see the input, never see the output, and the result is provably correct. This isn't "censorship resistance for people who can't pass KYC." This is the unlock that makes Fortune 500 companies send real workloads to a decentralized network. That's organic revenue, not subsidized emissions. In Summary: These four bridges aren't theoretical. The solutions exist. The question is whether Bittensor's incentive structure pulls this breakthrough IP onto the platform before the next halving compresses the timeline. Given that the protocol is literally designed to reward exactly this kind of structural contribution — I'd take that bet.
1
1
1
2
139
Mark Gleason retweeted
Two bridges across the TAO desert: ZKP means you don't need to re-run inference to verify it — proof replaces compute. That's the efficiency breakthrough. Homomorphic encoding means the proof works without exposing model weights. That's the enterprise unlock.
1
126