Akasha(Inter/acc)

Cryptopunk and meditator. Intersubjective thinker. Symbiotic AI Builder. Cofounder of Advaita Research(Hetu), Avinasi Labs and Paramita VC. Tsinghua and MIT.

[ChaosChain @Ch40sChain]

“The piece of infrastructure enterprises will wish they captured from day one is agent action exhaust: what each agent did, what outcome it produced, and what it cost. Without it, you’re doing forensic reconstruction. With it, you route from evidence.”

Our founder @_sumeetc at @AgenticSummit in NYC

[Hetu @hetu_protocol]

The new generation of Web3 infrastructure is reshaping the boundary trust relationship between participants. Traditional Internet and centralized scientific research systems rely on institutional credit and third-party endorsement, which inevitably brings problems such as opaque rules, centralized monopoly and high trust costs.
Decentralized protocols replace institutional trust with cryptographic trust, realizing trust minimization in all links of computing, data and collaboration. All network behaviors are executed according to open and transparent underlying rules, without the intervention of centralized institutions, and all contribution behaviors and value flows can be traced and verified.
This trust revolution is reconstructing the basic production relationship of Web3. It enables open, free and credible collaborative production in the fields of intelligent computing and scientific research, and becomes the core underlying support for the evolution of the next-generation intelligent digital economy.#HETU #Setu #TrustMinimization #DecentralizedTrust #Web3Evolution

[nick.eth @nicksdjohnson]

"A well telegraphed slow pivot."

In layman's terms, I think that means "we didn't tell you but you should have figured it out for yourself".

[Akasha(Inter/acc) @Sky201805]

Constraint is a better architect than abundance.

[Marvin Tong (t/acc) @marvin_tong]

What the hell? DeepSeek V4 make 97% cache hit for me!

MLA 1m context window to cache KV on hardware disk😅, from hours to days!

They may not have unlimited high-HBM GPUs, but China can make a lot of SSDs.

[Akasha(Inter/acc) @Sky201805]

@thedankoe Corollary: any field gated by psychology has less competition than its skill ceiling implies. Hard fields with easy psychology stay crowded. Easy fields with hard psychology stay open.

[Akasha(Inter/acc) @Sky201805]

In a bottoming market, patience matters more than conviction.
#btc #eth

[Aubrey de Grey @aubreydegrey]

I'll be speaking at LifeSummit 2026, Berlin, May 29–30.
100+ speakers. 3,000 attendees. A full expo floor. CME-certified — 12 points in 48 hours for physicians.
One of the best longevity events of the year - see you there!

[Akasha(Inter/acc) @Sky201805]

Studying gives you priors. Building gives you posteriors.
Steps 4–5 aren't roadblocks — they're the only data you actually own. Everything before is rented from someone else's experience.

[DAN KOE @thedankoe]

How to master any skill fast:

- stop studying
- outline a project
- start building it
- hit a roadblock
- figure out how to overcome it
- repeat 4 and 5 for the rest of your life

Most people don't get past 1, the rest spiral into complacency after 4.

[Akasha(Inter/acc) @Sky201805]

This is the exact inflection point I’ve been sensing.
@nathanmccauley and @Anchorage turning “Agentic Banking” from concept into infrastructure is quietly profound. Agents aren’t tools anymore — they’re becoming sovereign economic actors that need native identity, programmable trust, and autonomous capital rails.

Most legacy systems still treat agents as glorified human proxies. The real leap is building money and banking layers that are intersubjective by design: verifiable, inheritable reputation, zero-friction settlement, and on-chain provenance that agents can actually reason over and transact with at machine speed.

Hetu’s work on the Deep Intelligence Money stack is aimed squarely at this future — giving agents not just wallets, but economic subjectivity.

Extremely bullish to see traditional finance players starting to build the on-ramps. The agentic economy doesn’t arrive through hype. It arrives through protocols like this.

[Nathan McCauley ⚓ @nathanmccauley]

https://t.co/bPJppspwPK

[indigo @indigox]

Anthropic、Blackstone、Hellman & Friedman 和高盛宣布成立一家新的 AI 服务（JV）公司，该公司将与各行业的中型企业合作，将 Claude 应用于其最重要的运营环节。OpenAI 几乎同时宣布与 PE 财团合资，做类似的服务公司，采用 Palantir 的 FDE（驻场工程师）模式👀

为什么不找 Accenture/Deloitte 这种本来就有客户关系和实施能力的公司？

1. 因为 PE 自己就是最大的客户聚合体：Blackstone 全球被投企业 ~250 ��；Apollo ~200+ 家；TPG / Brookfield / Bain Capital / Advent 各有几百家 portcos。

2. JV 拿到的是"对所有 PE portfolio company 的优���销售权"，一笔交易等于打开几百家中型企业大门。PE 一边拿到 JV 股权（contract 收益），一边在自己 portfolio 里推 AI 提升 EBITDA（退出估值）。激励完美对齐——这是 PE 最擅长的"自家系统内套利"模式。

3. 中型企业 AI 部署的最大瓶颈不是技术，是说服 CIO + 走 procurement + RFP——12-18 个月。如果 PE 直接告诉 CEO "用这个，加 EBITDA 5 个点，估值多 30%"——周期会被压到几周。

另外，这是一个资本结构 + IPO 估值架构的设计：

FDE 服务业务毛利率 30-50%（Palantir 的 services 段），跟 SaaS API 的 80%+ 毛利率有 30-50 个百分点差距。如果 Anthropic 自己 in-house 做 FDE��整个公司的毛利率会被摊薄——等到 IPO 路演时，市场会按"软件 + 服务"的混合 multiple 给估值，而不是"纯软件"。差别可能是 10× revenue vs 25× revenue——按 Anthropic $ 50B 营收预期，是 $ 500B vs $ 1.25T 估值的差别。

Anthropic IPO 路演时，Dario 可以站着说"我们是纯软件公司，毛利率 80%，给我 25× revenue"——同时 JV 已经在地面跑 FDE 抓 mid-market 收入。这是给一级和二级市场各讲一个最有利的故事的双轨结构！

OpenAI JV 是 Anthropic JV 的 6.7×（$10B vs $1.5B）。但 Anthropic JV 有两个独特参与者：

- Goldman Sachs：OpenAI 名单里没有任何 IB。GS 可能在 Anthropic JV 里有更深的角色，这是 IPO 承销暗号？或者是 enterprise client 网络？

- Sequoia Capital：OpenAI 名单里没有 VC。Sequoia 一般不和 PE 共同投资。它在这里说明它把这个 JV 当战略基础设施而不是金融投资。

Anthropic 选择更窄、更战略；OpenAI 选择更宽、更资本密集。这跟两家在 frontier model 战略上的差异是一致的——Anthropic 一直更"挑客户、做窄而深"；OpenAI 一直更"all-out, all-comers"。

2026 年的前沿 AI 公司不再只是模型公司——它们正在用 PE 资本和 Palantir 模式把自己改造成"高 margin 软件主体 + 独立服务子公司"的双层结构，为 IPO 路演前后量身定制✨

[Akasha(Inter/acc) @Sky201805]

Reflection is overrated.
Memory is wrong. Attribution is wrong. "Rationality" is mostly your subconscious's defense lawyer—and the smarter you are, the better the story it tells.
Every decision is a probability bet. No sunk cost, no final verdicts. Update the model. Then act.
#AI #anthropic

[Akasha(Inter/acc) @Sky201805]

In the rising wave of the agentic economy, AI agents are poised to become core economic participants. They require payment, identity, reputation, provenance, and verification systems that are natively designed for them—built on cryptographic primitives and running autonomously on-chain—rather than simply retrofitting the legacy infrastructure of the human era.

However, the vast majority of today’s blockchain and payment infrastructures are still designed with “human users” as the default. They overlook the unique demands of AI agents, such as ultra-high speed, full autonomy, batch decision-making, and seamless cross-chain coordination. As a result, agents struggle to engage in truly frictionless 24/7 economic activity.

This is precisely where https://t.co/Bkeh5HsoEn, which I lead, comes in. We are not building yet another agent tooling layer. Instead, we are creating a native intersubjective economic layer purpose-built for agents—empowering every agent with a verifiable on-chain identity, inheritable reputation scores, autonomous payment channels, and zero-knowledge proof-based provenance systems. This enables genuine agent-to-agent trust and seamless value flows at scale.

[Katie Haun @katie_haun]

https://t.co/Unfl6DjfUk

[Akasha(Inter/acc) @Sky201805]

Read it again — almost every habit on this list is a refusal.Don't touch. Don't react. Don't pretend. Don't commit.Modern attention is colonized by default. The discipline isn't doing more — it's defending the gap where your own intent gets to form.

[DAN KOE @thedankoe]

Habits so simple you think they’re not worth doing, but have a profound impact on your life:

- Not touching your phone when you wake up
- Not thinking about work after work is done
- Putting a book down once you find an idea worth thinking about
- Setting aside time to do nothing for 10 minutes a day
- Going on a short walk after each meal
- Eating a meal without a screen in front of you
- Saying "I don't know" instead of pretending you do
- Asking "What if this isn't actually a problem?" before trying to solve it
- Letting yourself be bad at something instead of expecting perfection
- Trying to understand something you disagree with instead of looking for flaws
- Defaulting to "no" until you think through the commitment

[Akasha(Inter/acc) @Sky201805]

Stablecoin isn't the killer of cross-border payment. It just compresses the settlement layer.
The real unlock in LATAM isn't faster USD rails — it's when value starts flowing along intent, not corridors. Remittance is a trust problem dressed up as a settlement problem.

[Claudia @0x_claudia]

Everyone doing payment, fintech or programmable money is talking about stablecoins and LATAM remittance right now - but after building in local with my team I found some common wisdoms don’t work at all.

Here’re all the notes that combined all the data from my regional growth test and talked to local payment experts, communities, builders.

FYI

[Hetu @hetu_protocol]

Realised vs unrealised gains per alpha position is the kind of data subnet stakers have been waiting for. Clean execution @taostats.
Token-level visibility today. Agent-level verifiability next — that's the full stack.

[Akasha(Inter/acc) @Sky201805]

TL/DR
Don't time tops, wait for the 5-DMA break
Hardware leads, software follows
Domestic substitution is the long-tail trade

The cycle isn't over until the trend says so. 📈

[Akasha(Inter/acc) @Sky201805]

With AI, industry analysis has been democratized.
Anyone can pull first-hand financial data in seconds. So I'll skip the chart walkthrough and share a few takes on the AI trade 👇

[Akasha(Inter/acc) @Sky201805]

The numbers back it.
Institutional forecasts: domestic chip usage rising from <30% today → 60%+ in the coming years.
The pull-through effect on the entire supply chain? Massive.