Carlos Corrales

15+ years in tech. Sharing everything I build and learn 👇

✍️ writing in

[Hyperagent @hyperagentapp]

Fable 5 on Hyperagent is producing the most creative, ambitious work we've ever seen from our agents.

They're self-improving for hours towards open-ended goals. Visual reasoning has spiked noticeably. Outputs are consistently higher quality than Opus, occasionally at lower cost.

5 of our test cases below vs. Opus 4.8 👇

1. Visualize all asteroids in the solar system from NASA data
2. Design a site plan for a 100 acre fitness retreat
3. Reconstruct Apollo control panels from technical PDFs
4. Simulate the supply chain for World Cup jersey sales based on match outcomes
5. Show the effects of solar flares on aurora

Fable 5 is now available on Hyperagent.

[Molly O’Shea @MollySOShea]

NEW: Inside @PsiQuantum's Silicon Photonic Chipset

*Never-Before-Seen*

With Er-Xuan Ping, SVP, Barium Titanate (BTO) Development

Backdrop: the U.S. government just announced a $2B push into domestic quantum computing manufacturing - but what does that actually fund?

The @CommerceGov Department recently awarded PsiQuantum $100M to accelerate development of BTO, giving a rare look into the underlying manufacturing stack required to scale fault-tolerant quantum computing.

We profiled PsiQuantum last September following its $1B Series E, including how the company produces the world’s highest-performing optical switch - a core component of its silicon photonics platform.

PsiQuantum is currently the only company or institution in the world manufacturing this BTO material for optical switches at 300mm scale.

While the company built this silicon photonics platform for fault-tolerant quantum computing (FTQC)*, the implications may extend far beyond quantum itself.

As AI data centers increasingly shift from copper to optical networking, PsiQuantum’s photonics stack could also become foundational infrastructure for next-generation AI systems.

*Fault-tolerant quantum computing (FTQC) refers to quantum computers that can continue operating accurately even when individual quantum bits (“qubits”) are noisy or error-prone.

Shoutout to @PeteShadbolt

[Compute King @Compute_King]

继续思考，

华为在挑战里面没有谈散热，这是我比较诧异的。

目前两层堆叠，我觉得还有些散热的解法，但如果到三层Active Logic Stack或者更多之后，散热会从工程问题变成架构主问题。。。

目前流行的双层堆叠的技术AMD V-Cache，Intel Foveros和TSMC SoIC，还属于用冷cache叠热logic，因为SRAM功耗较低，热密度低，可用做Top Die，所以散热还能接受。

结构如下所示：

SRAM
||||
CPU

但华为的论文里是Logic-on-Logic。

即：

Active Logic
||||
Active Logic
||||
Active Logic

这就完全不同了，这种多层Active Logic，热无法横向扩散，所以中间的Logic Die直接变成了烤箱，传统散热是完全扛不住的。

三层或者三层Active Logic堆叠之后，必须进入主动式散热时代！冷却液必须进入封装内部。

变成，

Active Logic + Microfluidic Channel
||||
Active Logic + Microfluidic Channel
||||
Active Logic + Microfluidic Channel

液冷液冷液冷是关键！关键得说三次。。。

以后芯片设计里面需要Thermal Topology Architect，因为：热路径本身会决定Layout。

对的，本人的判断是：华为将来3层和3层以上的LogicFolding路径里面，Thermal将是最大的未解难题，甚至比EDA还难！

[AdiiX @adiix_official]

ONE GITHUB REPO AND $5 BILLION IN 5 YEARS.

Two guys from New Zealand took open-source code and built the backend now powering Netflix, Microsoft, Coinbase, and Uber.

Paul Copplestone CEO and co-founder of Supabase breaks down in 46 minutes how they actually pulled it off.

save this and watch it.

[Peter Steinberger 🦞 @steipete]

Looks like our focus on performance paid off.

[Carlos Corrales @_ccorrales_]

👏

[Suryansh Tiwari @Suryanshti777]

RAG might already be becoming obsolete.

A month ago, Andrej Karpathy dropped a simple GitHub gist called “LLM Wiki.”

Now the comments section looks like the birth of an entirely new AI category.

5000+ stars later, developers are rapidly building:
• persistent AI memory systems
• self-maintaining knowledge bases
• multi-agent research environments
• contradiction detection engines
• AI-native company operating systems
• local-first memory architectures
• graph-based reasoning layers
• evolving second brains

And the craziest part?

Most of them were built in DAYS.

Because the core idea is insanely powerful:

Instead of AI repeatedly retrieving raw chunks like traditional RAG…

…the model continuously maintains a living knowledge system.

Not temporary context.

Persistent synthesis.

The shift sounds subtle until you realize what it changes:

RAG:
retrieve → answer → forget

LLM Wiki:
ingest → synthesize → evolve

That one architectural difference is causing an explosion of experimentation right now.

People are already building:
• agent memory operating systems
• AI-maintained engineering documentation
• self-healing knowledge graphs
• persistent research environments
• conversational memory architectures
• contradiction-aware wikis
• context compression engines
• machine-readable company systems

The comments section alone feels like watching an ecosystem form in real time.

One developer built deterministic contradiction detection using sheaf cohomology

Another built “sleep consolidation” for AI memory systems inspired by human memory formation

Another created persistent multi-agent vault conversations

Another turned entire repositories into continuously maintained AI wikis

Another built local-first memory systems with audit trails, provenance, graph exports, and MCP integration

This is the important part:

Karpathy didn’t launch a product.

He introduced a pattern.

And patterns are what create ecosystems.

The same way:
• transformers created modern AI
• RAG created AI retrieval startups
• agents created orchestration frameworks

LLM Wikis may create persistent AI memory infrastructure.

That’s why this moment feels different.

For years, AI systems have been stateless.

Now developers are trying to build systems that actually accumulate understanding over time.

And once knowledge compounds instead of resetting…

…the entire interface layer of AI changes.

(Link in comments)

[Steve Burns @SJosephBurns]

A one-minute trading master class in what makes you profitable. ⬇️

[Rahul @sairahul1]

Anthropic's Head of Product just dropped a 28-minute masterclass on agent production.

Prompt caching. Tool search. Programmatic tool calling. Compaction. Advisor strategy.

28 minutes. Free. Worth more than 100 YouTube videos combined.

Watch it first.

Then read this.

The masterclass teaches you how agents work.

This teaches you what to build with them — a 5-agent content pipeline that does the work of a $300K creative team.

Full pipeline below ↓

Bookmark this. Start this weekend.

[Phosphen @phosphenq]

The math that powers serious quant models is Markov chains. Almost nobody breaks them down this clearly.

This 1-hour video covers how the framework actually works & how top funds apply it in practice.

Bookmark & set aside an hour tonight.

[AdiiX @adiix_official]

CLAUDE PROMPTING WORKSHOP 28 MIN
(Straight From The Team That Built It)

Anthropic just showed a 28-minute workshop on how to actually do prompts for Claude.

Taught by the people who built it.

Watch it and bookmark it now

[Rahul @sairahul1]

Karpathy just described what hiring looks like in 2026:

"Build a large project with Claude Code — like a Twitter clone. Make it secure. Have real agents using the platform doing stuff. The interviewer uses parallel agents trying to break in to verify security."

One person. Multiple agents. Shipping and defending production code simultaneously.

This is not a future job description.

This is happening right now.

The founders who get there first are not the smartest ones in the room. They are the ones who stopped doing everything themselves and built agents to do it for them.

Here is the complete playbook — 13 agents, exact prompts, 90-day build plan ↓

Read this before your competition does.

[Goshawk Trades @GoshawkTrades]

Jane Street just showed the inside of their AI training data center in Texas.

4,032 GPUs. 56 racks. 8,000 km of fiber. liquid cooling running through every server because air cooling can't handle the heat anymore.

but the part that got me was the origin story.

Ron Minsky, who co-heads their technology group. said their first compute cluster was literally six Dell boxes stacked on top of each other at the end of a desk row. they called it "the hive."

the trading systems sat out in the room with the traders because they wanted to be able to unplug them if something went wrong.

at one point, someone vacuuming the office unplugged a live trading system in the middle of the day.

from six Dell boxes and a vacuum cleaner incident to a liquid-cooled GPU data center processing trades in under 100 nanoseconds.

that's a 20-year arc.

[bodila @51bodila]

Jane Street hired this junior at $220k-$600k /year because he uses AI to analyse TRILLIONS of data

in this 1-hour lecture - he show how to research trillion of data points thanks to his machine

Bookmark & watch it, instead of Netflix to learn how to do the same!

[🌋🌋 Deep₿lueCrypto 🌋🌋 @DeepBlueCrypto]

"Somebody I know went to work for SpaceX and said, 'It's like being dropped into a zone of shocking competence.'

The best engineers in the world want to work for Elon Musk because he's the one CEO who's able to work with them as a peer."

[Carlos Corrales @_ccorrales_]

Omg it’s starting

[Min Choi @minchoi]

Anthropic shut down an entire company's Claude access overnight

60+ employees. No explanation. Just an email.

Want to appeal? Fill out a Google Form.

Integrations gone. Histories gone. Everything built on Claude... gone.

Never let one vendor own your workflow.

[Movez @0xMovez]

This 30-minute speech by the Head of Anthropic "Coding Agents" researcher will teach you more about vibe coding than 100 paid courses.

Bookmark it & give it 30 minutes today. This video will change the way you use AI forever,

[Marc Andreessen 🇺🇸 @pmarca]

Essential viewing.

[Evan Luthra @EvanLuthra]

Anthropic pays engineers $750,000+ a year to understand how LLMs work.

Stanford just put a 2 hour lecture that covers 80% of it for FREE.

Bookmark this. Give it 2 hours today.

It might be the highest ROI thing you do this month:

[Aakash Gupta @aakashgupta]

Leopold Aschenbrenner sold every share of Nvidia and Broadcom last quarter. Took the money and bought fuel cells, Bitcoin miners, and power companies.

Today Oracle signed a 2.8 GW fuel cell deal with his largest holding. The stock jumped 15% after hours.

This is the most contrarian AI trade on Wall Street right now, and the math behind it is wild.

Leopold wrote a 165-page essay in 2024 arguing AGI arrives by 2027. Then he translated that prediction into a pure energy play.

His logic: scaling from GPT-4 to superintelligence requires data centers consuming more electricity than most American cities. You can order 100,000 GPUs and get delivery in six months. You cannot add 500 megawatts to the grid in six months. The binding constraint on who builds AGI first is watts per rack.

So while AI funds stacked Nvidia at 30x revenue, Leopold built the opposite portfolio. Bloom Energy, his largest position at 15% of the fund, makes solid oxide fuel cells that can power a data center in 55 days. Grid interconnection takes 2-3 years.

He entered 2026 with $876 million in Bloom Energy. That position has more than doubled. His fund went from $254 million in equity positions in Q4 2024 to $5.5 billion by Q4 2025. Beat the S&P by 47% in its first six months.

He's 24. Got fired from OpenAI two years ago. Zero prior fund management experience. The Collison brothers and Nat Friedman backed him anyway.

Today's Oracle deal validates the entire thesis. Oracle contracted 1.2 GW immediately, with a pipeline to 2.8 GW, because Bloom delivered a fully operational system in 55 days last year. A month ahead of schedule. Oracle needs power faster than any grid can supply it.

GPU supply is expanding on a known curve. Electricity supply isn't. Leopold bet his entire net worth on the gap between those two curves, and so far the gap is only getting wider.