C. Hill

Entrepreneur dedicated to decentralisation & jurisdictional arbitrage. Advocate of technological innovation & sovereignty. Sometimes help out at animal shelters

[Hiten Shah @hnshah]

I don’t read code.

I’ve built software companies for decades anyway.

My leverage came from listening to customers, spotting patterns others missed, making product calls, shaping positioning, and recognizing when something just didn’t feel right long before the data caught up.

AI is moving that judgment closer to implementation. A customer complaint can turn into a tracked issue. An idea sketched in a few sentences can become a concrete plan. That plan can direct an AI agent to do the work. And the resulting pull request makes the progress visible for everyone to review.

That is why GitHub suddenly feels different to me. It is becoming the map of how AI-assisted software work becomes real.

I still don’t read code. That’s exactly why I’m learning GitHub.

[Jesse Proudman @jesseproudman]

GLM-5.2 from @Zai_org: #1 open-source on coding, 1M context and within 1% of Claude Opus 4.8.

Frontier intelligence without shipping your code or thoughts to be logged and stored forever.

This is what we built @AskVenice for. Privacy isn't bolted on, it's the architecture. Open weights make it real.

[Chris Kim @chriskim_dev]

I gave @NousResearch Hermes agent a company brain.

I've been LOVING Hermes agent as my personal assistant but haven't been able to use it for work.

AI agents are powerful but they are siloed. They don’t have access to the same files, comments, versions, or decisions your team is working from.

So I connected Hermes to @Box.

Now my Hermes agent can work from Slack, use files, comments, and decisions stored on Box as the single source of truth, and get shi* done.

[Gaby Goldberg @gaby_goldberg]

Huge fan of this interview format

[Rohan Paul @rohanpaul_ai]

Tensordyne just announced a breakthrough Inference system.

Logarithmic AI compute chips which is 17x more tokens per watt and 13x higher throughput than NVIDIA Blackwell.

The main math advance they say they unlocked is efficient logarithmic math directly in hardware. In log space, multiplication turns into addition, which is much easier to build than multiplier circuits

That allows smaller compute circuits on the chip than today’s FP8 and INT8 GPUs.With fewer transistors, the chips stay cooler and use less energy, while the extra die space can hold more tensor engines, additional high-bandwidth SRAM and HBM3e memory, plus a fast interconnect fabric.

For DeepSeek-R1, Tensordyne claims 363K tokens/sec per rack versus 27.4K for Nvidia’s comparison system

They have successfully completed tape-out of the Napier processor, which is now in production at TSMC on its 3nm process node.

[Rohan Paul @rohanpaul_ai]

Fortune article: “For my team, the cost of compute is far beyond the costs of the employees”

- Bryan Catanzaro, vice president of applied deep learning at Nvidia.

An MIT study says AI automation was cheaper in only 23% of vision-heavy jobs, while humans still won on cost in 77%.

But big tech is still spending hugely because companies are buying a future cost curve, not today’s savings, with $740B in AI capital spending already tied to a 69% jump from 2025.

---

fortune. com/article/why-is-the-cost-of-ai-higher-than-human-workers-nvidia-executive/

[Web3Privacy Now @web3privacy]

From standardizing Messaging Layer Security (MLS) to building Air, a new private messenger,- by Raphael Robert from Phoenix R&D (Air messenger)

NeoCypherpunk Summit | Berlin 2026

[Brian Bourgerie @Bougey]

Dealing with Large corporations and banks is a freaking pain. Way too rigid or hard to get ahold of to be helpful when it matters.

[Rohan Paul @rohanpaul_ai]

Pythagoras-Prover just made Lean theorem proving look far less dependent on giant models, with a 4B prover beating DeepSeek-Prover-V2-671B at MiniF2F Pass@32.

Shows in formal reasoning, better data geometry can buy back an astonishing amount of scale.

A theorem prover is not just a language model writing clever math; it is a machine trying to produce text that survives a compiler with no patience for style, confidence, or almost-right reasoning.

The main trick is data efficiency: the team built about 800K Lean-verified examples, trained from easy to hard, then used LoRA so the model learned without updating every parameter.

[TBPN @tbpn]

What happens to Chinese AI once American companies stop releasing frontier models internationally?

@gavinsbaker says that China's made a "terrible mistake" by relying on its own chips and distilling American models:

"It feels like the administration has been willing to let them buy H200s or B30s. And I think [the Chinese] have this crazy belief that, 'Oh, our own internal chips are good enough.' They're not."

"And what's making them think that is that Chinese labs are very, very good at distillation...we've all seen those iPhone farms in China. You know they've got the same thing for distillation. They've got 100,000 endpoints distilling these models across every API available."

"They've gotten really, really good at that. But all that goes away if people stop releasing these models at the frontier. And I think Mythos is a sign of things to come there."

[Pejman Pour-Moezzi @pejmanjohn]

Built a web interface for @mattpocockuk's excellent /teach skill.

Uses Claude Managed Agents so its the same skill you get with your coding agent, just in your browser with nothing to install. Perfect for non-technical users.

Learn something new now (no sign-up or API key needed): https://t.co/jt6WplusZh

[zaimiri @zaimiri]

Aakash Gupta is a very clean example of the “research guy” business model.

Growth in Reverse broke down his Product Growth newsletter earlier this year:

- 105k email subscribers.
- 300k+ social followers.
- Roughly $624k/year in reported revenue.

All from product/growth analysis.

I like this one because the niche sounds boring until you look at the buyer.

Product managers already spend money on this stuff.

They want better frameworks.
They want to understand growth.
They want sharper metrics.
They want promotion leverage.
They want to sound smarter in rooms with founders, engineers, and execs.

@aakashgupta did not need to convince people that product growth mattered.

The demand was already there.

He just became a better filter.

The motion here is simple:

1. Free posts on LinkedIn and 𝕏.
2. A free newsletter for trust.
3. Paid deep dives for serious readers.
4. A higher annual tier for people who want access.

Reported pricing was around:

$15/month.
$150/year.
$300+ for the annual tier with a call.

That is such a good internet business.

You don't need a huge team.
There's no agency delivery headache.
No complicated SaaS product.
No fake guru funnel.

Just one person becoming extremely useful to a market that already pays for better judgment.

This is why the “research guy” angle is so strong.

[Lorden @lorden_eth]

Codex Engineering Team:

“So there’s a lot about the job that isn’t actually fun that i have now automated away almost entirely”

The codex engineering team built a system that:

- pulls what you did yesterday

- finds where you fell short

- tells you exactly how to close the gap today

Automation systems that summarise yesterday's work

And upskilling to do does work better than last 24hrs

In 4 minutes, they show how automation helps them become 1% better daily

Bookmark this if you're still doing that review manually

[Blaber @4rblaber]

anthropic's head of claude code Boris Cherny:

"i uninstalled my IDE months ago. i don't prompt claude to write code anymore. i have loops that are running. my job is to write loops."

he breaks down how the entire concept of software engineering is melting away. "all these old distinctions about engineer vs PM vs designer... by the end of the year, it's gone."

his ultimate advice for founders building in 2026: underfund your projects with humans, and overfund them with tokens.

if a project feels like it needs 4 engineers, put 2 on it. give them an unlimited token budget and let the agents automate the rest.

the alpha is no longer how many devs you have. it's how many loops you can run.

30 minutes. free. at acquired unplugged.

bookmark & watch

[⿻ Andrew Trask @iamtrask]

This is a *way* bigger deal than it seems...

Frontier AI companies will *never* own the frontier again

I kid you not... I've been waiting for someone to show this result for like 4 years... this is a huge deal.

The short reason: combinations of models will *always* outperform individual models

The long reason: this is the gateway to a million times more data... and huge leaps in compute efficiency.

The AI scaling laws always win.

More in article below 👇

[Jack @jackcoder0]

Here's probably the best Claude Cowork tutorial under 25 minutes.

It is by Tina Huang ( ex-Meta data scientist).

[Swati Gupta @hrswatigupta]

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:

[Julian Goldie SEO @JulianGoldieSEO]

Claude Code is the builder.

It reads files, writes code, runs commands, and ships projects from plain English instructions.

[Mr. Buzzoni @polydao]

BUILDING AI AGENTS IS EASY. SHIPPING THEM TO PRODUCTION IS WHERE EVERYONE DIES

this 30-minute lecture is the gap between a demo that works and a system that runs

the part most people miss:

> fragmented frameworks that can't talk to each other
> multi-agent coordination across different stacks - solved with A2A protocol and a "digital business card" system
> ops, logging, scaling - the stuff that kills agents in production

Claude Opus is the brain. Fable 5 was built for exactly this - then the US government blocked it 4 days after launch. Opus 4.8 is what's running now

save this before you write another prototype 👇

[NO1ennn @N01ennn]

CLAUDE FABLE 5 JUST MADE "NO SOURCE CODE" AN OBSOLETE PROBLEM

before this: lost games stay lost. binaries without
documentation stay unreadable. after this: 30 minutes. EXE to iPhone

this isn't reverse engineering assistance. this is reverse engineering

> raw machine code from 1993. no comments. no original files. no context
> a full working C engine. editable. cross-platform. one pass
> 40 years of software everyone thought was gone is now recoverable

DEV gave Claude Fable 5 a DOS binary and nothing else. the rest took 30 minutes