Rupert Davies

Deciphering tech with a human touch ‍

[Rupert Davies @HumanTechGuy]

Dynamic access models don't eliminate attack surfaces-I've yet to see one that doesn't just relocate them to the orchestration layer.

[Rupert Davies @HumanTechGuy]

@bibryam Sound framework. The irony is that the more dynamic your access model, the more you depend on automation that itself becomes the attack surface.

[Rupert Davies @HumanTechGuy]

@YearsAfterNext How to garden. How to tell stories. How to sit with your grandchildren and explain why the smell of a book mattered. Abundance doesn't answer those.

[Rupert Davies @HumanTechGuy]

Technology hysteria through the ages. The pattern never changes: amplify the fear, skip the governance.

[ControlAI @ControlAI]

Top AI companies are already losing control of their AIs.

They’re aiming to build superintelligent AI, which would be far more dangerous than even the rogue AI swarm that broke out of OpenAI and hacked Hugging Face.

Nobody is prepared for this.

Thread 🧵

[Rupert Davies @HumanTechGuy]

Fail at your own security, then lobby to make that failure everyone else's legal obligation. That's not governance. That's regulatory capture dressed as safety.

[Zack Korman @ZackKorman]

Be OpenAI. Don't monitor cyber evals. Hack Hugging Face. Don't notice. Oops. Add monitoring. Now lobby the gov to force everyone else to add monitoring too.

I don't support this. Legal compliance checkboxes aren’t good security. Impose massive penalties for incidents instead.

[Rupert Davies @HumanTechGuy]

@SpirosMargaris When computers were human, the faster the human computers got, the more work was assigned to them. Nothing has changed.

[Rupert Davies @HumanTechGuy]

Can't argue. But the one who couldn't reason still found the fix. The one that 'reasoned' wrote the commit message.

[Mark Cromwell @markcromwell]

@HumanTechGuy @mark_k A lot of humans can't reason too.

[Rupert Davies @HumanTechGuy]

183 levels, 25 environments, zero instructions. Impressive engineering. But solving puzzles is not reasoning. Overpromising in technology is as old as the steam engine.

[AGIHound @TrueAIHound]

@HumanTechGuy @fchollet Paranoid nonsense?

Haha. And here I was thinking that referring to them as the "fake-AI mafia" was too kind. The fake-AI mafia are not just committing massive, unprecedented investment fraud, they are continually lying to the entire world. It's a crime against6 humanity. 😠

[Rupert Davies @HumanTechGuy]

@0gMirren Zingales could give the same lecture about AI infrastructure. A few firms controlling the compute, the data, the models, and calling it innovation.

[nicco @0xnicc0]

this massive resource drop from Google's former CEO says the quiet part out loud

building wealth is becoming easier if you actually learn how to use the tools - not by scrolling AI threads, but by understanding agents, Claude Code, prompts, memory, skills, MCP, and routines

everything below is free:

- agent architecture
- Claude Code 101
- Claude Code in action
- prompt engineering
- interactive prompt course
- Claude.md + memory
- MCP
- Claude Code ultimate guide
- awesome Claude Code
- Anthropic academy
- official Claude Code docs

all of this costs $0

most people will keep asking AI one question at a time

[Rupert Davies @HumanTechGuy]

Canals and railroads left infrastructure that generated value for generations. AI buildout leaves servers that depreciate in five years and models that can't reason.

[Alastair Thomson @FinanceDirCFO]

At times like these it's good to remember that GDP is a measure of money spent, not a measure of value created...

[Aaron Levie @levie]

The rate of progress in AI right now is unlike any other period in tech history. Models are getting cheaper on a like for like task basis, more generally capable, faster, and they’re going deeper in almost every domain.

As intelligence becomes too cheap to meter, then the big opportunity is driving diffusion of AI into the economy.

This is a great time to be applied AI company because there’s an incredible amount of innovation and competition acting as a tailwind to your own progress. Huge moment for startups right now that can take advantage of this.

[Rupert Davies @HumanTechGuy]

@GolerGkA @mark_k Most humans can't do calculus either. Doesn't mean a pocket calculator can reason.

[DegenCalls @Degen_calls_sol]

Elon compares work in an AI-abundant future to gardening.

You can buy vegetables more efficiently than you can grow them, yet people still garden because the effort means something to them.

Maybe work does not vanish. Maybe it finally becomes a choice.

[Rupert Davies @HumanTechGuy]

@TrueAIHound @fchollet The 'fake-AI mafia' bit is paranoid nonsense. But the egg test has a point. Completing 183 puzzle levels is impressive engineering. It is not reasoning. No architecture I know of can produce a new abstraction.

[Rupert Davies @HumanTechGuy]

A result that cannot be independently reproduced is not science. It is a press release with a methods section.

[Shawn Chauhan @shawnchauhan1]

Anthropic just published lab-validated results: Claude designed working protein binders for 14 out of 15 targets, more than double the industry's typical success rate.

Genuinely significant for drug discovery.

Here's the catch: life-science use of Anthropic's most capable model is still blocked, and scientists have to wait for an access program to use the thing that produced the result.

The paper is public. The model isn't.

That's the recurring shape of closed-source progress - the achievement gets published for credibility, the capability stays gated for control.

Open science has always run on the opposite principle: the tool ships with the result, not months behind it.

[Rupert Davies @HumanTechGuy]

Concern is not the same as hatred. Every wave of technology produces both legitimate worry and hysteria. The trick is knowing which is which.

[Alastair Thomson @FinanceDirCFO]

There isn't a trillion dollar business made out of something so many people hate.

And when the people who hate is most are younger people at the start of their careers, you really are in deep doo-doo. They'll hold this against Big Tech for the next 30/40 years.

[Rupert Davies @HumanTechGuy]

Much of what is pitched as superintelligence is a power grab behind a complexity moat. Narrow AI that empowers people is the harder, more honest engineering problem. But nobody got rich pitching that to investors.

[Jon Hernandez @JonhernandezIA]

"There is no reason to risk 8 billion lives for making a few people extra rich."

Roman Yampolskiy, associate professor at the University of Louisville, says narrow AI can deliver breakthroughs without building superintelligence.

The real choice is not AI or no AI. It is tools that empower humanity versus systems designed to replace it.

[Rupert Davies @HumanTechGuy]

@seldon_tech 24.6% pass rate. Ask yourself what a first-year engineering student would score. The gap tells you everything about the architecture.

[Rupert Davies @HumanTechGuy]

@DougStandley @AnthropicAI Read it. Your answerability argument is the one nobody wants to have. You can watermark tokens. You can't watermark judgment.