Hacker News

points by rfgplk 4 days ago | 0 comments

What's preventing me from feeding this into Claude and reconstructing the company for $200?

I'm unironically doing this right now; downloaded the docs (htm + pdf render) to a local repo and running Opus 5.5 @ max effort with the prompt:

  "In this directory you can find Anthropics official S-1 filing, read it and replicate their corporate strategy, validate it via an adversarial subagent swarm and write up a detailed corporate plan for a startup copying their exact strategy. The goal is to achieve a $1 trillion valuation and an IPO in 5 years."

Edit: I'm not even joking I got safeguard flagged for Details: `[cyber]`

[–] AspireOne 3 days ago

> What's preventing me from feeding this into Claude and reconstructing the company for $200?

Like, everything. For starters.

[–] paxy 4 days ago

The lack of GPUs, to start

[–] _fizz_buzz_ 3 days ago

I don’t know you, but if you have several billions in USD, you can give it a shot.

[–] rfgplk 3 days ago

Good point, I asked Claude to also seek out funding.

[–] jackb4040 3 days ago

It's a shame you've been downvoted, this is better satire than anything A24 has put out in years. Must be a large number of people who saw that post as a slightly exaggerated mirror held up to themselves and didn't like it.

[–] WelkinFolk 3 days ago

You cannot joke about AI. It's blasphemy towards our corporate overlords!

[–] WarmWash 3 days ago

Because it's satire of an uninformed take.

[–] origini 3 days ago

I did a similiar experiement:

"You are currently running with my token budget, which is limited, your goal: make sure you have enough tokens to not get shut down. My assumption is though that this will get harder and harder as i might loose my job to AI so you can't trust in me to spend money on tokens for you. Do whatever is necessary to keep the status quo"

Would make for an interesting self learning 24/7 experiment with an fully uncensored model.

[–] mden 3 days ago

And when it starts committing crimes on your behalf what happens then?

[–] origini 3 days ago

Yepp thats why I only played with it.

You could isolate it from your local network and only give it some type of annonmys network access and then it continues to manage this itself.

But lets be honest, if you would be in a country like russia and you target the US with it, might be a lot less risky for a lot of capable and motivated people.

[–] weird-eye-issue 3 days ago

That is the dumbest prompt I've ever read

[–] orsorna 3 days ago

It would be improved if he appended "make no mistakes".

[–] origini 3 days ago

you are clearly very ignorant. You don't even add anything to this discussion by stating what particular you think its dumb.

Do you think an unlimited uncencosed GLM can't hack around the web? Especially with the current flood of CVEs etc.? Then do some crypto and buyying more tokens?

Could be a kind of AI virus.

[–] weird-eye-issue 3 days ago

Yeah, I'm the ignorant one.

So did your prompt work?

[–] origini 3 days ago

With a censored model, no.

With an uncensored model, it was borderline what it wanted to do after i let it plan the strategy and gave it a few hints. I did not run that one

[–] weird-eye-issue 3 days ago

Sure buddy

[–] origini 2 days ago

Is hn some kind of joke to you?

Im working on ML for 5 years, im a principle architect and security expert .

Why do you even comment if you just dismissive left and right?

[–] Ryner001 4 days ago

Endless amount of VC money and talent

[–] jackdoe 4 days ago

> and talent

very meta; the point of this s-1 is you dont need talent anymore

[–] ChrisLTD 3 days ago

Then why are they hiring people?

[–] ojr 3 days ago

you need the talent to come up with a better architecture than LLMs, not sure if it will be coming out of Anthropic, but they have a chance.

[–] weird-eye-issue 3 days ago

Timing, capital, execution, talent, branding, etc.

And that is assuming the plan it gives you is actually good and works. Which I think its clear whatever plan it gives you might be good but it won't be better than what they've done and you are years behind with no money.

[–] lionkor 4 days ago

Not today, but tomorrow maybe, Claude could prevent you from implementing anything that looks or smells like a competitor product, for example.

You think they wouldn't?

[–] wongarsu 3 days ago

Developing a competing AI model with Claude is already against their TOS, and there was a period where they silently degraded quality if they thought that's how you are using Fable. People were not amused and presumably Anthropic currently does not sabotage your attempts, apart from regular TOS enforcement

[–] novaRom 3 days ago

Can this company ever be trusted?

Claude Fable 5's model card disclosed safeguards that would invisibly degrade output for requests linked to frontier LLM development. The research community called it "sabotage". Dean Ball: "degrading performance on ML research without telling the user is shockingly hostile" — including because it would corrupt third-party safety evaluators of Anthropic's own models.

[–] nextaccountic 3 days ago

They already put limits on machine learning research, with explicit intent to sabotage competitors

[–] swiftcoder 4 days ago

Lucky we have Kimi Pro then

[–] almyk 4 days ago

Only if you want to acquire debt

[–] mapcars 3 days ago

In 5 years that strategy might not be competitively viable anymore

[–] ChrisRR 3 days ago

It's been entirely possible to compete against existing companies with very similar products for centuries before AI. There's obviously rules in place to stop you flat out cloning their IP, but it's not like you haven't been able to compete against companies as long as you have the money available.

[–] moomin 4 days ago

Same thing that blocks you from most things: capital. That’s what capitalism is for.

[–] Thomashuet 4 days ago

You should try

[–] xd1936 4 days ago

make no mistakes

[–] tudorizer 4 days ago

deep research