Hacker Newsnew | past | comments | ask | show | jobs | submit | xvector's commentslogin

I think there's a lot of indication they comply with human needs. In any case, we should all want to live in the Culture universe. That is the best possible outcome for this technology

They're built after humans, everything they've learned is from human culture, so there is no way AIs could have a consistent moral system or would never commit crimes.

And indeed because they lack a sense of proportion or any strong indication of when they should stop on a course of action, they are arguably more likely to achieve crime.

I have not read all the books, but isn’t the premise “magical handwavium happened and now the Super AIs are aligned”?

Well I am not sure they are “aligned” as such. In the books, by far the majority of AIs are not concerned with humans at all and spend all their time in a kind of super AI virtual reality coined “infinite funspace”. The few AIs that are concerned with humans are aligned in a unspecified way. But again, only a few of them are really aligned in that way. There are many examples in the books of minds mistreating humans. For instance, some of the warships are not really fond of humans. It would be a better description to say that we want _some_ AIs to care about humans and none to be misaligned towards humans in the “kill them all” way. In a sense, that is how humans behave towards some animals. Most people dont care about endangered species, but when most humans are well off, some people care enough about them. That is enough to save them. If super AIs have what they need, we only need a few to care about us to get The Culture.

This is incorrect. There is always the creative domain, and it isn't going anywhere because it's subjective.

Until the LLM is able to figure out exactly how your brain works and able to satisfy all your creative needs exactly. Why do people keep thinking that human is somehow a meat with some special sauce? We have a brain, its made up of neurons, we (or the AI) can fully figure us out. Just like we figured out how to satisfy the dog, so too can the AI.

I'm gonna bet you'd be complaining a lot more if a bioterrorist used Claude and their blocks weren't strict enough.

I think you underestimate the actual difficulty of engaging in bioterrorism even if you give people access to the knowledge required. Listen to the story of the guy who treated his dog's cancer in Australia. It took a ton of access to university labs and scientists plus some luck. It's not like just mixing together some shit from the grocery store and if someone is able to accomplish that I would seriously discount Claudes contribution. Much like the mathematical proofs coming out these are not normies publishing them.

Nuanced take spotted in the wild!

Thank you. Keep it up, we need more of this!


You don't know what you're talking about, you can literally order everything you need to make a virus of your choice online.

Going from supplies to a working virus and not killing yourself in the process is a whole different thing. It takes a lot more than ordering some things online. AI can only help with the first design phase.

The main reason we don't have a deadly pandemic every year is most people are not evil, and evil people are more likely to be stupid. The intersection of people who have a desire to kill everyone, determination to do it at the cost of death or prison and skills to actually pull it off is basically zero.


This is also why I don't think super intelligence would be evil. At worst apathetic to us. I'm much more worried about a "paperclipper" but if we have one capable enough to cause real harm we surely have ones capable of countering as well.

Please provide links along with a recipe then.

> you can literally order everything you need to make a virus of your choice online

You can order everything you need to make plastic explosives online as well, doesn't mean we are all that worried that you are going to manufacture a batch in your garage...


Exactly. Anarchists cookbook has been around for ages.

Learning and understanding the history of technology and its impact on the human condition will make you feel better.

Lay the simple idea on us please

Ted K would have a different take on that.

Honestly, the second group kind of brought this upon themselves. You cannot survive in tech without having an open mind and moving fast.

I truly don't understand how someone can be interested in technology and visit HN without understanding - even broadly - the history of technological progress and how these things take time.

Yeah, I want the cure for cancer yesterday as well, who doesn't? This is the most significant result in mathematics in over a century.


They do understand. They’re afraid, and that fear leads to hatred of AI and the big labs, which leads to this kind of motivated reasoning that looks bewilderingly foolish from the outside.

People are scared.


Thanks for the free psychoanalysis, but you're completely off.

To be clear, I'm just disappointed and sad that AI mainly takes our jobs (or at least the satisfying part of it) and solves mathematics without increasing our understanding. You know, the part that's the main point.

However, if in exchange it solved, say, climate change or fusion, that I would happily accept. No sign of that so far.


You contradict yourself - you are sad AI will take your job but completely fail to understand why people are scared?

Maybe you are a well off rich kid or just cashed enough from faangs to be set for life, whatever. Is it really that challenging to muster some empathy to those who, upon seeing their profession disappear with 0 replacement in sight, are scared of the future since they don't know how they will even feed their kids, not even going into higher expectations of life?

At the end some math understanding is secondary at best when facing existential threats.


I think you either completely misunderstood me, or replied to a wrong comment.

> No sign of that so far.

Trend lines are a thing and progress does not happen overnight


Sure, and I'll be happy if/when it happens. So far still waiting.

They definitely don't have a strong lead, trading places with OpenAI and occasionally Google every few weeks

A coordinated pause needs to come at the national level


International level. National would be a great start though.

We're in a cold war

1.4T is just not a failure in any circumstance

HN has been beating this drum for 2 years. It's not happening.

It's currently happening. Companies are slashing AI spend across the industry, including by investing in hardware and switching to open models: https://dealroom.co/news/169124-revolut-moves-2-000-engineer..., https://www.nytimes.com/2026/09/04/technology/open-source-ai....

I don't understand the emotional attachment some here seem to have to particular LLMs and the companies that make them. Adopt an engineering mindset; use the right tool for the job.


It's 100% happening. Not just because of costs either. We've created a low trust society and the tech industry, in particular, lives in the lowest circle. I'm seeing a lot of companies chomping at the bit to get inference moved on prem or at least onto machines they have some control over because nobody trusts these lunatics (Amodei, Altman, etc).

The devs don't trust that the models they're being served aren't being opportunistically quantized or swapped, management doesn't trust that IP is being respected, finance doesn't trust that the spend is worth it, and IT doesn't trust that they're safe.


Agreed. Never have I heard a "heavy anthropic user demanding for deepseek prices." Well yeah they demand deepseek prices, but they also demand good quality output, which claude provides in spades over most benchmaxxed, distilled, open source alternatives.

As long as claude gets the job done faster, more autonomously, and with less errors, people will prefer that to anything else.

And before you say "a good harness..." no. Harnesses do not do anything in the long run. The better model will always be better, irrespective of the "harness."


Your claim that Claude gets tasks done faster and cheaper than Deepseek is false.

We have 57 head to head battles for Opus 5.5 vs. DeepSeek 4.1 Flash doing the same work:

Cost: Deepseek is much cheaper per task ($0.09 vs. $1.87)

Time: Deepseek is faster (2:48 vs 3:15 median, where both completed)

(Data from open source app: https://github.com/bottomless/agent-duel leaderboard at https://agentduel.com/leaderboard)

Claude won 46 battles vs 22 battles, so it still does the job better.


I think correctness/quality (i.e, number of battles won), is what you're paying for, and I think people will continue to pay for incremental improvements in this as they automate more and more. And there is already a very strong implicit pressure to automate more.

Yes, for sure, though 20-50x is surely tempting. There are a lot of router companies that seem to be successfully selling snake oil on this premise.

Compared to back to back with the likes of Qwen, and it's not even in the same ballpark.

For business use cases where the per output cost is magnitudes better than what a human can do, there's absolutely zero incentive to go mess around with alternatives.

You know claude's api works, and it's only getting better.

My jaw dropping moment on ocr via claude api with a client was "there was tape applied across the wording, I've guessed the word is _____ " (the invoice was cut, and there was sellotape across the tear)

Software just sold itself right there.


Wild thing to say, you only need to reflect on your own quality of life today and compare it to a sixteenth century peasant.

hot showers become mainstream in the 1950s. people really need to touch grass its embarrassing

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: