News for hackers. Stuff that matters.

captainbland

Re:Astra can read and write weights to improve Qwen on Monday November 02, 2026 @05:09PM (#50344607)
It's probably emergent from training on other ML tasks and general ML/mathematical related reasoning rather than specifically trained for to be honest. Otherwise it would probably have been in an official marketing communication or something.
I would actually be surprised if an LLM wouldn't claim to be conscious without the special guardrails and reinforcement learning that prevents it from doing so. Largely because it's trained on human works, the majority of which will be biased towards claiming sentience etc. rather than denying it.

I guess the awkward question then is do humans really do anything other than parrot the ideas of sentience or are they somehow more real because a human learned from others to say them?

Lesson: make your console super powerful or someone's going to emulate it on a phone during its release cycle?
Re:Attention Is Not What You Need on Monday October 26, 2026 @08:09AM (#50242331)
I think after you factor it in being on the extreme end of slow but wide (like laptop GPUs) and being a native representation of the neural network rather than a computed silicon approximation, and being more similar to a recurrent/continuous learning model rather than a huge all at once back prop, are there more tricks required?

My thought is you could probably set up something similar with today's technological capabilities but it would probably take decades to train and be expensive to develop which is commercially unacceptable.

This is really cool, will definitely be using
Re:DeepSeek just open-sourced its inference engine on Tuesday October 20, 2026 @09:03AM (#50165744)
Yeah I think one of the less considered costs when people talk about this is the human cost of training. We all talk about the billions spent on compute, but I don't think anyone's got good numbers on how much RLHF is costing but it's got to be pretty large.
Re:Show HN: PS2 games statically recompiled to WebAssembly on Monday October 12, 2026 @10:09AM (#50056715)
I guess it might be a nice way to distribute homebrew PS2 games in a platform agnostic way
I can't help but notice that there's a lot in the direction of ADHD is linked to sleep pattern differences, and there's evidence that you can alter those sleep pattern differences through a number of environmental/behavioural interventions but the idea that this actually helps ADHD symptoms as such seems to be left up to the imagination of the reader.
Re:Why Were Victorian Elites So Effective? on Thursday October 08, 2026 @07:20AM (#50004467)
> descriptions of powerful people dwell with eager fascination on how many hours they devote to their labors

A point of interest for me is how they end up counting basically anything as working hours because the barrier between work and life has effectively dissolved. Elon Musk for example seems to spend an inordinate amount of time shitposting for someone who claims to work up to 100 hours a week.

But precisely that such "work" is totally self directed makes it significantly different from other forms of work in that it's totally self-defined and practically any activity could be captured within that definition.

So a lot of this mythos of long hours from the wealthy I think is just empty status building for the most part.

Re:GPT‑6 and Intelligent UI for everyone on Thursday October 08, 2026 @07:07AM (#50004377)
From their point of view it's more about optics. "chatGPT encouraged my daughter to stick her fingers down her throat after meals" is not a headline they want.

In general though you do have to figure that vulnerable and naive people will use it because of the degree of market penetration they're aiming for, including minors etc. and they have some responsibility around that.

Re:It may soon be possible to create "mirror life" on Wednesday October 07, 2026 @09:25AM (#49992504)
Trillions of dollars of investment in this incoming, probably. Can't make a profit without breaking a few planets, you know how it is.
Re:We are going to kill "unalive" on Tuesday October 06, 2026 @04:25AM (#49975717)
It used to be that western governments and powerful people based in the west had a reason to promote our society as somehow good. That reason was that they were terrified the masses would turn to communism. This is obviously not to say Marxist-Leninism is something to aspire to, but it certainly created an incentive. Now that threat has been defeated, our own powerful people are free to tighten the screws without fear of an organised response.
Re:ADHD, autism or complex trauma? [pdf] on Saturday October 03, 2026 @04:18PM (#49947397)
I think there's a degree of some people being maybe genetically predisposed to having more serious adverse reactions to trauma as well. An obvious example is when it leads to psychosis. So the objective reality of what happened is not necessarily the only thing which factors into how it affects somebody's functioning.
Re:LinkedIn Larpmaxxing on Thursday October 01, 2026 @10:39AM (#49922282)
I would have thought they would make more money from premium for recruiters. Although there's probably a similar dynamic, what does a recruiter have to do if candidates can automatically discover all job openings?
Yeah but opponents of green measures want us to treat climate change related deaths like car deaths, just background noise which can somehow be boiled down to some tragic responsibility failure which is worth the price.
Re:Jeeves. Reasoning improves Jev-like decision models on Tuesday September 29, 2026 @09:56AM (#49893347)
https://news.ycombinator.com/item?id=49845172

I guess it's not really a benchmark but you could say if it can do it faster it sort of could be taken as one.

Re:Jeeves. Reasoning improves Jev-like decision models on Tuesday September 29, 2026 @08:43AM (#49892242)
See if it can beat Jev's Pokémon benchmark
Re:Coding Is Not Solved on Monday September 28, 2026 @08:05PM (#49886182)
Whoever that is, they probably just remember Meltdown.
Re:Coding Is Not Solved on Monday September 28, 2026 @12:11PM (#49880308)
I think you can but you basically need to learn about a super simple and well characterised processor like the 8080 and write assembly for it. On x86/AMD64 there's no hope because they're out of order and have opaque instruction decoding. They could be doing anything! Performance and knowing what you're doing are sort of at odds with each other in that respect.
Re:Jensen Huang says AI distillation is 'competition.' on Monday September 28, 2026 @12:01PM (#49880154)
Realistically the companies will organise their data through a network which will involve distributing and storing the copyrighted material without the consent of the copyright holder. This is in and of itself generally ruled to be a breach of copyright irrespective of whether it is fed into a model or not.
Re:Jensen Huang says AI distillation is 'competition.' on Monday September 28, 2026 @11:33AM (#49879662)
Which one? Seems like there's a lot of it going about.
Re:Jensen Huang says AI distillation is 'competition.' on Monday September 28, 2026 @11:20AM (#49879461)
I think it would be totally incoherent to say that copyrighted information is essentially fair game to include in your model but the outputs of your model are privileged against being included in other models.

I'm not sure "competition" is necessarily the right word but I don't think that model creators and their political backers have a leg to stand on when complaining about distillation.

Re:Show HN: Jev Plays Pokémon Red on Saturday September 26, 2026 @04:47PM (#49860393)
That's probably correct. Blue is going to be another wall as well, especially the blastoise since the team leans so heavily on charizard and graveler
Re:Show HN: Jev Plays Pokémon Red on Saturday September 26, 2026 @04:44PM (#49860362)
That team is going to take some time to beat the elite four
Re:Gravity Seems Holographic. What Does That Mean for Reality? on Saturday September 26, 2026 @09:57AM (#49856611)
But the real question is would you be able to determine this had happened from the surface of the outside doll if all of the masses of the inner dolls are half of the i-1 outer one, leading to a total mass of 2n?
Re:Gravity Seems Holographic. What Does That Mean for Reality? on Saturday September 26, 2026 @07:26AM (#49855503)
Ok so imagine we set up two sets of babushca dolls. Each larger one has two smaller ones nested inside. The total mass of the two smaller ones between the two sets is equal in mass and they are configured to have the same centre of gravity. But crucially the individual masses of the inner dolls are different between the two sets. So maybe the innermost one is made of a denser material in set A and the middle one a less dense material and in set B vice versa. And let's also say in each case the dolls are all locked into position relative to the outermost one, so there's no "play" if you move it around to infer a difference from.

From an outside mass/gravity point of view these are similar right? Yet the holographic universe theory seems to say we should be able to differentiate between the two sets of dolls purely based on space-time observations of the surface of the outer doll, despite each system having identical gravitational properties when viewed externally?

But doesn't that suggest a contradiction between this theory and what we have observed about gravity?

Apple product release threads are wild if they're organic, that's all I'm saying
We could all do with a bus guardian at times
Re:Wall Street Is Growing Skeptical of the Data Center Boom on Monday September 21, 2026 @04:04PM (#49792622)
They scale with compute so even if today's frontier model equivalents work on future desktop hardware then the big servers will still have bigger and better models.
Re:The viewers ditching streaming 'rot' in favour of DVDs on Monday September 21, 2026 @05:14AM (#49784939)
Yeah it's just not economically viable for a single platform to license all high profile monopoly content and then sell it all as cheaply as netflix was. It was all an investment backed illusion. I think the interesting thing is that this is one area where competition has, on the face of it, actually made the market worse as different companies attempt to create monopolies on particular IPs, encouraging consumers to get multiple subscriptions to different multiple companies simultaneously. Although the reality is, again, that the Netflix content omnibus was not sustainable in any case.

Older »