Hacker Newsnew | past | comments | ask | show | jobs | submit | random3's commentslogin

Some are cautious about hinting their competitors. Incremental pushes are hints for everyone else and they get scooped.

That seems like an outdated model and not really in the spirit of scientific collaboration

Scientific collaboration happens through publishing, not giving away the ideas/credit before publishing. Sure, there's collaboration, but that's not with the entire community, especially not in math which for big results is an all-or-nothing endeavor.

Also outdated since when? We've just seen a scandal about OpenAI scooping Levent on Navier Stokes https://en.wikipedia.org/wiki/Navier–Stokes_priority_controv...


For a proof like this, I assume many of the components are interesting enough to be published on their own. The authors just don't.

Compared to solving RH say in a few months they are not interesting enough to be published on their own prior to that.

I agree that it is inefficient, but I understand why mathematicians work like this.


As I understand how credit goes for mathematical discovery, if you do 99.9% of the work and then I come in and put the last couple steps that make it actually work as a proof instead of a bunch of useful lemmas, the proof has my name on it.

You gotta eat.

Agreed. I know nothing about the field, but I always assumed that mathematics was the gold standard of an open, collaborative scientific process (especially considering how mathematicians are known to hold most other fields in contempt for not being as "pure" as them). If one mathematician sitting on a work-in-progress proof in total radio silence, for 12 years, is not uncommon, then clearly I thought wrong.

Mathematicians in general do not have contemot for other fields.

Not all mathematics is done in secret solitude, but it certainly exists. FTL by Wiles etc.

I feel that many commenters on HN have a very negative view of mathematicians. As if they are all arrogant, insecure and only seeking status.


IDK, it feels like “climbing” K1 then deciding to take the helicopter and call it a win. Nothing wrong with helicopters, it’s just not the same race. First, all will question your ability and, second you have to ask what could have everyone else tackling the same problems could have achieved in a similar manner.

I feel that this is a surprisingly apt analogy. Flying a helicopter at high altitude takes a lot of skill, and the only person to "land" a helicopter on Everest has a Wikipedia article, but he's no Sir Edmund Hillary.

https://en.wikipedia.org/wiki/Didier_Delsalle


its more like manual agriculture vs agriculture with a tractor

The difference is that a proof is true or not independently of the human sphere. It’s not an endeavor like “get to the top of that elevation, but only use your legs so you can demonstrate your mental fortitude”.

The race hasn’t changed, people are now retroactively pretending that the race was always about something other than it was commonly understood to be, because they don’t like that machines are beating them at it.


This is not at all historically accurate. The purpose of mathematics was always understanding. As far back as the computer proof of the four-color theorem, people have largely felt that "having a proof" wasn't as good as understanding why something is true. We still basically don't know why the four-color theorem is true, other than the fact that you can rule out the 600+ minimal counterexamples.

> people are now retroactively pretending that the race was always about something other than it was commonly understood to be, because they don’t like that machines are beating them at it.

I think this is uncharitable.

It is true that a math paper is useful to the world regardless of whether it was authored by human or machine.

It is also true that people have a deep-seated need to feel esteemed and valued by their community. Many people derive that from their profession and many mathematicians derive it from being able to manipulate symbols in their head in a way few others can to discover truths about the world.

It is cruel to watch that meaning get ripped away and shrug simply because the same math papers are coming out in the end anyway.


I see the platonic angle, but the proof being true is less of a concern from a "race" perspective. It's just saying that there's a K2 peak, vs climbing it and vs climbing it for the first time.

Obviously there are more aspects here:

- is a conjecture true or false - this has utility on its own

- finding a proof - this is an ability game/race - this used to be a part of math, but I'm arguing at this point this got muddy unfortunately

- finally, is there utility (other than personal desire to understand) in having people do math

Surely, just like we don't mix F1 cars with Usain Bolt, we can still have IMO.


You could use a forklift to lift a barbell but you would miss a point on why most people lifting weights.

The proof of the blowup has zero effect on our ability to predict weather.

Most mathematicians have long expected the Riemann hypothesis to be true — it matters whether techniques from the proof may be used for something, whether there is an insight , not just whether there is a proof.


Can you share more, or direct us to more resources around local setups? For starters what's the upfront cost, what are the space, weight, energy requirements, etc.

Really depends on what GPUs you go with but I hate proprietary software so I went with AMD AI PRO r9700 cards which you can still pick up used on ebay for $1500 used on auctions. Started with 1 then expanded over time as I won auctions. Though on a lower budget my $400 16gb 6800 XT can run Qwen 3.8 27b which I could absolutely get by with at significantly reduced speed.

My entire 12u rack has roughly the same capability as an H200, though at maybe 15k all in with mostly used parts vs 50k for a used H200 checking todays ebay prices and that is -just- the card you still need to build a system for.

Consumer cards are a lot more power hungry than enterprise cards and they take up more space. I run my setup next to my dryer, running on 240v for 15% better power efficiency.

Also using transparent memory encryption on a Xeon and an EPYC for confidential and private inference guarantees no third party can offer me. Of course I can run abliterated models that do any security research I want without refusals.

Strata and Radiance are the state of the art ATM for running big MoE models on small consumer GPUs at high performance.


Re: running qwen 3.8 27b on a 16gb card - that isn't a MoE model, was it just heavily quantized? How much context window were you getting?

Just starting to dive in on local AI with my own 16gb card, and the struggling with evaluating what tradeoffs are worth it.


On a 16gb card check out Bonsai2. I run it on my 6800XT.

Was pleasantly surprised to get 25-35tps on 3060, 12GB VRAM with Bonsai2. Quite smart, tool calls, vision, 95+% intelligence of dense qwen3.8 27b apparently.

Impressed with your setup @lrvick. Hoping Bonsai type improvements bring local inference to more modest setups. Still use frontier models for now, but very ready to stop.


I would laugh heartily if this user turned out to be a Black Hat who just wants to gush about his gear, because this is exactly the rig I would deploy if that was my aim.

As a career security engineer that has found and set 0days on fire that could have made me tens of millions of dollars many times, I am very curious what point you are trying to make.

I'm not a big fan of reddit lately but the localllama subreddit has tons of useful info on getting started, hardware and software-wise.

There are tons of resources out there on how to run a local setup. Once you have something that's somewhat functional you can then have the system self optimize your software stack to suit your needs.

As for cost a 9700 is running at about 2k USD MSRP right now and the rated power is 300w.


People complaining about how difficult self-hosting is seem to be stuck in the past and missing the point. It's the very AI that we're talking about that removes most if not all friction points related to software configuration and maintenance.

While this doesn't remove the upfront cost (which could be fixed with basic financing like Apple does) and physical manipulation (e.g. replacement on failure, etc.) it does lower the barrier.

Finally, there's another path I find interesting and interested to think more of which is sharing hardware locally, from your house to your neighborhood, etc. We've done this in the late 90s - early 00s with Internet and it worked really well. I think it could work great for compute.

I see a lot of opportunities in this direction, actually and most are actually enabled by the latest improvements in AI, so most of the "old" thinking doesn't apply anymore.


>sharing hardware locally, from your house to your neighborhood, etc

It is interesting to think about, but I think economy of scale would still make data centers cheaper for this, if the "customers" are not getting the privacy anyway. Unless you could sell your excess compute for less than the data centers?

I do think that businesses would want their own onsite AI infrastructure though.


There’s a book for very young kids

“What if everybody did that?” - by Ellen Javernick

https://www.prindleinstitute.org/books/what-if-everybody-did...

I think the topic is great for 4-7yo and honestly wondering how it became a topic for grownup philosophy students and then top of HN.


go on..

Awesome series. This came up in a related post not long ago https://news.ycombinator.com/item?id=50000780

What’s the alternative to building on an assembly line? Is there a “theory” of the assembly line building relative to more ad-hoc approaches?

Moving the article being built means it moves to the machines that do each stage of the work. As opposed to moving the machines to where the article is being assembled.

I powered the house off a much less fancier 4XE Jeep, which has a generator, and so, arguably more versatile as you don't need spend time charging.


I just got flashbacks of stuff I forgot long ago like OpenStack and EMC Rex-ray and so many things like that https://www.openstack.org https://github.com/rexray/rexray


I am seeing a lot of openstack around since Broadcom enraged the vast majority of their vmware customers


Just deployed OpenStack by using the Kolla Ansible installer. Works great coupled to Ceph for storage. Thanks to Broadcom we are moving in the right direction. Proxmox as fallback plan but it looks like OpenStack is solid.


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

Search: