Hacker Newsnew | past | comments | ask | show | jobs | submit | emil-lp's commentslogin

I use Perlin noise in all games I make, and in my research. It's marvellous!

+1 to Perlin.


Great work. It looks as the space in "No. 12" is too large. Is it possible to distinguish between space and sentence-end-space?

Good eye! The space in "No. 12" would traditionally get a thin space [0] in professionally typeset documents, which is smaller than the standard space used for inter-word and inter-sentence spacing.

The solution here is to fix the sample text itself to use U+202F (NARROW NO-BREAK SPACE,  ). As opposed to the standard width no-break space it currently has for "No. 12" and "Fig. 7".

0: https://en.wikipedia.org/wiki/Thin_space


Not quite. A "thin space" is thinner than an ordinary interword space, which in turn is thinner (or at least less stretchy) than an intersentence space. The space in "No. 12" or "Fig. 7" or "Mr. Jones" or "cf. page 5" is by definition an interword space! The problem is that HTML renderers tend to prefer to set it as an intersentence space instead, because they think "." always indicates the end of a sentence. Using   can help; that's like "~" in TeX. TeX also supports a breaking-but-never-intersentence-sized space via "\ " (backslash space), which is what you'd probably want to use in "cf.\ page~5." AFAIK, HTML/Unicode have no equivalent of that.

The "standard" method would be to double space after the end of a sentence. This is how "filling" paragraphs works in org-mode to avoid breaking a line in the middle of an abbreviation.

No, that's a convention from typing.

The actual approach used by a printer would be to be more generous with spacing material at the end of a sentence.

The relevant TeX commands are \frenchspacing or \nofrenchspacing.


It was accepted in a top journal for seemingly no reason?


Well, in this case, it's your pension.


It isn't my pension, and if it was my pension the major issue would be that I was relying on people I think are untrustable with money to fund my retirement. That is one of those ideas so strategically bad that no tactical success or failure matters. I personally wouldn't choose to let more than a fraction of my money get pushed into that vortex. It looks like a disaster waiting to happen (hopefully I'm wrong and everyone goes home happy).

As a rule of thumb in life, if someone is managing your money then you should by and large agree with their judgement of the markets.


> But as always, the methodology is buried: how many open problems were tried until they found a success?

Not only that, but they have like 500 world-leading experts in mathematics and IMO alumni, so how do we know one of the agents wasn't hardcoded to return a proof that the mathematicians had found?

I'm a mathematician/graph theorist, and I've tried ChatGPT 5.3, 5.4, 5.5, and now 5.6 on a bunch of simple-ish open problems, and I've never gotten a solution.


I’ve had a similar experience in physics. Excellent domain knowledge and semantic search, but the intellectual sparkle and reasoning just isn’t there and it still often throws out a lot of wrong ideas. It is very useful for coding.

But there is a discrepancy between (the implication behind) these reports and how I subjectively feel talking to LLMs. Granted, I don’t have access to whatever cutting edge model is out there for as many credits, but I also don’t feel like I’m talking to an IMO silver medallist.


What is a "simple-ish open problem". I guess it falls under "solving this open problem in mathemathics is left as an exercise to the reader" ;)


Proof of work doesn't help if the abuser is massively decentralized and is using other peoples moneys.


> if the work of a programmer can be automated then this means that any work can be automated

This is a very typical programmer thing to say.


i think most things can be automated, but not everything

i think AI will just replace jobs with new jobs, and humans will continue with open-ended goal setting, and probably jobs where being human is necessary: responsibility and authenticity, so judges, politics, leaders, chefs, etc.

so in a way, the above commenter was correct, anything a programmer can automate, AI can do


It is a little bit suspicious that the big replacement didn't happen, yet.

If it was so valuable then AI companies would directly create valuable things, not just sell API tokens.

A proof like this is a good effort at trying to provide value directly, but still far away from real use.

When OpenAI cures cancer I'll accept the fate, but till then I'm still seeing a lot of gambling going on.


Reminds me of crypto. A lotta tokens. A lotta gambling. Very little supplanting of crypto from fiat. Very few use-cases that revolutionize society, etc.


LLMs are genuinely useful but the cult surrounding it makes it super sketchy.

Tech needs no anti-humanist and low empathy cults. It's a disgrace to the technology.


There's really no good proof system mature enough to do advanced graph theory. The leading library in Lean is Graphlib, and it's really not ready for research level theorems.


How many tokens would it cost to write some library functions to fill in the gaps?


You could try solving that in Lean perhaps


what kinds of proofs would it be good at? I thought that combinatorial proofs would be easier to reason over than ones that required analysis


Graphlib? Do you have a link to this for me?


The prompt was released, but not the cost of the result.


Assuming all 64 subagents were running for a full hour (the tweet states just under an hour):

  Throughput                    Output tokens   Output cost
  ----------------------------  -------------   -----------
  40 tok/s  (5.5 low)                   ~9.2M         ~$275
  55 tok/s  (5.5 base)                 ~12.7M         ~$380
  70 tok/s  (5.5 high)                 ~16.1M         ~$485
  750 tok/s (Sol Fast, $75/M)         ~172.8M       ~$13,000
Claude estimates that tool use / input tokens might add 10-15% on top of that depending on exactly how the model went about the task.

Edit: better tok/s estimate buckets based on GPT 5.5 actual speeds since I couldn't find real benchmarks on 5.6 published anywhere. Also account for Sol Fast pricing.


Sol fast isn't the Cerebras 750 tok/s version, it's just 1.5x speed at 2.5x price

I assume they didn't use the Cerebras version for this since it's probably very supply-constrained right now


But Sol is running on Cerebras. That’s the whole point of this. That’s how they get 750 tokens per second. There is no other way.


Regular Sol does not run on Cerebra’s. I don’t think anyone public has access to that.

https://x.com/thsottiaux/status/2075596669958472146?s=46&t=Z...


Yeah they posted an update below it

> apparently that is just Sol being Sol on fast mode, not 750 tps. :x O.o

> This is real and it is not 750 TPS. Anyone with 5.6 Sol Ultra on fast mode can reproduce this! It’s all GUI interactions with CUA. No MCP or bpy needed.

https://x.com/kimmonismus/status/2075493505011482922?s=20


Who’s they, who’s chubby, why would I care.


And not how many times it was prompted before it returned a working solution.

Or how many prior variants of this prompt were tried.

Or if proof checking software was used to hone in on the final winning prompt / LLM output.


Statement of AI use. The proof in this note is entirely due to GPT 5.6 Sol Ultra and the writeup with Codex (with GPT 5.6 Sol).

Clearly that sentence isn't AI generated ...


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

Search: