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

LLMs by themselves are not able to but you are missing a piece here.

LLMs are prompted by humans and the right query may make it think/behave in a way to create a novel solution.

Then there's a third factor now with Agentic AI system loops with LLMs. Where it can research, try, experiment in its own loop that's tied to the real world for feedback.

Agentic + LLM + Initial Human Prompter by definition can have it experiment outside of its domain of expertise.

So that's extending the "LLM can't create novel ideas" but I don't think anyone can disagree the three elements above are enough ingredients for an AI to come up with novel ideas.


You're proving the GP's argument - LLMs aren't creative you say as much, it's the driving that is the creative force

You can tell an agentic system. "Go and find a novel area of math that has unresolved answers and solve it mathematically with verified properties in LEAN. Verify before you start working on a problem that no one has solved this area of math"

That's not creative prompt. That's a driving prompt to get it to start its engine.

You could do that nowadays and while it may spend $1,000 to $100,000 worth of tokens. It will create something humans haven't done before as long as you set it up with all its tool calls/permissions.


Let me know when the Fields medal arrives in the mail.

It won't because even though it looks clever to you, people who /do/ understand math and LLMs understand that LLMs /are/ regurgitating

Why does your LLM need you to tell it to look in the first place? Why isn't just telling us all the answers to unsolved conjectures known and unknown?

Why isn't the LLM just telling us all the answers to all the problems we are facing?

Why isn't the LLM telling us, step by step with zero error, how to build the machine that can answer the ultimate question?


Here's a Fields Medalist commenting who doesn't seem to believe that.

https://x.com/wtgowers/status/2057175727271800912


Um - all I see is

> Timothy Gowers @wtgowers

> @wtgowers

> If you are a mathematician, then you may want to make sure you are sitting down before reading further.

If your refutation requires someone to have an account, login, and read something - it's meaningless


Try https://xcancel.com/wtgowers/status/2057175727271800912

it's readable to most, it's annoying having to swamp through ex-Twitter .. but there are work around's.


Thanks - I'll read that and the above linked OpenAI PR

But, I remain sceptical


The (linked by OpenAI) comment paper by various tangential mathematicians was the most interesting read from my PoV:

https://cdn.openai.com/pdf/74c24085-19b0-4534-9c90-465b8e29a...

it includes the longer remarks by Gowers & others.



I believe when we have AI Agents "living" 24/7, they will become creative machines. They will test ideas out their own ideas experimentally, come across things accidentally, synthesize new ideas.

We just haven't let AI run wild yet. But its coming.


So are self-driving cars - as they have been for the last... decade or so

AGI has been "just over the horizon" for literal decades now - there have been a number of breakthroughs and AI Winters in the past, and there's no real reason to believe that we've suddenly found the magic potion, when clearly we haven't.

AI right now cannot even manage simple /logic/


If that’s a requirement, aren’t LLMs driven by pretraining which was human driven?

Who decides at which the last point it’s OK to provide text to the model in order to be able to describe it as creative? (non-rhetorical)


So a mishmash of ideas from other papers(Not to downplay the results). This is exciting times of hackery and basically using puzzle pieces and piecing together stuff.

This is the kind of stuff that can only be done so quickly by having more and more people brought into the field to try these ideas out. The more people the more permutations of ideas.


128 bits*


I assume he means with the encrypted metadata in HTTP/3 / QUIC that it makes it harder as a security admin to "peek" at what is going on in the network.

In my opinion its short sighted, because if we care about security, then we should care about user security and privacy as well. Because if the security admin has the ability to packet inspect stuff, so does a potential malicious app.


Odd, surely SSHv2 already suffers from inability to inspect on the wire.


From the Github:

SSH3 is a complete revisit of the SSH protocol, mapping its semantics on top of the HTTP mechanisms. In a nutshell, SSH3 uses QUIC+TLS1.3 for secure channel establishment and the HTTP Authorization mechanisms for user authentication.

So, it has nothing to do with SSH2; more about HTTP/3-QUIC security theater: hostname is still being sent over TLS/1.3 negotiation.


To be clear, my reading of the parent post is that the grandparent doesn't like HTTP/3-QUIC making it harder to read data off of the wire (ie: for internal security analytics).

But I don't see how this is worse than SSHv2. In both cases retrieving the hostname / IP is obviously trivial since you just instrument DNS for the hostname and, of course, the IP is cleartext.


The owning organization or user should already have full admin on all endpoints.

Malicious apps and attackers should not.


More like incomplete state machine for HTtP/3-QUIc


Change up the model. That allows it to see previous frames and 1-2 future frames.

Then train the model on movies that are color and then turn them black and white.

That way you can train temporal coherence.


Probably the same sentiment between DLC vs Expansions in gaming.

But fundamentally no difference because you are getting a discounted price in exchange for the soft locked feature. So 2 and 3 are the same imo.


I'm not asking what's the conceptual difference, I just don't get what the two options are. Is #2 one-time purchase and #3 subscription?


For context Terence Tao is often referred to as one of the greats of modern mathematicians of our time.

He also has a Mastodon account where he sometimes goes over the implications of LLMs and in this post is his musings of how he sees it's current potential and possible impact in the near future for mathematics.

"The 2023-level AI can already generate suggestive hints and promising leads to a working mathematician and participate actively in the decision-making process. When integrated with tools such as formal proof verifiers, internet search, and symbolic math packages, I expect, say, 2026-level AI, when used properly, will be a trustworthy co-author in mathematical research, and in many other fields as well."


Thanks, I just hope this is sustainable. I like newsletters and the site design is simple and basic. I hope there's some way you can set up like a monthly contribution of like $1-$5 to pay for some of the hosting/API costs.

The main concern though for me is long term vision for the site. I'd really like it to stay user/community oriented and hopefully the users can pay for the costs with donations/contributions.


EM’s UI designer here. Simple and basic are some of the most common words when I explain my designs. And Hacker News’ UI was an inspiration.


Thanks - I appreciate the concern about the costs to run the site, but they're minimal for now and I'm more than happy to cover it. Might eventually try to monetize the site and upcoming newsletter, but there are better ways to do it than charging readers.


> there are better ways to do it than charging readers.

No, there isn’t. Let me guess, running ads?


It seems to me that they heard a suggestion for the site to be made available only to people who pay and the suggester was instead suggesting that there be a donation / "buy me a beer" link.

Indeed, there are ways to monetize such a thing that exclude both charging all of your users and selling advertisement space.


AI can efficiently teach your daughter how to solve math problems, but as a human, you can provide her with the context and understanding of why and what she's learning. AI is great at handling the technical aspects, freeing us to focus on cultivating our uniquely human qualities of empathy, creativity, and critical thinking. Letting the AI teach the how and the human teach the why and what can lead to a more holistic and enriching education.


I'd like to agree, but it does seem like AI is getting better rapidly at providing the context and the "why" too (as well as creativity and critical thinking)

the language models can tell you how to do a fourier transform, and then you can have a conversation about what it all means, why it's important, the discovery, etc

    [me] why are fourier transforms relevant to learn? what is the history behind them

    [bing] Fourier transforms are relevant to learn because they are used in many fields such as signal processing, physics, and engineering. They are used to convert a function from the time domain to the frequency domain, which can help to identify the frequencies present in a signal and in what proportions. The Fourier transform was first introduced by Joseph Fourier in 1822 as a way to solve the heat equation. Would you like more information on the history of Fourier transforms or its applications?


Me with not much background experience in programming not knowing what RoR stood for.

Me asking ChatGPT:

In this context, RoR most likely refers to Ruby on Rails, which is a web application framework written in the Ruby programming language. The second comment is suggesting that people had similar dismissive comments about Ruby on Rails (RoR) when it was first introduced, but it has since become a widely used and respected framework in web development. The comment implies that similar trends may apply to AI tools used for web design.


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

Search: