> How do people not understand that some laws only make sense at a certain scale?
Consider that any other laws you may want in place could be even worse, and what we have with AI is the logical culmination of technology and the laws we as a society have established over centuries of dealing with hairy issues based on sound principles:
Tl;dr: AI has harvested that which we as a society have very explicitly decided should belong to the commons.
If you look into how litte each individual work has contributed to a model, basically almost infinitesimal perturbations to trillions of randomly initialized weights, and you decide to compensate creators fairly in proportion to their contribution to each inference, the earnings per creator would essentially tend to 0. Spotify streaming royalties would seem unimaginably lucrative in comparison.
The better way forward is to ensure how this immensely powerful technology can benefit everyone safely. New forms of compensation will need to be evolved, for sure. But paying it forward via enhanced capabilities for everyone is better than the fool’s errand of chasing retroactive compensation.
My hunch (or intuition, hah!) is that intuition is an instinctive mental shortcut required to navigate large problem spaces that can’t entirely fit into our heads.
Maybe LLMs do not need intuition because they can scale their “cognitive capacity” with hardware and brute force their way through these problem spaces.
> Maybe LLMs do not need intuition because they can scale their “cognitive capacity” with hardware and brute force their way through these problem spaces.
My view is that is certainly true of smaller LLMs but becomes less true as they scale up.
To quote the parent bananaflag in a sub-comment:
> I believe the LLM weights have some internal representation of math in the same way brains do that allow them to produce proofs
I think as the sort of spare space adjacent to pure language processing in LLMs grows the probability of the sort of reasoning bananaflag is getting at (or spatial reasoning, or anything else) emerging in that space grows enormously.
One of the questions for AI development over the coming months or years is going to be if deliberately cultivating the architecture of those sub models for specific reasoning types beats any emergent reasoning mechanisms or not.
Hmm we may be talking of two different interpretations of intuition here. I agree that LLM weights contain representations of abstract concepts, as a lot of prior research has shown. This surely includes Mathematical concepts.
But to me that is analogous to what human brains do, and a bit different from intuition. I think of intuition as “heuristics”, typically developed through experience, that may link seemingly unrelated concepts via vague, hard-to-define associations, but which let us make mental leaps (or shortcuts) while reasoning. (Maybe analogous to System 1 / 2 thinking.)
On the other hand, LLMs can do both: build “intuition” from patterns in data AND brute force a huge amount of potentially unrelated concepts. This gets fuzzier when we realize that even these “concepts” themselves are gleaned from patterns in data! But my point is we necessarily have to take shortcuts to scale, whereas machines can scale with hardware.
This is of course a layman theory! But it could explain why these models are progressing so fast.
Yes, in my case "intuition" comes a lot from visualizing things spatially, manipulating them, and being able to capture their properties in equations/proofs, and it's that which is (currently) conspicuously missing when dealing with LLMs. (And may yet appear with world models).
With the alternate view of intuition that many of you are describing it is clear LLMs are somewhat either there or heading there now.
This is an intriguing observation! LLMs were famously bad at spatial reasoning, until Astra which apparently has a huge improvement. I wonder if that has any bearing on the recent jump in Mathematical performance?
One thing that struck me from Dario's last podcast with Dwarkesh was that he said training LLMs on a diverse set of tasks does not make them better just at those tasks, but they get better at unrelated and other tasks overall. What you described could be a concrete example of how that dynamic works!
Case in point, Russia has been fought to a standstill by its much smaller neighbor using asymmetric warfare. There is no reason Russia would not turn around and do the same to other countries.
One may argue it has already done so in the past via election interference, except now it also has AI in its arsenal, which won’t even need ICBMs to be delivered to its targets.
Dude, you have no idea about the modern warfare. The only country Russia is actually afraid of is the USA. And nobody except probably the USA or maybe Turkey is even remotely ready for modern war. Russia is having "military operation" right now. If they start the "war".
Second of all, the "much smaller neighbour" is any European country to them. Ukraine is the biggest country on European continent if we include the occupied territory. And it's actually has the most capable armed forces military-wise. Russia took over 116k sq km of Ukraine. All three Baltic states are 175k sq km combined and they literally can't defend themselves in case USA decides to withdraw.
And last but not the least, Russia's war is irrational. It makes no sense from any rational viewpoint. So relying on common sense in this case is wrong. You need to understand russian ideology and Putin's specifically.
Didnt they start almost immediately trying to setup like mineral processing facillities in areas they were able to nominally gain control of?
Wonder what the impetus for this was, were the elites embezzling too much or low oil prices or something that motivated them to pillage to make up the difference? Or typical greed in the absence of any asscovery...
I’m not quite sure what the disagreement is here. I was just saying asymmetric or other unconventional warfare can be extremely effective, as Ukraine’s resilience has shown, and there’s no reason Russia could not adopt similar means. In fact it has been doing so in the past. Somehow that makes me a Russia apologist, as sibling comment implies? Genuinely confused.
Just leave it. There are so many Russian apologists on this forum it’s exhausting. You’d think the fact that Russia is in the middle of an unwise and unnecessary war of aggression wouldn’t itself be an argument for concerns of more attacks on NATO countries and allies.
But to prebut whataboutism, I also don’t like American imperialism.
We may not like this, but let us contemplate what laws would be in place to prevent a thing like this; I suspect we would like those laws even less.
The laws at play here are related to Intellectual Property, specifically Copyright. Yes, it is terribly flawed, but it is the product of centuries of case law dealing with very hairy issues, and I believe it is fundamentally sound, and here's why.
As the name implies, it deals with only verbatim copies of works or subsantial portions thereof. It very expressly does not cover abstract things like concepts, ideas, themes, facts, or patterns, and rightfully so, because we really do not want anyone owning something that broad.
But these abstract things are precisely what have been extracted, at unimaginable scale, to build these models! Each pattern in the tokens derived from these works contributed imperceptibly tiny perturbations to randomly initialized weights, interacting in incomprehensible ways into vectors representing concepts and ideas and facts, the cumulative aggregate of which has somehow created a form of intelligence.
There is no copying, only gleaning, and so Copyright Law falls short. But what is the alternative, and do we want it?
To prevent something like this would require some sort of legal protection on the more abstract things. We do have a legal framework for those: Patents! But as is very clear on HN and in many Tech circles, those are an extremely contentious topic (even though they actually protect much narrower ideas than most presume.) I don't think anybody anywhere really wants any protection on broader abstractions, and rightfully so.
So: we as a society expressly decided these abstract things belong to the commons, and those are the exact things these labs harvested. This is probably the only logical culmination of our technological journey, and is within the very reasonable legal frameworks we have evolved over centuries.
As such, it is not productive to dwell on fighting this or bemoaning this. Instead we should focus on ensuring that this technology -- with its immense potential and opportunities and dangers -- benefits everybody as much as possible. That is a better way to compensate everybody's labor, and that is a much richer and fruitful discussion to be had.
I would highly doubt it, because news outlets and blogs rehash each others' reporting all the time. How many times have you seen an article start with "Today <XYZ publication> broke the news that..."?
Since this is a copyright fight, rights extend only to verbatim copies of the full work or significant portions thereof. Abstract things like facts, ideas, concepts, themes, and patterns are explicitly not protected, and rightfully so. Yet those abstract things are what get repeated and distributed, and are what get encoded into model weights.
This is probably plaintiffs' biggest challenge because it has been very hard to get models to regurgitate entire works except for a very small handful of extremely popular works (and now there are guardrails against even that.)
I should add that we must also consider the extent and speed of replacement. If the replacement rate is too high, employment levels can’t recover, both scenarios need to be avoided.
Technological innovation can be sped up, but it cannot be slowed down without mechanisms that create worse problems than technology. Adaptation is a political problem.
Right? I've been automating things and eliminating jobs using technology for over 20 years. No you do not need a team of five to manage your monthly reporting. We can build automated reports that are delivered automatically to everyone who needs them! No you shouldn't be managing all of this data in an excel file on a shared computer. I don't care if 90% of what Dave does during the week is maintain that file. Let's streamline that.
Suddenly it's very different when it's our jobs being automated away.
Yes, I don't like it much either -- though I continue to be absolutely fascinated by all this -- but I realize this is unstoppable, and I must trust that on the balance technology has been extremely good for humanity and the source of all our progress, and so I must adapt to a new future.
But I also realize the impact of technology, good or bad, entirely depends on how society uses it, and that is where our focus must lie.
Yes! The best part is, at any point they are right there to answer any questions, or to challenge or validate and experiment with any ideas! You can go down fairly deep rabbit holes in either case to understand a topic, and -- unlike any other medium other than a live human expert -- you can hone in only on the areas that matter to you most. Or you can just ask it to do something (as long as it's in the digital realm) and see for yourself what happens.
I always say that since one of the best ways of learning is by doing and given that with LLMs we can do so much so quickly, we are in a golden age of learning for those who really want to learn.
It's often been a pun combining the technological and political, at least in recent years. Last year's motto was Power Cycles. The previous year was Illegal Instructions. The one before COVID was Resource Exhaustion and the one before that was Refreshing Memories.
Any lawyers here who have used AI agents heavily for their work? From what I've heard, they're currently very good at searching, analyzing and drafting documents like contracts and patents, but some say they suck at interpreting the law.
I’m sure there are a wide variety of experiences out there, but here’s my perspective as a former biglaw associate and current solo litigator:
I have had some success using frontier models from the last 6ish months, but only when I can break up my work into discrete and verifiable tasks. For example, I had ~15k pages of discovery I needed to dig through for a summary judgment motion. Instead of just asking Claude to find the best evidence, I asked it first to run a clean, high quality OCR pass (it was almost entirely PDFs). Then I had it generate embeddings and write some reusable python scripts to make keyword and semantic searching easy for agents. While I was writing the brief, I would routinely ask my agent (Claude Code) to use both keyword and semantic searching to find the best evidence supporting whatever assertion I was trying to make. I trusted it because there were traces I could follow.
In other cases/situations, I’ve tried just giving a model access to all the docs and saying “write a brief arguing X,” but it’s always terrible at this. It writes briefs with lots of evocative jargon and rhetorical flourish, but a low signal-to-noise ratio.
Again, I’m sure others’ experiences differ based on workflow, legal area, etc.
Agreed. Six months ago, it was basically a gloried grammarly.
But lately, I’ve been taking hints from the “company brain” models, where it develops a running model of the case, and assesses each new piece as it comes in and updates the file.
I’ve also been using “Ralph Wiggum”-type models where you pass letter or contract drafts back and forth between agents with different goals (rules compliance, grammar, conciseness, ai slop detector, an opposing counsel critic, etc.). After a few rounds, it’s not perfect — but I start with a very good first draft in my hands.
They are excellent, especially the latest models. That said, (a) I wouldn't feel safe filing something without a real lawyer looking at it; (b) it can't (easily? legally?) do oral arguments for you; and (c) a lot can happen in the hallways outside the courtroom to move a case forward that the AI can't easily do.
In my experience it basically doesn't even try to interpret the law. It just summarises the publicly available law/guidance out there and, if there is a question about how to interpret some provision, might set out the arguments for each interpretation. It doesn't really take a position. It is pretty good at drafting though. (Legora)
Interesting! In other unrelated domains, models seem more willing to take a position. It may be nuanced, but they do tend to take a stance. I think it's good that it leaves the interpretation to humans, but I wonder if this is also some sort of a guardrail to minimize liability...
reply