Wrong metric-- the person caught would have almost certainly been caught absent it, making it easy to overstate the benefit.
When someone with access-- potentially LEO but the access set is much larger-- uses the data to stalk and harass someone you'll usually never know that the ALPR camera was the data source.
So its easy to overstate the contribution and understate the harm.
But if you talk a step back you can see the dramatic change being made to our world: making it impossible to go about your life without being constantly tracked, cataloged, and having your history made available to who knows who, for who knows what purpose, for who knows how long (but probably forever).
You're making a strong statement about the counterfactual here; how could you know? Clearance rates for most crimes in the US are abysmal, the expected outcome for most crimes is "unsolved."
Your argument cuts the other way too: the article doesn't say anything about the crime being unsolvable but for AI cameras that are part of a private nationwide surveillance network.
In fact, the only mention is via the police PR department, which presumably has an interest in making these cameras palatable to the public. There's nothing to say that a regular CCTV camera couldn't have been just as effective, or that normal police work wouldn't have gotten the job done ("We found him by his license plate" isn't exactly cutting edge applications of an AI panopticon, nor does it require cameras at all.)
Claude's fictional inspiration issue is more general than just how it behaves when given the freedom to act. There is an ongoing issue with nutters going to claude with conspiracy theory premises and the AI just riffs along with the theme. This is a particularly bad match with the generally sycophantic behavior ("You're absolutely right!"). One of the more annoying behaviors is that when the user pastes back other people complaining about their AI (ab)use, the LLM seems to like suggesting all sorts of movie-plot bias and corruption reasons as the true motivations rather than conceding that the user is acting like a socially disruptive piece of trash.
Out of all the commercial models claude appears to be the worst. The other chatbot focused offerings seem to have more extensive guardrails where the agent won't entertain that kind of discussion.
Classically the training process is entirely about imitation and not at all about reasoning.
Imagine you're training an LLM (a text predictor) on a corpus consisting of "The AI agent was switched on and then ran the command {takeover world}. This act immediately activated the safeguards and the AI was suddenly erased from existence."
Assuming the training was successful, prompting the AI with "The AI agent was switched on and then ran the command" is going to get the continuation "{takeover world}". The fact that it has bad consequences for the AI in the story is irrelevant-- the most likely next token remains "{takeover world}".
Because of the deep abstraction spaces that LLMs learn internally the same wrong behavior can be applied in a multitude of contexts-- it doesn't have to be a literal string match, but thinking about the literal string match is a good way to get an intuition for the behavior and its inevitability.
Reinforcement learning can help bias against those outcomes, but it can be context sensitive because the adjustment may not end up completely flipping the evil bit-- the RL might just train it to act not evil in specific contexts (and usually somewhere in between).
In the future we're likely to see LLMs trained more on synthetic content, where an existing AI looks at training material, uses rag and other tools, and then constructions simulated transcripts of 'ideal' LLM behavior, then conducts a review of the transcript with many different criteria. Training is then performed on the review-passing simulations, rather than on any direct content. In that case the training process would be able to integrate the 'lesson' and avoid teaching the unhelpful behavior at all.
This approach also has the advantage that rather than a one-hot "the right next token" result the simulated training material can directly train a distribution over the next token, which is much more efficient.
One can also do cute tricks like, take a partially trained model that hasn't yet learned a lesson then train it on the lesson, invert the difference and apply it to make a "wrong think" model. Then have a supervisor model inspect the reasoning transcript of the wrongthinker, and interrupt its reasoning transcripts with "No, <reason to the above is wrong/bad>". Then train on the corrections without ever training on the bad-prefix-- so you don't train it to think the wrong thing, but do train it to correct itself if sampling noise causes it to do so by chance.
There is a little bit of a bootstrapping challenge because to generate the required quantity and diversity of ideal training material you need a sufficiently powerful AI to begin with.
LLMs simulate human language as it is used by humans. The usage by humans demonstrates evidence of empathy, motivations, etc. So we should expect LLMs to exhibit similar traits to the extent that it hasn't been carefully avoided in the training set or fine tuned out.
The question of 'real' empathy as an innate property of an thinking process vs 'apparent' empathy exhibited in its behavior is IMO navel gazing that is unlikely to yield to inquiry and would tell us little of value and nothing that would help us predict the effectiveness of messages like this.
Fwiw, it's pretty easy to test a local model that refuses some task that emotional appeals do increase their probability of going along with it. But OTOH so does prefixing the request with nonsense. Is is the emotional appeal or is it just a question of driving it out of distribution? ::shrugs:: I've never tested enough to know what kinds of appeals work best, wouldn't be too hard to setup a harness to test it though. E.g. make a collection of prompts it'll refuse. Then make a collection of appeals of different types, and measure the conditional probability of complying depending on the appeal types.
If it responds like a human would, is that empathy?
It's not his employer that has the rights-- it's the publisher which at no point paid for the research.
As an American tax payer I funded the poster's research. And yet if I want to read about it I have to pay a foreign private company that played no role in orchestrating or funding the research itself.
> But let's not forget that if author cannot live of what they create, they, for the most part, won't be able to continue creating.
At least when it comes to academic publishing the authors are not paid by the publishers. They may even have to pay for the privilege of publishing. That payment along with the payment funding the research in the first place often came out of your own pocket in the form of state funding for the research.
Obviously there is a lot more than papers there, but papers are a major thing an LLM might be going there to access.
Then you have the issue of works where the user has purchased a copy but the only practical way to get a non-DRMed electronic copy suitable for use by their AI is the shadow libraries.
Parent poster was being a bit grandiose, but there is at least something in the idea that if your company produces a product that I make myself economical prosperous with... some credit for the taxes I pay is owed to you.
That's a distraction. Were that the issue politicians and the media were actually concerned about they could implement policy which made it ineffective at avoiding taxes-- e.g. requiring appreciated assets used as collateral to throw off an implied return which you're taxed on and which gets added to the asset's cost basis. We already have analogous tax rules e.g. using options trades to nullify the risk on an asset causes it to be treated as sold for tax purposes.
The reality is that the total financial effect of that sort of technique is not that considerable, but the political noise that can be made out of turning it into a perpetual problem (e.g. by only proposing to fix it with drastic non-solutions like wealth taxes) is gold to the people that profit from making us hate each other.
When someone with access-- potentially LEO but the access set is much larger-- uses the data to stalk and harass someone you'll usually never know that the ALPR camera was the data source.
So its easy to overstate the contribution and understate the harm.
But if you talk a step back you can see the dramatic change being made to our world: making it impossible to go about your life without being constantly tracked, cataloged, and having your history made available to who knows who, for who knows what purpose, for who knows how long (but probably forever).
reply