We are getting the "intrusive nanny state" regardless of financial incentives of a given healthcare system. That ship has sailed. It's a goal, not a side effect.
I don't think soft skills describes it in the traditional sense. The skillset largely needed with LLMs is more akin to being an editor or qa tester.
I suppose you could describe having the modesty to admit to yourself when you don't understand and research something deeper could be described as a soft skill, but I'd say it's a stretch. You are dealing with yourself in that scenario, not others.
Neither are skills that a large portion of users of those services pursue to any meaningful extent, I'll grant you that. They also certainly are not synonymous with the term "soft skills" as I know it. So I think I am on your side of the fence on that part any way.
I feel like if they were skills under a reasonable definition, you should be able to name an expert in these skills, and how do we know they are an expert.
But I think you can't. It seems to me, instead, one is better at googling/prompting the better they are in a particular domain, but it only applies in that domain. Like knowing a jargon is not a skill, knowing the domain is.
> you should be able to name an expert in these skills
There are certainly local “finding information online” experts in many families and social groups.
> and how do we know they are an expert
They are the ones who get mentioned a lot in conversations in the manner “I'll have to ask [name]” with the implication that [name] will look up the information or know it from previous occasions people (possibly this specific person and [name] is getting sick of them asking and not remembering simple answers!) have asked.
Soft skills include: handling change under ambiguity, critical thinking under pressure, self-awareness, prioritizing, motivating and guiding others without relying on authority, navigating disagreement constructively.
And of course if you point any of this out people go “Well many people are good at it, people getting bad results are just bad at it.” The constant refrain of “you’re just using it wrong” has become very tiresome.
If you can always say “just use it right” every time a critique comes up, then we aren’t having an honest conversation about the limitations of these tools.
It is impossible to duplicate results with an LLM. This strikes me as a serious barrier to calling it a proper skill. If you can’t even somewhat replicate the results you can’t really improve the input consistently. You can’t become “skilled” at it if you can’t even reproduce what you did.
If you enter the same prompt 3 times the results are of pretty significantly different quality. ChatGPT literally has you A/B test for them sometimes. They’re right to call it superstition - it feels like we’re making incantations and hoping for the best a lot of the time.
Prompting LLM’s still feels like a constant game of guess and check. At best you can argue it’s an educated guess. I don’t know about you but I didn’t learn math by guessing and checking, I frequently had to work backwards and review where I went wrong and/or I had the answer given to me with the work shown so I can learn. I can’t do that with a prompt. When I get bad results (which we all frequently do) I just guess what it didn’t like, try again, and pray for a better result.
Me typing into emacs is also not replicable, until I git push. Me with emacs and a good coding LLM can generate systems that I see as good, in my extensive professional judgement. For coding, for production grade code, the LLM will mostly be a tool in the hands of the professional. There is a new category of disposable code, which I think will be useful for many adhoc investigations; in neither case is replicability a serious requirement.
I do a lot of guessing and validating in learning maths. It is pretty efficient way to build that conceptual understanding. I even try to predict the next big theorem as I am listening to a lecture or reading a text book. It's engaging.
And honestly, I am enjoying learning this new way to make code I am pleased with. Using the LLM effectively and for quality deliverables is different from typing in many surface ways, but modularity of thought, iterative design and implementation, simplicity and generality, documentation, all still pay off.
These conversations always veer into “well I find it useful so I disagree.” I’m not saying anything about utility, I just think that system prompting is still very much hopes and prayers. We’re having a discussion about whether or not it’s a hard skill one can learn, I think that’s incredibly debatable
> It is impossible to duplicate results with an LLM
But that's the fundamental property of it - it is stochastic by nature. The skill is to learn how to sandwich deterministic logic between layers of randomness. Determinism doesn't live in the model. It lives in the harness you build around it. You can't make the model deterministic (it simply cannot be), so you make the system deterministic instead. Validation before, validation after, the randomness stays contained in the middle.
Talking to an LLM is not a skill, having a meaningful conversation leading to practical outcomes is.
If you ask me to write an email, all 3 variations will be more or less the same, except it will almost always improve with each iteration. 3 cracks at a prompt is 3 separate, unrelated attempts. None of them informs the other. In fact, you run the risk of making it worse if you include previous versions.
> 3 cracks at a prompt is 3 separate, unrelated attempts. None of them informs the other.
Thats where the "skill" part comes in. Like your 3 attempts at email that `almost always improve`. This is where you input the "previous versions", not directly back into the model. It's the "soft skill" of being flexible and adjusting based on how an entity responds to the input. Not learning how to adjust the input (by using previous attempts to inform the next) to more optimally direct the output given the state of the llm (chat? agent? model, effort lvl etc) it will seem chaotic.
The skills wont make an llm* deterministic, same as applying these soft skills to people. Give the same input to a person in `3 times separate, unrelated attempts` you are likely to get 3 different seemly chaotic outcomes. The "skill" is in being able to take what you received as output the first time, and make adjustments based on previous attempts while accounting for the state of the entity for the next attempt.
* In no way am I saying they are conscious beings or whatever nonsense by using people in the analogy. There are, however, parallels in how a set of soft skills (and this is why they are "soft" skills) can be used to get more optimal results from an entity that should never be expected to act as a pure function.
Anyone that has seen someone take a strong dose of datura can attest that what you are saying is all correct. It is not a traditional psychedelic in any way, shape, or form.
It's wild seeing someone fully interact with a reality that isn't there and then come back not knowing anything about what happened. I had a friend in the early 2000's that had us video tape and babysit him. He had a half hour long conversation with his dead mother on a telephone that wasn't there. Also, he chain smoked non existent cigarettes for hours. He knew we were there and interacted with us, but he had no continuity.
With photos, I could see a cryptographic solution. Of course it would still need some kind of centralized trust, but it's doable if people cared enough. It could be applied by cameras themselves.
This is already a thing. Leica cryptographically signs images. Useful for establishing trust for photojournalists I guess. I’m not sure how deep the chain goes. Do they have hardware attention down to the sensor? You could take a picture of a screen, but that would likely have some other tell-tales. Especially if Leica took another step like putting a depth sensor in the package and added its data to the signature.
A camera could be hacked with unrestricted physical access, but that doesn't make the suggestion unsound, it only requires there be a process for revoking trust in specific signing keys. This is already part of C2PA.
And the value of the signature is also going to vary by who claims it. Improbable photos signed by a random camera body sold to an anonymous consumer should be treated with more suspicion than one a newswire agency publicly claims, for instance.
A camera should not need to be hacked. What do you expect to accomplish by hacking the camera?
We should be able to move certificates on and off it because they would most likely expire anyway. So, you can get the keys from the camera, what then? You use openssl to sign an image that shouldn't be signed... what then? You do this enough and get caught, you lose your cert and can never pass the kyc to get another one.
Then every picture you used it for in the past would start showing a big red exclamation point with a note, "This is a scumbag user known for forging images".
Same way it works for tls and code signing. Public reports bad actors to the signing authority. They and/or independent firms investigate. You self report theft. It's not like we are inventing pki from scratch here and wondering what an implementation would look like. We have decades of use to look at.
Besides wasn't your argument "hacking" a minute ago? So you concede that then? You seem to have moved on.
Yes, "the public" includes experts in the subject field and independent witnesses to events.
If your argument is that images can't be verified, you are going to need to provide some kind of backing to that. We know well that there are watermarks and tell tale signs to go off. Regardless, if they are depicting real world events, outside confirmation can be used.
You also don't seem to understand the point of the signature. It's not to prevent fake images, it's to tie an identity to images so that that trust can be established and permanently lost in the case of a malicious actor. Your arguments repeatedly have failed to engage with the actual system proposed.
> You also don't seem to understand the point of the signature. It's not to prevent fake images, it's to tie an identity to images so that that trust can be established and permanently lost in the case of a malicious actor.
It was proposed as a solution to: how is any future viewer going to tell your images from "AI" fakes?
You seem to be saying it is not a solution. I agree.
I have already said, it's done by establishing trust by tying identity to images. Your assertions aren't a counter to that. You thinking they are shows a lack of understanding of how to verify truth on a fundamental level.
Let's also list the points you have conceded by quietly giving them up:
1. Hacking the camera is irrelevant.
2. Images can be verified.
Do you want to actually engage with the conversation? The funny thing is that there are engineering challenges in this space, but you don't seem to understand the basics well enough to even get to that point.
> I have already said, it's done by establishing trust by tying identity to images.
You've said it, but you've not credibly substantiated it.
Imagine a famous photo journalist dies and there emerge "AI" fake pics signed with creds stolen from his hacked camera. Trust isn't part of a solution. It is part of the problem.
Then the cert is pulled with a date and the timestamps on the images can be used to verify when they happened. This is generally how you'd handle theft. You can then also resign with a new cert at a later date.
Besides, after a famous journalist dies, their work is already famous, people already know it or don't. This is a contrived and irrelevant example.
Again, this is not hypothetical, we already know how this works. Next one please.
And that's problematic if the certificate is also stolen at the same time, but that's why there is an authority to validate issues and address what needs to be done to resolve them.
The premise is not that it couldn't be faked. It would be more practical to remove the key from the hardware and just use it to sign images if you wanted to do that.
The idea is that a centralized source of trust would revoke certificates belonging to bad actors or those that were stolen.
Cool. Every time I want to pass off AI images I'll buy a new Nikon camera. Always a Nikon until their cert is revoked. I'll short their stock before that.
The foundational failing I see in my countrymen is the notion that China can't innovate. Their economic reliance on exports is not untrue. Just like their reliance on energy imports is a fact (though one that we are seeing right now become more questionable.)
The average American will keep shouting over and over again that "China can't innovate" while they quietly and slowly tick off every box on their checklist of issues. It doesn't end well for us.
So your professional opinion is that the attack surface of mobile banking apps is limited to tokenized payments? Honestly, I'd be appalled if tokens were routed through my banking app. There is no reason the local client needs that data.
My professional opinion is that APKs can be de-compiled regardless and that has nothing to do with tokenized payments themselves which are like you said handled through server-server communications at the payment processor level. Your phone simply sends a one time use token to authorize the transaction.
What exactly was my example? You agreed with me already. Kind of odd to flip around now.
> which are like you said handled
Also, I corrected your misstatement about payments. They have absolutely nothing to do with decomp, and certainly can't be used to dismiss the entire attack surface of a mobile banking app. Your example is complete nonsense.
Pretty sure most people have seen their page at least once, so they know, Cloudflare is what they see if nothing is working (because the actual server is overloaded).
Yeah, this is the point. I don't understand the bulk of this conversation. Copyright doesn't matter, the books themselves don't matter. All that matters is that their corpus of training data grows faster than their competitors.
