you can have a TOTP variant which has the browser display a challenge string & QR code, authenticator generates a 5-6 digit response. include more entropy in the response if you want to do away with the password entirely.
passkeys are trash, I would never opt for dedicated hardware (or a cloud service) and using them with a password manager is extremely messy when you want the password manager to be properly isolated from the rest of your system.
by buying a mass produced generalist device (preferably from a randomly chosen vendor) and then restricting it to a specialized purpose you are effectively negating a targeted attack.
I always assumed the people buying those crypto gadgets were idiots.
also, if Apple falls to a supply chain attack like this where their laptops are backdoored crypto theft will be problem number #10.
in order for your model to accomplish that, you would get very rich.
there will also likely always be more. in the limit in order to get an edge your model would start to infer insider information. for example, it's common knowledge by now that satellite imagery is used to measure car numbers in parking lots, that's a proxy for insider information.
so it's not even so much the model as it is the data.
even being able to forecast weather better than publicly available methods can be leveraged to gain a significant edge.
this is why any detection armsrace will have an ending. where the classification classes will be indistinguishable.
for some reason too many people assume it will go on forever. humans aren't perfect, humans have variance. AI systems will nestle in the variance. the same applies to any camera (and lossy compression) you put forward.
even in the event that perfection exists, an AI generator and a discriminator can asymptotically approach it. and it assumes there will be no variance, which isn't realistic (even lossless video has sensor variance).
since GLM 5.2 (perhaps even earlier?) it has been remarkably easy to reverse engineer any closed protocol you want as long as you have a binary. previously, it required so much manual effort as to simply not be worth it 95% of the time.
now all you need is IDA or Ghidra MCP, a binary and some vague sloppy instructions.
some more recent models even started instrumenting a running binary (when possible) to enumerate the protocol without being explicitly instructed to, which is even better.
You don't even need an MCP, I just let the agent write scripts for headless ghidra. The MCPs are fragile and there isn't a whole lot of knowledge about how to use them in the training datasets, whereas there are plenty of ghidra scripts (and proper docs for the APIs).
Seriously, though, the popular Ghidra MCP is really badly architected; it's way better to rearchitect it or just script Ghidra directly. With that said, Opus 5.5 seems to have been trained on CoT from the popular Ghidra MCP. This makes it work better, but also makes it even more inefficient if you modify the MCP without changing its name and shape significantly (it will try to make tool calls using the "mainline" format, then have to retry them when they fail).
Even with Opus 5.5, IMO it's better to just ditch the MCP and let the LLMs eat with bintools and headless Ghidra; with both GLM and Opus this produces significantly more efficient results than the popular MCP. On the other hand the IDA Pro MCP is much better architected and seems to be pretty good.
I will note that Opus 4.8 and below was unable to use the MCP very well, misused it really often even with instructing. Good to see that they have improved some things.
Reversed an old video game pretty extensively for me. Got it running in 30 minutes, then patched some bugs, reversed the entire game logic VM and rendering pipeline, and we're now in process of expressing game logic as Lisp and upscaling assets for rendered that it already made to operate at 4x its hardcoded resolution...
Starting point was literally original installer copied over from a CD, which I hold on to and dug up from an old hard drive, plus a short prompt asking to make it work. Installing Ghirda and developing a whole framework for patching the binary, as well as substituting DirectDraw with its own DLL shim, was Claude's own invention.
It depends on what you're reversing. It tends not to like DRM-shaped things or security-sensitive things (for a very vague definition of "security-sensitive"), but e.g. game engines are fine. (For the things it doesn't like, dropping down to Opus 4.8 or 4.6 usually works)
Works fine for me. I didn't explicitly say decompile but it had no issue digging into executables to give me information on how it and various file formats were designed.
> there isn't a whole lot of knowledge about how to use them in the training datasets
if you include a SKILL.md for the MCP (or just dump it into the prompt) it's not a problem.
I don't use Ghidra but IDA Pro MCP works extremely well for me for all manner of tasks. for example, some software likes to call home for license checking (and I wish to run it with networking denied to it). it no longer does.
Without admitting to anything, I was surprised by how little time it took to add support for the new protocol to the library. And I didn't even need an MCP server.
I've had good luck aiming for easier targets. Android APKs, Java, and .NET binaries are usually easy to reverse engineer with AI and without any advanced tooling.
those benchmarks aren't actually indicative of capabilities.
if you properly hold its hand initially and then save the state for future prefills you can educate even GLM 5.2 to be pretty much everything you need.
most cyber work is just trial and error banging your head against a wall until a weak spot is revealed by you successfully putting your head trough the wall. you can offshore this work to an LLM.
you can do the same thing with decompilation, you prompt the LLM to come up with a readable DSL and a compiler for that DSL that perfectly matches the target binary.
just get an LLM to think and then force it to output a specific json with prefill post-think.
make sure to include good conditioning text in the prompt with examples of exactly what the output should be like. you don't want dissonance in the probabilities on the prefill.
A classifier is a subset of generative text, so I think responses like this miss the point. Jev is cheap enough and fast enough to sprinkle across your app in ways that LLM would be infuriatingly laggy and unnecessarily expensive, and it's never going to be injected to provide a sorting algorithm in python.
The point isn't that new type of problem has been unlocked, rather a new approach that can unlock new use cases.
when you want a machine to reason about the prompt and generate a structured output not using an actual LLM makes no sense. I have been doing it since the first chain of thought open models became available.
perhaps there may be a way to get a Jev-type model to think for a very specific number of steps to gain control over its latency, if so that would be the next step. truncating LLM thinking like this does not work well, and its thinking isn't efficient anyway.
I always found it odd that people who are intelligent enough to work in IQ-loaded fields are as easily manipulated by either rhetoric or ideology as anyone else.
growing up I always assumed that everyone would grasp some aspect of game theory intuitively, I still remember the day I found out game theory was a thing - it was like finding out that someone successfully systematized common sense.
most people take the things people say as if they were worth considering. signals without cost are only useful as knowledge of what the signaler wants fools to believe.
> I still remember the day I found out game theory was a thing - it was like finding out that someone successfully systematized common sense.
Game Theory as a field is actually very complex and often reveals dominant strategies that would never occur to someone as the "common sense" approach to a problem. It was arguably created to solve for problems where there was no obvious correct answer, even for a very intelligent person.
The world is not full of idiots, rather communication channels are heavily managed to prevent people from establishing consensus around the obvious.
No matter how many people think press release X is obviously wrong - 80%, 95, 99% - it'll always be "the comments section criticizing the article," or "bloggers sharing incisive criticism."
If you know how your emotions and opinions and neurology can be used against you and don't use that to inform your media consumption habits then that's just stupid.
There's no safe way to watch propaganda. There isn't even a safe way to watch it secondarily: an endless stream of debunking videos pretty quickly still does the main thing propaganda needs: repeats the core message.
Perhaps it stems from the the mind projection fallacy[1] (your comment is an example of this :) ) where in these people are so enamored by the rhetoric that they fail to consider the underlying game theory involved which needs second or even third order thinking. This needs some self awareness.
I would say much of the futility of debate and discussion originates from [1].
Depends on the angle. It's easy to manipulate someone once you exceed their ability to keep up with you. People "living in symbols" aren't easy to manipulate about easy things, but you can ramp up complexity until they lose track and accept an unwarranted reasoning leap without realizing it.
With people "living in dirt", you can't pull that off because you'll jump ahead so far they'll immediately realize they can't possibly understand what you're selling right now, and shut the argument down. But they can get confused about things outside their direct area of direct experience.
Note: I'm not saying either is smarter or less smart. Rather, "living in symbols" get confused "higher up", while "living in dirt" get confused "low and to the sides", but they probably have more solid grounding.
I think it’s more precise to say it’s easier to change the mind of someone who lives in symbols than who lives in dirt. Whether that is changing their mind for the better (oh new maths proof, I’d better update my priors) or the worse.
Surely it depends on what you're trying to do? If someone is deeply knowledgeable about how AI works, then it will be difficult to fool them with misinformation about the capabilities of AI. Someone who knows very little about AI will be more likely to have misconceptions about AI.
the problem here appears to be that police officers are generally of average or below average intelligence. combined with the fact that they face no consequences for incompetence, you have the perfect mixture for nonsense like this happening.
I expect the solution they will settle on is to form a department for interacting with technology like this, who understand its outputs are not edicts.
you can't really expect any meaningful police reforms (such as liability and insurance) because anyone who is in any position to enact them is perfectly fine with dumb muscle that holds a monopoly on violence and is deferential to them. in fact the public disliking the police is an ideal outcome for those decision makers as it gives them more leverage over the police.
The problem is that they're treating the output of the AI algorithm as 100% fact. Countless examples of Flock matching incorrectly or police otherwise using the match found by Flock as "here's who to go arrest" and not "here's a starting point for further investigation".
Partial matches, correct matches but no investigation into things like paint transfer, it's happened way too many times (hint: one is too many times for things that can incarcerate people).
It's like they're speedrunning unworthiness to hold the monopoly on violence. This is how you get riots and revolutions. Sadly, today's Western world is complicit and complacent. Nobody who is pro-surveillance state will change their mind until it affects them or someone close to them.
the cold political war turned hot with the prosecution of Trump (he made it easy by being an idiot). there will be no considerations for retaliation now, both sides will be using whatever means they have available to damage each other.
also if you look at how the two factions separate from each other decade by decade (there used to be overlap, almost none now) it's obvious that there will be no reconciliation.
it's a musical chairs game at this point, whoever is in power when the democratic facade cracks will be able to seize power and transition into whatever single party state follows.
I think the blue team has been in a cold civil war since 2016. They can't come to terms with their progressive hegemony ending, but also can't pivot to a new politics because elections are too close to risk a mass preference cascade needed to revitalize the party.
I think the red team is only now starting to come to similar terms, and will increasingly do so as the blue team continues to maximally exercises power against their opponents instead of revitalize the tarnished reputation of instituions.
Around 2030, the US will need to spend 100% of it's tax revenue to service the debt, and California will need to a federal bailout or risk either default or wiping out their state pensions. I believe the coming financial disaster will likely be the catalyst that quiets the 'nothing ever happens' crowd, and both teams will need to reset their politics to the new world, distinctly different than the post-WW2 open society liberal consensus.
The cold political war turned hot when Trump tried to overthrow the US government. I understand that this bizarre both sides framing is comforting to you for now, but it's just going to leave you even more alarmed when he tries again.
are you seriously equating empty rhetoric to legal action?
Trump was an idiot who made himself an easy target for nobodies with no strategic bones in their body to go after him. note that his guilt or lack thereof is irrelevant, what matters is that legal action was initiated against a past president. this one action doomed the republic just like it did the previous Roman one.
you can have a TOTP variant which has the browser display a challenge string & QR code, authenticator generates a 5-6 digit response. include more entropy in the response if you want to do away with the password entirely.
passkeys are trash, I would never opt for dedicated hardware (or a cloud service) and using them with a password manager is extremely messy when you want the password manager to be properly isolated from the rest of your system.
reply