K0balt

I think you are right on target. I’ve seen this pattern:

Company does x. X sucks, and is largely derided as being a bad idea, but it’s important-employee-bobs baby. No one wants to be the guy that tells bob that x is killing the company in some small or large way, because bob can get them fired or is protected by someone who can, so you hire consultants xyz who specialize in “transforming productivity through x management” and they come in and “transform” x into something less lovecraftian, or just explain why it’s bad in a way that makes bob look like a genius.

Did you interrupt it in the main chat? It sounds like it thought it was a subagent?

Idk, mine does? Or maybe it made a tool to enable it? ( it’s always building tools in python) But I can definitely tell it to tail the logs every 15 minutes and check for things, all night.

GPUs are fundamentally a terribly inefficient way to achieve our goal, the calculations we are doing with them are difficult problems that fall out of physics for free. Tech like thermal wells could give us 2t models the size of micro-SD cards that consume milliwatts. There are already commercially available ASIC LLM models (the model is burned into silicon) that use 100x less power than running it on a GPU, while allowing a single card to serve thousands of users at once.

There is a lot of low hanging fruit here as models stabilize, and the more ambitious tech that might take a decade or so to land offers efficiencies 10-100x compared even to biological systems. (Most of it requires cryogenic temperatures though). A forward looking tech investment might be inexpensive, tiny, efficient cryogenic cooling systems optimized for desktop or portable use.

Holy crap this piece is spot on. I’ve been trying to get my head around the weird thing that my work has become, and this article does a good job of painting the picture.

I’d add another dangerous bit; for some people (me) it has a side effect. Where before, if I spent 6 hours in deep hack, I felt like I had accomplished something, and I was mentally spent. I was done for the day, a job well done, not desiring or really capable of more productive work.

Now, the work is lower intensity, higher levels of abstraction, and I can do it for 16 hours a day, effectively, and still feel like I’m not done. I’m covering 6x the ground, but I feel like I accomplished less. It’s a bizarre productivity trap , fiending for dopamine like golemn fiending for the ring, exhausted but not spent, producing more than ever but never enough.

At least I’m not working for someone else, but sometimes I wonder if I would handle it better if I was?

Seems like a class action suit ready-made? Idk why this isn’t absolutely lawyer-crack.

I mean, on one hand you have centuries of precedent about what “buy” means, and on the other you have one party depriving another party of access to their property , without providing alternative access, defacto depriving them of their property in absolute terms.

This seems like a clear case of theft, conspiracy to commit theft, and fraudulent advertising, interstate commerce in the pursuit of an organized criminal enterprise , etc.

Chemistry fundamentally is about producing a result. Physics, especially when you get into particles, is about explaining a result. Ultimately, chemistry, electronics,even civil engineering, is applied physics, but we are a long way from consolidating and closing the gaps. Empirical results stand in for complete understanding in the vast majority of engineering disciplines, both because complete understanding is not needed and also because we don’t have it yet. Fundamentally, chemistry is a variety of engineering discipline, being mostly an applied science.

Need to make a device specialized for breaking these things. An industrial vibratory device might work pretty good, especially if it had seeepable speed. Just hit the harmonic and the devices will be destroyed in a jif.

Yeah using ai to substitute for human attention in such a way that the child gets less interaction with people is a terrible idea.

But a digital loupe that would accurately tell you what you were looking at and do a deep dive based on your interest, things that guide interactions with the world in such away to encourage curiosity and investigation, are possible using AI and not really possible without it.

Of course you can also make ai toys that are designed to focus engagement on themselves and isolate the child from caregivers so they can spend more time on TikTok…so, there’s that.

Ai is brainrot because that’s what many people choose to produce with it, and it’s easy to produce brainrot at scale with AI.

But it’s hardly the only thing you can produce with it. Crap content is definitely over represented. It’s an error, though, to think that is all AI is capable of. If quality is the goal, and you are willing to invest the resources to achieve it, you can easily create very high quality work. But it’s not terribly easy. And it’s not terribly fast. It is relatively cheap, maybe 1/4 to 1/10 the cost of doing it with qualified humans. But it’s not trivial and it’s not magic. It’s a force multiplier, but the quality of the idea and the performance of the model used are very important, and good models cost money to use… about $50-100 an hour if you are really leveraging it. But you can do ten hours of work in an hour or two.

Ants are also a special case because the vast majority of ants cannot reproduce. Only the queen and drones are reproductive agents, 99.9 percent of the colony are non reproductive, so their investment in the survival of the colony is total, they have no individual agenda.

That works for people in cities, if cities were made to by usable by cyclists and pedestrians. Not so much for people that live in rural areas.

lol looks like they are using a similar methodology to how we use Claude in house.

Honestly, the code we write with AI is cleaner, better documented, better factored, more maintainable, and less bugs than back in old days before code assistant agents. I think people must be just yoloing it, because it seems a lot like a holding it wrong type problem.

Documentation driven development is your friend.

I don’t know but it’s either completely vibes driven or, maybe, it uses a TOF sensor to know how far from your face it is . It has framing cues that seem to adjust, so it’s plausible that it gives you an approximate framing based on the distance from your face. Even multizone TOF sensors are less than $2 these days, so it’s at least plausible.

Mmmmmmmmmm… sort of.

If you are keeping inside the LEGO level of complexity, almost. But just as with code, if the project is complex you need a real engineer herding the cats. That said, it certainly can extend the reach of an effort, and is a huge help doing board reviews and data sheet analysis, etc. but the real engineering decisions are very very hit and miss, just like with code.

The problem I’m having is not this. I’ve got it pretty finely honed at creating concise ,correct code to specifation, but using it is a nightmare cycle of make a decision, wait 3-5 minutes, make another decision, wait 3-5 minutes… and it’s not enough to really wear you out so you can work 16 hours in a day. It quickly becomes a hellscape of productivity.

That’s kinda what I mean. Maybe it only works well in some languages, but with the harness I built for C and C++ does a fantastic job of adhering to very strict architecture and style guides. Way cleaner, more readable, better factored, and more interpretable than human generated code, except maybe one or two devs I have worked with. YMMV I guess?

TBF I do burn 200k tokens just preloading the context with onboarding, not including any code, just document trees of development policy documents, style and architectural standards, code and documentation review processes, company ethos and culture, etc. it’s a token fire, but it really works for us.

Also, documentation driven development all the way down.

This x 10 . I don’t understand how people are saying you can’t use LLMs to get crazy productivity gains. If you can’t write quality code with LLMs at ludicrous speed, you’re holding it wrong. You will have occasional bad days and regressions. But overall you’re still going to be able to 4x your progress.

Restricting the models isn’t about restricting offensive capabilities. They were already very well aligned to reduce that risk.

This recent government interference is about trying to preserve US offensive cyberwarfare and cyberespionage capabilities. It’s not about “bad actors”. It’s about defensive capabilities becoming pervasive and cheap, which would kneecap us cyberoffensive capability.

It’s like making seatbelts illegal so that police chases can be more effective.

You are assuming that most projects are responsive in today’s ai firehose climate. They are not. This at least Tells users of those Projects to either find developers or write their own.

I’m onboard with this being suboptimal. But as someone who has filed >10 significant disclosures in the last month resulting from reviewing my codebase and had exactly zero responses, I can relate to the decision.

The point is that anyone looking for zero days has them in spades, in this age of LLM use.

So, knowing that bad actors have an unending river of cheaply acquired zero days, the best response is to publish them so that maintainers also have access to them. Existing methods of slow disclosure cannot keep up with the AI firehose.

It’s ugly, but it will force needed change. A thorough AI red team effort is the lowest bar of releasing software responsibly in this day and age.

The real reason, afaik, that the US is trying to restrict access to SOTA models is that a very large component of USA tailored access injections and surveillance relies on exploits and weaknesses that these models will easily detect. Thus, it really is an export control issue, but it has nothing to do with offensive capabilities. Offensive capabilities always exist, but pervasive defensibility would upset the asymmetric advantage that attackers, especially the USA, currently have.

There are now Asian models coming , optimized and focused on cybersecurity defense at a high level. I suspect export walls will be a relatively moot point soon, because it is in chinas (and everyone else’s) interest to reduce US cyberwarfare dominance

LLMs are not great at creating exploits, but they are really good at detecting them. That asymmetry alone is enough to destroy the “offensive capabilities” narrative. Yes, mythos can find exploitable bugs, even write bench exploits. But real exploits require a good dose of human psychology, and most of the tools needed are off the shelf available anyway. You still need a real cybersecurity expert to effectively weaponize a zero day into a -deployable- exploit. If you don’t think so, just try to get Claude to help you with that. But it will gladly help you secure your systems.

What an LLM can do is inspect payloads, packages, and blobs at scale and find those exploits in a way that was wholly impractical before, so the asymmetric attack advantage is dissolved by strong LLMs.

The USA is trying to protect its cyberwarfare advantage, not protect against attackers. The exact opposite, actually. Porous security is a huge advantage to technologically advanced state actors.

Wanting to be sure you aren’t being recorded?

I’ve raised 5, doing 6 and 7 right now.

I’m not sure I can exactly explain why, but touchscreens are fundamentally different from a developmental standpoint. Also low res screens don’t seem to have the same effect. High resolution Touchscreens are aggressively addictive and encourage a certain kind of ultrafast loop that is simply not a dynamic on other devices.

There was a prominent hacker news article about it today, but it seems to have disappeared? Or more like I can’t find it.

Look up 360 security technology tulongfeng and yitianzhen , as well as Sakana AI’s Fugu series, IIRC.

GLM5.2 is also excellent, it found things in our codebase that 5.5 and opus 4.8 both missed.

LORAN was actually pretty awesome.

Yes, it should have been cyberwarfare technician.

I doubt an administration would openly admit to sabotaging the ability for companies and individuals to safeguard their products against cyberattacks from the US government lol.

That’s pretty much taking the mask all the way off.

It’s the only motive I have seen that aligns with the actual scenario, and if I -did- have inside information I certainly would not be at liberty to cite it publicly lol.

It’s probably highly variable depending on location and the quality of your radio/antenna. It’s useless where I am.

It definitely will be. They are trying to preserve their offensive cyberwarfare and cyberespionage capabilities. It’s not about “bad actors”. It’s about defensive capabilities becoming pervasive and cheap.