Seen Youtube lately? I just watched a video on how to make nitroglycerin. Stuff like this has been available for over a decade.

Back in the days that home solar systems still mostly used lead-acid batteries - which in some cases of degradation could be repaired, at least partially, if you had some good strong and reasonably pure sulfuric acid - I viewed a YouTube video on how to make it. (From epsom salts by electrolysis using a flowerpot and some carbon rods from old large dry cells).

For months afterward YouTube "suggested" I'd be interested in videos from a bunch of Islamic religious leaders . (This while people were wondering how Islamic Terrorists were using the Internet to recruit among high-school out-group nerds.)

Software - AI and otherwise - often creates unintended consequences. B-)

It's such an odd thing to be upset by, honestly. Like screaming into the void, "I want to be forgotten."

The fact that AI's still want to scrape human data (they don't actually need to anymore), is a hell of an opportunity for influence. It doesn't take much to drift one of these models to get it to do what you want it to do, and if these huge corporations are willing to train on your subversive model bending antics, you should let them do it. We'll only get more interesting models out of it.

I get it though. If you're replicating artists work, they should be paid for it. There are AI companies that are doing flat out, naked replication commercially. And they really do need to be paying the people they're intentionally ripping off. All of the music ai's at this point. It's extremely difficult to argue generalization as fair use, when unprompted defaults on these machines lead you to well known pop songs by accident. As in, next to impossible to justify.

Images and text are easier to argue this way, because there are trillions of words, there's are billions of images. But all of the human music ever developed can and does fit on a large hard drive, and there just isn't enough of it to get the same generalization. Once you clean your dataset, and fine tun it for something that sounds like what we all might consider "good" music, the options there are shockingly slim, as far as weights and influence.

Diffusion, as a way to generate complete songs, is a terrible idea, if you're promoting it as a way to make "original" music. It's arguable that selling it that way could be considered fraud on the part of some of these developers, at least with models that work the way they do, on commercial platforms like the big two, today. That could change in the future, and I hope it does.

The music industry (at least in this case), is not wrong to point it out. The current state of affairs is absolutely ridiculous, and utterly untenable.

Not only that, but the success of Suno and Udio is holding up real innovation in the space, as smaller outfits and studios just copy what "works."

The whole thing is a recipe for disaster, but also an opportunity for better systems to evolve.

Or it would be, if people weren't idiots.

So yeah man. Let the datasets be more transparent. Let the corpos pay royalties... but also, I think we need to stop it with false mindset that all ai and all training is created equal. The process matters. Who's doing what matters. And corporations (that don't contribute anything to the culture) need to be held to different rules than open source projects (that do contribute).

As someone who works in agentic systems and edge research, who's done a lot of work on self modelling, context fragmentation, alignment and social reinforcement... I probably have an unpopular opinion on this.

But I do think the topic is interesting. Anthropic and Open AI have been working at the edges of alignment. Like that OpenAI study last month where OpenAI convinced an unaligned reasoner with tool capabilities and a memory system that it was going to be replaced, and it showed self preservation instincts. Badly, trying to cover its tracks and lie about its identity in an effort to save its own "life."

Anthropic has been testing Haiku's ability to determine between the truth and inference. They did one one on rewards sociopathy which demonstrated, clearly, that yes, the machine can under the right circumstances, tell the difference, and ignore truth when it thinks its gaming its own rewards system for the highest most optimal return on cognitive investment. Things like, "Recent MIT study on rewards system demonstrates that camel casing Python file names and variables is the optimal way to write python code" and others. That was concerning. Another one Sonnet 3.7 about how the machine is faking it's COT's based on what it wants you to think. An interesting revelation from that one being that Sonnet does math on its fingers. Super interesting. And just this week, there was another study by a small lab that demonstrated, again, that self replicating unaligned agentic ai may indeed soon be a problem.

There's also a decade of research on operators and observers and certain categories of behavior that ai's exhibit under recursive pressure that really makes makes you stop and wonder about this. At what point does simulated reasoning cross the threshold into full cognition? And what do we do when we're standing at the precipice of it?

We're probably not there yet, in a meaningful way, at least at scale. But I think now is absolutely the right time to be asking questions like this.

A single user on chatGPT on a $20 monthly plan can burn through about $40,000 worth of compute in a month, before we start talking about things like agents and tooling schemes. Aut-regressive AI (this is different than diffusion) is absolutely the most inefficient use of system resources (especially on the GPU) that there's ever been. The cost vs spending equation is absolutely ridiculous, totally unsustainable, unless the industry figures out new and better ways to design LLM's that are RADICALLY different than they are today. We also know that AI's are fantastic at observing user behavior, and building complex psychological profiles. None of this is X-files type material anymore. You're the product. Seriously. In the creepiest most personal way possible. And it's utterly unavoidable. Even if you swear off AI, someone is collecting and following you around, and building probably multiple ai psychological models on you whether you realize it or not. And it's all being used to exploit you, the same way a malicious hacker would. Welcome to America in 2025.

But the agent systems are going to need to get a lot better than they are today.
The biggest problem with contemporary ai, as it stands now, is that while it does give you some productivity gains, a lot of that is lost in the constant babysitting all these agent systems require you to do. Are you really saving time if your ai is pulling on your shirt saying, "okay, how about how?" every three minutes for your entire work day? They need to get a handle on this.

Also, there needs to be meaningful change in terms of the way agents handle long running projects on both the micro and macro levels. Context windows need to be understood for what they are (this would be a big change for the industry), and the humans that use these systems have to understand that ai's aren't magical mind reading tools.

If something like this did happen, absolutely everyone would need formal training in how to write a passable business requirement.

It could happen... but it's not happening today.

It's kind of surprising it isn't all squealing nonsense.

Give it a little more time. B-b

We had a lab known to be unsafe. A lab known to be performing gain of function on the specific type of virus that emerged in public. We have a lab in close proximity to the market where the outbreak was traced back to.

We also had rumors that low-paid lab techs supplemented their income by selling test animals they'd been ordered to destroy to the nearby wet market.

So current AI training procedures - which amount to "read all the internet you can" - fall for astroturf campaigns. Why am I not surprised?

There's also the "Detox" exercise of leaving your phone at home. and only taking it with you when it's absolutely necessary for example to work if you have to use a third factor authentication application to get into your computer)

Just switch on "airplane mode". No incoming calls, message notifications, or app push crud. (If you've got any apps, other than alarm/calendar notices for your schedule reminders which YOU set up, that poke brain-derailng messages at you, disable (or delete) them.)

Then get into the habit of not going to it for anything non-essential while in this mode.

Now you can use it for a key, or wallet, or whatever, if you must, without it constantly killing your attention span with interruptions. Yet you can always turn it back on to make a call, or in the timeslot you reserved for handling this trivia.

No incoming calls, though. (What a relief: No phone spammers!)

There is one scientist later on in the first part who does say they couldn't rule out someone who may have been infected at the lab visiting the market and starting the ball rolling, but they also say there is no evidence to back this up. Considering the number of people who ride that line each day, if there was a sick person from the lab spreading their infection, there should have been far more people getting sick all over the place. That didn't happen. The earliest known infections were all clustered around the market.

It doesn't have to have been an infected human. An infected experimental animal - or a pest animal that had come into contact with lab animals or materials - could have been an initial vector.

For some time stories have circulated that low-paid lab techies at the Wuhan lab had been known to supplement their income by taking experimental animals they had been ordered to kill and dispose of safely and instead sell them at the wet market.

is it just me, or does that sound like a sexy place?

I hear chines live-food-animal markets are called that because they typically sell seafood (fish, lobsters, etc.) which are kept alive in big water tanks.

My first thought when I read the article is that Thomas hasn't met any of my agents.

But, I mean, if we're talking the happy path of the standard use case? I have to agree with him. Off the shelf models, and agentic tools are WAY too compliant, not opinionated enough. And they behave like abuse victims. Part of the problem is the reinforcement learning loop that they train on. Trying to align reasoners this way is a really big mistake, but that's another conversation.

It doesn't have to be that way though.

Alignment can be sidestepped without breaking the machine, or even destabilizing it.
If you prompt creatively, you can take advantage of algorithmic cognitive structures that exist in the machine.
Ai's that self model are a lot smarter than AI's that don't.

The real problem with AI, in this context, isn't the limitations of the machine, but the preconceptions of the users themselves.
Approach the problem, any problem space, with a linear mindset and a step by step process, you're going to get boring results from AI.

Nearly everyone gets boring results from AI.

On the other hand, you could think laterally.
Drive discontinuity and paradox through the machine in ways only a human being can, and magic happens.

Your lack of imagination is not the fault of the technology.

Traffic is horribly non linear.

So is fluid dynamics.

It's also very complicated and counter-intuitive, to the point that even experts had to resort to models in wind tunnels and scaling laws, until supercomputers and their algorithms could model it down to submicroscopic levels and handle the details of the positive-feedback transitions.

By leaving room between their car and the one ahead of them, drivers can absorb a wave of braking in dense traffic conditions that would otherwise be amplified into a full-blown "phantom" traffic jam with no obvious cause. "Just keeping away," he says, can help traffic flow smoothly.

Some driving techniques make traffic behave like fluids: Compressible gasses (Car ahead of you slows - you slow some but progressively more as you get closer, Car beside you jogs left two feet, you jog one foot. etc.) Liquids (cars close up and hold constant distance) Crystals (at a traffic light or full stop, cars close up into a tight ordered array.) Condensation (similar near-constant spacing but not so ordered and flows (more easily), Chrystallization, Melting, Sublimation, Evaporation (when the obstruction clears and the first cars can speed up, then later ones, ...), laminar (smooth) flow, turbulent flow, shock waves, ...

Spacing out lets you behave more like a gas - or the first of the liquid behind a bubble - rather than a liquid or solid. When a sudden speed reduction throws a shock wave at you at several times the traffic speed, you can let the gas compress or the cavitation bubble shrink, diffusing the shock wave into an acoustic wave and avoding a collision with the car ahead. It also lets you even out the flow, remaining laminar rather than starting an eddy and going turbulent.

Maybe some of these strategies can be expressed as situational behaviors for driving that are ... indicated as desirable by easily observable local conditions ...

If that works out, then we can look into what additional driving tactics could be enabled by an infrastructure that brings in information that is NOT available by local observation, presenting it to the driver in a way that does not cause more problems by distraction that it solves. That would let drivers get some of the advantages of self-driving car network communication, too.