In principle, one could train the AI to insert ads in its answers. So no, if you only do inference locally with an open-weight model you are still not in control.

I think ads can be removed with abliteration, just like refusals in "uncensored" versions. Find the "ad vector" across activations and cancel it.

Despite the shortage, RAM is still cheaper than mathematicians.

It's also less frustrating to organize world wide ram production and logistics than to deal with a single mathematician.

Constantly sitting around trying to solve problems that nobody has made headway on for hundreds of years. Or inventing theorems around 15th century mysticism that won't be applicable for hundreds of years.

Now if you'll excuse me I need to multiply some numbers by 3 and divide them by 2 ... I'm so close guys.

The comment feels a bit like Verdex may have dated a mathematician at some point and it went sour.

I don't know, I think if you weighed up the costs of AI related datacentre spend vs. the average mathematics academic's salary you could come to a different conclusion.

Raising, nurturing, training, and mentoring an expert mathematician is not cheap; it never was, perhaps the first time in history when we can witness that rule to change - spinning up a bunch of math-savvy agents, each smarter than Ramanujan maybe will get too cheap.

You dont have to raise them, someone already did it, you have to hire them

You're oversimplifying the message I'm trying to convey. "you just hire them, someone already raised them" - treats mathematicians as a commodity stock rather than a flow. The conversation frames it as "mathematicians vs. RAM" - a cost comparison. But that's like comparing the cost of a GPS unit vs. a ship captain. The captain isn't expensive because they can calculate routes; they're expensive because they know when the route is wrong. AI makes the math cheaper but makes the mathematician more valuable, at least until true AGI genuinely surpasses human mathematical creativity - at which point we have much bigger economic questions than mathematician salaries.

The topic on itself is quite interesting, and far complex than supply/demand norms. Even before AI, there was and both wasn't shortage of mathematicians - academic pure mathematics - there's a glut. High school teachers - people exist; but they won't work for teacher salaries. Applied math - acute shortage - quant finance, ML research, cryptography, pharmaceutical modeling - we don't have enough. NSA - always struggled to hire - private sector salaries pull people away. Interdisciplinary - mathematical biology, climate modeling, materials science - domains where math is the bottleneck but the job title isn't really "mathematician" - acute shortage.

Doubt it. You have to pay these mathematicians once and then you can deploy to millions of sites.

But not everyone has to pay mathematicians, like RAM :-)

At the same time, processing is much cheaper than memory

Without memory you have no data to compute on. Memory and compute scaling only makes sense in tandem.

This pipe dream will soon be replaced by "let's have the first degree of judgment be ChatGPT; human judges should only deal with appeals".

Sounds like every self-driving startup

Can you use git's Copilot from the command line? If you can't, then you have nothing to opt out from.

It's not git's Copilot it's Microsoft, or at best github's, Copilot.

And Copilot is integrated with IDEs. Doesn't need any interaction with the github site beyond the initial sign in...

By that logic, you can't use any user input to train an LLM, because what if they decide to write their own name.

Indeed, you can’t unless you have appropriate consent. Which isn’t difficult to obtain if you have clearly defined purposes, but you have to do it.

Let me add that the typical GrapheneOS user will probably prefer to install the OS themselves rather than trust what comes preinstalled.

The typical GOS user generally doesnt want to do that. Flashing is a hurdle that increases barrier for entry. Reducing or eliminating that burden is ideal. Greenboot support would make flashing a little easier.

> typical GOS user generally doesnt want to do that

How do you know this? Is there an official (or even unofficial) source of GOS preinstalled devices that a substantial amount of "typical GOS user" has acquired?

Or maybe you are talking about "potential user of GOS"?

In any case: if you installed it yourself you mostly have to trust the source of the installer. If you purchase a pre-installed device you're basically back to the android/ios model: you have to trust the manufacturer AND the maker of the OS

I have helped a significant number of GOS users install GOS to their device. If you perform post install steps correctly then you do not need to trust where you got it from, as the post install steps are there to verify your install is genuine. If GOS gets greenboot support for motorola devices, then not getting a yellowboot screen will show it is genuine and you wont need to trust anything.

It's still less space for other things in the L1 cache, isn't it?

If your goal is reducing the number of multiplications, I imagine it would make sense to factor that polynomial into degree-1 and degree-2 factors.

It's abs(x) only over the reals, for complex numbers it's more complicated.

So basically a homing missile?