Follow Slashdot blog updates by subscribing to our blog RSS feed

 



Forgot your password?
typodupeerror

Comment It is NOT autoconplete the way you think it is (Score 1) 163

You're confusing the task with the mechanism. Classic autoconplete uses statistical methods, often using some variant of a Bayesian algorithm. The task is to predict the next word, the method is statistics.

But if I asked *you* to predict the next word in a sentence, you would not be using a simple statistical method. Neither is the AI. It doesn't have the breadth of multi domain training data that your neutral network has, so it doesn't really think like a human does, but the way it functions is much closer to your brain than it is to a classical autoconplete.

It's hard to stress enough how profound that difference is.

Comment Re: doesn't have to be bad (Score 1) 218

If I may offer a modification to your statement, it would be that car infotainment systems age poorly. Not the cars themselves. I currently own three examples that have aged quite well. All are near or over 20 years old. New enough to have OBD-2 and great reliability and repairability, old enough to be focused on real utility and avoid the invasion of useless (or worse) tech and complication. No $700 LED tail lights or $1500 mirrors with cameras, or everything-is-a-module-connected-to-the-canbus. Now get off my lawn! :)

Comment Work with what you have (Score 2) 22

Why not deploy more, cheaper, less efficient AI processors, but run it only during the daytime when the solar farms are pumping out excess energy? It won't be the highest performance and have more heat output but you don't have to worry about energy availability or build grid storage infrastructure to support it. You could then offer it at a discount for customers that are willing to wait a little longer for training tasks to complete.

This is obviously not good for on-demand inference tasks (e.g. talking to AI customer support agent), but inference is orders of magnitude less demanding on the hardware.

Comment Re:Some Evidence. (Score 3, Informative) 107

Dude there are *four* surviving space shuttles. One in DC, one in NY (those are close together, fair enough -- the one in NY was only for atmo testing and while mostly capable of flying in space, but never received the refit to be able to do it) but the other two flown shuttles are at KSC in Florida and in Los Angeles. If your argument was that people have to travel too far, then we'd move the NY one to Nebraska or something to minimize distance traveled from any point in the country. That would also be a lot cheaper to stay at a hotel there than in Houston.

But, conveniently, Florida, New York, DC and California are some of the most visited places in the US. 64% of Americans have visited Florida (far more than any other state), 56% have visited New York, 54% DC, and 50% California. Texas just barely beats California at 51%, so you could probably improve accessibility a tiny tiny tiny bit by moving the LA one to Houston, but that would leave the entire western US with worse access (distance from LA to Houston is 1500 miles, vs distance from Houston to KSC is 1000 miles, and that's not even taking into account places like the Pacific Northwest.)

If you want to see a shuttle for less money, you have a couple of options. Go to Florida and drive to KSC, or stay at a cheap place somewhere along the Northeast Corridor or Metro North train lines and take a day trip into NYC -- you can stay late as the last Northeast Corridor trains run late into the evening and the Metro North trains leave as late as 1AM. (And you can take an uber to Penn/Grand Central to get the short distance to the train stations if you want to avoid the subway at night -- it's not as dangerous as the news makes it seem but you do see some tweakers on the subway, but the regional trains out of the city are clean and comfy)

Comment I'd rather talk to an AI chatbot than a human one (Score 1) 81

On average, I've actually been very happy with the use of AI chatbots for phone support.

The reason for this is that, for lower tier support (as well as a fair chunk of things I need done that can be handled by lower support), the support agents are largely working off of scripts that they are not allowed to deviate from, nor do they have the expertise to understand what they are doing.

While the AI is not necessarily as intelligent or capable as a human *CAN* be, in practice it is often more capable than the first-tier support agent that it has replaced, due to the breadth of its training data. If I need something that requires cognitive tasks that exceed the AI's context window, I can request escalation.

This is in stark contrast to the bad old days where I waste time talking to the bottom tier support where I usually need to spend a long time explaining what I want done, wind up requesting to be escalated anyway, or try to battle with a dumb non-AI menu-based agent bot to even get to speak to a human (and usually that human is still bottom-tier support who I have the same problems with).

Comment Re:training models versus running them (Score 3, Insightful) 28

"I don't see how this is economically feasible"

We're in pre-enshitification of AI. Once one or two dominate the technology, they will start "monetizing" it. Also, the more AI shit you can spout on earnings calls, the more horny wall street gets for your sweet sweet stock.

Comment Re: scale (Score 4, Informative) 158

I see these arguments a lot and they're kind of contradictory:

1. It won't work in the US because there's too many people in the US! Norway is a sparsely-populated country!

2. It won't work in the US because there's too few people in the US! The US is a sparsely populated country!

It's a dumb comparison because you can just take any given US state in isolation the same way you can take European countries in isolation. Norway is about the same size with the same population density as Utah, and like Utah has most of its population concentrated around one major city, and has about the same median income as Utah. (Norway is much richer in GDP per capita but the average joe actually buying a car makes about the same).

If Norway can do it then surely Utah can do it.

Slashdot Top Deals

"It's a dog-eat-dog world out there, and I'm wearing Milkbone underware." -- Norm, from _Cheers_

Working...