It's still only working in 2D, needs a roomful of computers and doesn't really "know" what it's looking at

The biggest weakness for a computer is the amount of data it can work on in an instant, which at the moment is usually 64 bits

It does a whole bunch of them dead quick, but your eyes look at, say 5MB when they look at something, plus X-megabytes for depth perception

It's a bit like looking out through a tiny toilet roll tube and only "seeing" one pixel at a time

A computer has a 64bit address, your brain has a 64000 bit address kinda thing

Even optical character recognition is still struggling, it's good, but only for the obvious stuff

That Arnie terminator robot is still a long long way away, a long way

Edited By Ady1 on 08/08/2017 15:50:05

When searching for images you have to understand that it is using the text surrounding the image. Therefore it helps to think about what the person posting an image was thinking when they titled the picture. So as with ebay additional words like hand, power, metal, Boxford, Alba can set it on the right track.
I think it has for some time understood similar words, but the AI bit will progressively allow it to appreciate context, so perhaps marking lots of sites and blogs as being engineering related. The other part about learning is that it should already be noticing that the results you click on are the one related to engineering sites. So next time round it should put more of those sites at the top. Added to that it will start to take into account that your youtube viewing and ebay transactions label you as an engineer.

I have noticed how my ebay/google searches get reflected in the advertising at the start of youtube videos. Can anyone suggest searches that will get me more interesting adverts. I'm a bit fed up with the fallout from just a few minutes looking for a plumbing part.

It won't get you more interesting adverts but you can get rid of a lot of the annoying ones by eliminating tracking cookies on those sites:

In Firefox go to Preferences > Privacy > History > Exceptions and add something like:

ebay.com Block

ebay.co.uk Block

google.co.uk Block

Russell

Edited By Russell Eberhardt on 08/08/2017 16:59:56

A VERY interesting recent development (sorry to keep yapping on about this stuff) has come about via the bitcoin and blockchain industry which does those internet ecoins

They can now produce relatively cheap and efficient hardware chips which can only do one complicated thing, but they can do it extremely efficiently

A normal computer chip can "do everything" via software, but it is actually pretty slow compared to hardware

So in the past, dedicated hardware like a 1980 video recorder could record movies but it took microchips and software 25 years to develop and catch up with that ability

Nowadays a single intel 3 Ghz core will use about 20watts to hash 3.4Kilohashes for the ferociously difficult Litecoin scrypt algorythm but hardware chips using 10watts can do 345 Khs (So 2 KW of work can be done with 10 watts)

I reckon it's only a matter of time now before someone comes up with "smart" hardware chips which can interface with a slower software based logic controller chip, sight, hearing etc will be via a dedicated hardware chip, with decisions being made via the logic chip

Then things will really take off

Edited By Ady1 on 09/08/2017 09:32:27

Moore's "Law" may have held because it is not in the Interests of the computer industry to run too fast.

When Mobile phones first came out, Motorola Gurus predicted phenomenal demand and built 3 factories in Scotland to meet the projected demand. The factories were expected to work 3 shifts. Only one factory was ever used and peaked at 2 shifts.

People accept change slowly and eventually grow tired of continual upgrading.

I find this video of an ABB Flexpicker robot and vision system very impressive, to process the image and send the offsets to the robot at that speed impresses me.

Mike

Haha I wouldn't rely on anything Motorola ever said. That's the same company that also invested massively in 3G and nearly killed itself in the process and that was just the PCS telecoms group. I was working there at the time and the leadership team were a bunch of deluded corporate retards. It's no great coincidence that since then the company has slowly but inexorably disappeared up its own back passage, to the point that there is now pretty much nothing left.

During my time there, the CEO was Chris Galvin, grandson of one of the founders. He enjoyed a suspiciously meteoric rise to the top, fuelled by a rich diet of management bulls*it and sycophantic, self-interested thrusters. I almost used to enjoy the CEO's monthly corporate email, it being so far beyond parody. There is an expression here in Lancashire "clogs to clogs is three generations" ie the first generation founds a business, the second manage to keep it going and the third run it into the ground. Chris Galvin was your man, although it's clear from his Wikipedia page that he feels he actually did them proud….

Murray

Specialised hardware ICs have been around for decades. Any PC will have a dedicated graphics processor which processes vast amounts of data in parallel. Likewise many consumer product processors will have dedicated IP hardware blocks for things like graphics and audio. They can be run with DMA so little or no need for processor intervention.

The key problem with any type of hardware processor is getting the data in and out, not the processing per se. When parallel processing can use true parallel data paths as well as multiple processors then that will really put a tiger in the tank.

Andrew

(is this Model Engineering? why not??)

I write code that runs mainly on graphics chips; the latest bit for a customer runs something around 250 million "threads" to complete one computation; takes a few seconds on my $100.00 (Canadian) GPU card. It creates a 1024×1024 "grid" as output in the end. The client loves it. (whew)

I'm currently porting it to my Android phone for a test – will be interesting to see how well it works.

Relevant? I think so – computers have invaded the workshop to an extent, and computer vision, or olde plain CNC, will be the way of the future, whether we like it or not.

In the home, we've already seen the rise and fall of Mach-3/4, maybe LinuxCNC will become embedded (as Tormach took it and package it as PathPilot) and there are a lot of other pre-packaged controllers out there. (just ask John Stevenson!)

Back to work… (sigh)