Artificial Intelligence & Robotics

[nudge]

Have you heard anything about this @Mel81x?

https://www.theguardian.com/technology/2019/feb/14/elon-musk-backed-ai-writes-convincing-news-fiction

[Mel81x+]

35 minutes ago, nudge said:

Have you heard anything about this @Mel81x?

https://www.theguardian.com/technology/2019/feb/14/elon-musk-backed-ai-writes-convincing-news-fiction

i knew they were working on GPT2 just didn't know they had gotten this far. I think its safe to say that it won't see the light of day soon considering what it can do in its current state. I quite liked that it was able to take text from the internet and start trolling people as well. It's bad enough we have users who shit on quality content but now we have a machine that is relentless too and who knows what people could come up with (as stated in the article). Imagine if it started pushing out ads to users with the smarts to know enough about demographics so that it never had to ever worry about a team telling it what it needed to push. Then it takes it one step further and starts creating text which incites enough hate about something that someone uses it to slander a competitor to a state where it could then fabricate more text to corroborate why its slandering the competitor. 

All this said I have to say I am very impressed. Playing DOTA was always going to be the first stepping stone to learning how to beat a human while pretending to be human. This is a whole level of awesomeness that I don't think the world is ready for.

[Mel81x+]

https://arstechnica.com/information-technology/2019/02/researchers-scared-by-their-own-work-hold-back-deepfakes-for-text-ai/

@nudge if you want to see some of GPT-2s work look at the middle of that article. It's quite entertaining.

[nudge]

5 hours ago, Mel81x said:

https://arstechnica.com/information-technology/2019/02/researchers-scared-by-their-own-work-hold-back-deepfakes-for-text-ai/

@nudge if you want to see some of GPT-2s work look at the middle of that article. It's quite entertaining.

This is really impressive. I was mostly surprised by the fact that it is able to actually pull off the specific tone that is required based on the prompt it's provided with. That nuclear material theft report was the most believable one both in its style and contents; certainly wouldn't be able to recognise it as a fake. The unicorn one contained factual and logical errors that were easy to spot but that's probably what the article meant when it said that the AI does well with well-known popular topics while it struggles with more exotic unusual material. The JFK one was hilarious (even with footnotes and what not!!!) and made me think that at some point we might actually have algorithms like this to create decent fictional literature works. 

[Mel81x+]

7 hours ago, nudge said:

This is really impressive. I was mostly surprised by the fact that it is able to actually pull off the specific tone that is required based on the prompt it's provided with. That nuclear material theft report was the most believable one both in its style and contents; certainly wouldn't be able to recognise it as a fake. The unicorn one contained factual and logical errors that were easy to spot but that's probably what the article meant when it said that the AI does well with well-known popular topics while it struggles with more exotic unusual material. The JFK one was hilarious (even with footnotes and what not!!!) and made me think that at some point we might actually have algorithms like this to create decent fictional literature works. 

I can see some school kids going out of business if this AI is put as open source and some kid finds a way to write all the essays for the class. Some poor teacher will never be able to tell if its a human that did it or a machine. The other thing in that article is the 40GB payload to learn which I found incredibly small considering all the sources it was gathering its writing skills from. I was half expecting Reddit style postings in there but it kept things very articulate which means it learned which posts got higher ratings then started filtering them out based on language usage, incredibly smart. If they took this same thing and taught it to code and self-heal broken code systems that would be even more cooler. 

[Mel81x+]

https://arstechnica.com/cars/2019/03/in-1959-volvo-gave-us-the-seat-belt-heres-what-its-safety-team-is-building-now/?comments=1

Only reason I even put this in here is the driver assistance feature.

[nudge]

Elon Musk's Neuralink announced plans to start testing its BMI implants on human volunteers next year. Basically a “high bandwidth” brain-computer connection, made using thin flexible threads and implanted via sewing machine-like robot surgery and they apparently successfully tested it on rats and monkeys already. Short-term initial goal is to help people with brain and spinal cord injuries or congenital defects, the ultimate goal is symbiosis of the human brain with AI. Now it's Musk after all, so we should take this announcement with a pinch of salt, but this is definitely very interesting and ambitious.

White paper of their technology: https://assets.documentcloud.org/documents/6204648/Neuralink-White-Paper.pdf

Video of the recorded livestream: https://www.youtube.com/watch?v=r-vbh3t7WVI

Informative CNET article with a lot of details: https://www.cnet.com/news/elon-musk-neuralink-works-monkeys-human-test-brain-computer-interface-in-2020/

 

Thoughts?

@Mel81x

[Mel81x+]

On 20/07/2019 at 03:47, nudge said:

Elon Musk's Neuralink announced plans to start testing its BMI implants on human volunteers next year. Basically a “high bandwidth” brain-computer connection, made using thin flexible threads and implanted via sewing machine-like robot surgery and they apparently successfully tested it on rats and monkeys already. Short-term initial goal is to help people with brain and spinal cord injuries or congenital defects, the ultimate goal is symbiosis of the human brain with AI. Now it's Musk after all, so we should take this announcement with a pinch of salt, but this is definitely very interesting and ambitious.

White paper of their technology: https://assets.documentcloud.org/documents/6204648/Neuralink-White-Paper.pdf

Video of the recorded livestream: https://www.youtube.com/watch?v=r-vbh3t7WVI

Informative CNET article with a lot of details: https://www.cnet.com/news/elon-musk-neuralink-works-monkeys-human-test-brain-computer-interface-in-2020/

 

Thoughts?

@Mel81x

So from what I've read and seen I like the idea of brain wave scanning and pattern reading as well. But for me the bigger implementation here is behavioral change now because the application here is far more than just reading and integration with an AI. I also think this could be used for degenerative muscle disorders as well because if they can get the strands to operate with flexibility on electricity they could strengthen core areas of the muscle structure to help people with accidents and injuries to recover faster with more stimulation as well. 

[Azeem]

https://m.huffpost.com/us/entry/us_5d0d0faee4b0a3941861fced?guccounter=1&guce_referrer=aHR0cHM6Ly90LmNvLzN5cEdPeHdxSm4_YW1wPTE&guce_referrer_sig=AQAAAErbm73YByPgwHDh0xrp9BRe50NT7vGJw1VAVI6qG5TlgqD3udv-Q7ReoxqjjA8ZnoY1hzUNb0opuiXNLZD6LPftPPM09EYXZJdlYgfmFLC0Nz7EV7O63dk5xMxt-ATG-S3IsnwpAhAQl2kmplPyAw7kWXItV1aR0s3rdUAEtve8

Deep fakes and there threat

Edited August 14, 2019 by Stick With Azeem

[Mel81x+]

6 minutes ago, Stick With Azeem said:

https://m.huffpost.com/us/entry/us_5d0d0faee4b0a3941861fced?guccounter=1&guce_referrer=aHR0cHM6Ly90LmNvLzN5cEdPeHdxSm4_YW1wPTE&guce_referrer_sig=AQAAAErbm73YByPgwHDh0xrp9BRe50NT7vGJw1VAVI6qG5TlgqD3udv-Q7ReoxqjjA8ZnoY1hzUNb0opuiXNLZD6LPftPPM09EYXZJdlYgfmFLC0Nz7EV7O63dk5xMxt-ATG-S3IsnwpAhAQl2kmplPyAw7kWXItV1aR0s3rdUAEtve8

Deep fakes and their threat

The problem with the technology is that its actually really good when it comes to using AI and face recognition to generate what is the best fit for a body. I remember reading an article on ARS not so long ago where a site was shutdown not because it was illegal but because it was unable to handle traffic. This trend you've posted is far more disturbing because putting porn aside we now have a platform that allows people to auto-generate anything they literally want and the day isn't very far where the system could potentially use text to generate speeches or do other things as well. Most of the code that does this isn't even hidden and could probably be found with some serious searching on groups with a ".TO" extension as well like that hacker group which went out and stole and distributed another hacking group's data. 

Used to be governments were worried that people were stealing classified information, now they'll have to worry that people/groups can actually mimic and generate false content and generate misguiding propaganda too.

[Mel81x+]

https://techcrunch.com/2019/08/19/the-five-technical-challenges-cerebras-overcame-in-building-the-first-trillion-transistor-chip/

Great name for a company and a big leap in finding processing power that is faster than the current connector standards. Curious what the tests in the lab were like but I am sure those will come out soon enough too.

[Mel81x+]

https://arstechnica.com/science/2019/09/paper-leaks-showing-a-quantum-computer-doing-something-a-supercomputer-cant/

We dont really have a section for Quantum computing not like we really need it so I figure I'd put in this section. What's interesting here is how they went about claiming Quantum Supremacy and the fact that they used observation to achieve it. Now, in the comments section someone has posted that someone at Caltech said its a great feat and he was the one that coined the term but I suppose until its solving practical problems and doing all this theoretical testing it really won't amount to much in the end. 

[nudge]

15 minutes ago, Mel81x said:

https://arstechnica.com/science/2019/09/paper-leaks-showing-a-quantum-computer-doing-something-a-supercomputer-cant/

We dont really have a section for Quantum computing not like we really need it so I figure I'd put in this section. What's interesting here is how they went about claiming Quantum Supremacy and the fact that they used observation to achieve it. Now, in the comments section someone has posted that someone at Caltech said its a great feat and he was the one that coined the term but I suppose until its solving practical problems and doing all this theoretical testing it really won't amount to much in the end. 

The whole quantum mechanics and computing stuff flies well over my head really... What would be practical uses and applications of an actual working full-fledged quantum computer (bar cyber security)? If I understand it correctly, it's advantage mainly lies at solving very specific extremely high level mathematical problems and simulating extremely complex systems that would require infinite time  or would be fundamentally impossible on "regular"computers? 

[Mel81x+]

On 25/09/2019 at 20:36, nudge said:

The whole quantum mechanics and computing stuff flies well over my head really... What would be practical uses and applications of an actual working full-fledged quantum computer (bar cyber security)? If I understand it correctly, it's advantage mainly lies at solving very specific extremely high level mathematical problems and simulating extremely complex systems that would require infinite time  or would be fundamentally impossible on "regular"computers? 

I think that's the biggest issue surrounding quantum computing right now. Well two big problems, the first being that it has no real-world applications that trump the user of a super-computer which suffices for what it does right now and isn't really something you'd even want quantum computing for. The second being the fact that qubits are highly unstable and to maintain something or create something for observation seems like its only practical use right now. The bigger problem with quantum computing as a science as well is the fact that its all just theoretical and while we'd like to think that most things around us warrant complexity of that nature they really don't just because of how we perceive them to bring some kind of stoical measurement into place. 

Long story short, your i3, i5, i7, AMD variant would beat the shit out of a quantum processor if you tried to use it for daily tasks which is why its where it is, in theoretical science space. 

[Mel81x+]

https://www.gatesnotes.com/Health/Professor-Hawking-Fellowship-lecture

Since we don't have a health section figured I'd put this in here since its pretty much technology-based with data.

[Mel81x+]

https://arstechnica.com/gadgets/2019/10/fieldbit-is-bringing-augmented-reality-to-industrial-plants/

I like this. One of the projects I work on involves providing field engineers with AR schematics of motherboards and other devices in server centers so when I saw this it makes complete sense that they'd take some of that tech and go onto the industrial floor. The ability to tell if something is reaching critical mass or not functioning properly, with good tech, is something that isn't really light years away. We have the sensors, we have the AR and its high time someone put it to good use. But, as the article suggests, this does introduce the probability of a less skilled engineer being present to not be able to diagnose a problem properly. However, if you don't put it out into the world you'll never make it better. 

[nudge]

Scientists used stem cells from frogs (and evolutionary algorithms in supercomputers) to build first living programmable machines. This is fascinating stuff.

https://www.popularmechanics.com/technology/robots/a30514544/xenobot-programmable-organism/

Also the link to the original paper for more details: https://www.pnas.org/content/early/2020/01/07/1910837117

 

@Mel81x you might be interested in this if you haven't read it yet!

[Mel81x+]

5 hours ago, nudge said:

Scientists used stem cells from frogs (and evolutionary algorithms in supercomputers) to build first living programmable machines. This is fascinating stuff.

https://www.popularmechanics.com/technology/robots/a30514544/xenobot-programmable-organism/

Also the link to the original paper for more details: https://www.pnas.org/content/early/2020/01/07/1910837117

 

@Mel81x you might be interested in this if you haven't read it yet!

And you send this in a 10 days span when I've just watched Carpenter's The Thing and then the prequel as well. Scary stuff, very scary stuff.

[nudge]

This is heartbreaking. VR helps a mother get some closure after losing her young daughter to an illness. 

 

[nudge]

https://www.technologyreview.com/s/615189/what-ai-still-cant-do/

A good read about one of the limitations of AI currently and what is being done to overcome it. Also there's an interesting takeaway there that the same limitation also applies to human minds and how it has been causing (hehe) problems in scientific research for ages. 

@Mel81x 

[Mel81x+]

1 hour ago, nudge said:

https://www.technologyreview.com/s/615189/what-ai-still-cant-do/

A good read about one of the limitations of AI currently and what is being done to overcome it. Also there's an interesting takeaway there that the same limitation also applies to human minds and how it has been causing (hehe) problems in scientific research for ages. 

Excellent article which can spawn myriads of questions into how data and its usage in scenarios ultimately leads to a deeper understanding (even if its skewed to fit a purpose). I remember reading a Nassim Nicholas Taleb book called "Antifragile" that deals with how disorder creates scenarios of importance in situations where you'd least expect them to. This of course was backed by math and then by numbers but you then have to ask yourself how good the AI is ever going to be if it doesn't truly have all the data it needs to make that math work. Just because an equation works doesn't necessarily mean its being applied correctly. This is where that genius bastard Yoshua in Montreal made some serious inroads. His paper on neural networks and how they can be used to further learning by simply introducing metas into formulas as a way of testing learning is astounding (i'd recommend a read if you like formal languages and automata theory but it turns into a dull affair if you don't grasp that concept as hes not the kind of person who likes to do things which can be understood by people who dont have their fundamentals right). 

All said and done this is a part of AI that goes beyond what we think Machine Learning is going to be useful for. Just because a machine can churn out data based on what its given to produce results is still looking at the problem as a box versus a universe. I also think that statement about this being the last hurdle is ridiculous and the kind of thing you'd expect from someone at MIT or Columbia just because they themselves are victims of causality and trying to put it in a box with a limitation that math can solve when the reality is that there are way too many variables in play to truly say it will work without seeing everything (and that imo isn't happening for the next 100 years or maybe more just because of how much society and our living and understanding changes every day)

[nudge]

https://www.technologyreview.com/s/615434/ai-machine-learning-social-outcome-prediction-study/

@Mel81x not surprising but an interesting read nevertheless. Bad news for Rehoboam

[Mel81x+]

1 hour ago, nudge said:

https://www.technologyreview.com/s/615434/ai-machine-learning-social-outcome-prediction-study/

@Mel81x not surprising but an interesting read nevertheless. Bad news for Rehoboam

"Three sociologists at Princeton University asked hundreds of researchers to predict six life outcomes for children, parents, and households using nearly 13,000 data points on over 4,000 families."

If they think that's enough data they are sadly mistaken. Also, Rehoboam would have looked at the problem and said "thats okay i'll just change society to suit my needs" haha.

[Mel81x+]

@nudge figured you might find this interesting from the Director AI and how it works.

Edited April 14, 2020 by Mel81x

[nudge]

21 minutes ago, Mel81x said:

@nudge figured you might find this interesting from the Director AI and how it works.

Fascinating. The idea of an omniscient director AI is very cool.  I also love how the Alien unlocks its behaviour trees based on the player's behaviourial patterns and then adjusts its own actions accordingly. Very neat.

Also I would poop my pants if I had to play it