SoylentNews is people
SoylentNews
From the RooshV Forum:
I constantly get the vibe from people that they think our technology is skyrocketing, that we're living in a new tech age, "where was all this ten years ago?!" etc.
But I disagree with this assessment of our technology. It has made steady improvements in one specific space: software and electronic hardware. That is all. On top of that, the improvements on the hardware have not even been ground breaking. GPS is a ground-breaking invention. Smaller screens are not: they are just an incremental improvement.
Smartphones are merely the result of incremental improvements in the size and quality of electronic components. The only breakthroughs involved are ages old. The invention of the transistor, the laser, etc. The existence of google, facebook, uber, and so on, are merely inevitable "new applications" stemming from these improvements. They are not breakthroughs, they are merely improvements and combinations upon the telephone, the directory, and the taxi.
In my opinion, technology as a whole is borderline stagnant.
A list of why technology is still shit:
The posting goes on to list examples of incremental, rather than breakthrough, changes in the areas of:
- Electronics & Machines
- Energy
- Medicine
- Clothes
- Food
- Finance
Have we really stagnated? Have we already found all of the "low-hanging fruit", so new breakthroughs are harder to find? Maybe there is greater emphasis on changes that are immediately able to be commercialized and less emphasis on basic research?
(Score: 0) by Anonymous Coward on Monday May 29 2017, @03:08PM
I really hate to respond to a post with just an article, but really: A Brief History of Neural Nets and Deep Learning [andreykurenkov.com]. You just repeated an oft repeated line, but it's simply not true. However, proving that is no trivial task and is not a wheel I care to attempt to reinvent.
And you're never going to get even an approximation of strong play (let alone a solution) to Go through brute force. The search space in that game is absurd, and tactical nuance (which can decide games) is not present to the degree in a game like chess that helps not only with performance against humans but also in rapid culling. Another issue is that hardware sees extremely rapid diminishing returns beyond a baseline which depends on the task at hand. For instance here [wikipedia.org] is a wiki page listing some configuration:performance results for an earlier version of AlphaGo. It's interesting that the first doubling of GPUs (from 1 to 2) lead to an increase in performance of about 26%. The next doubling saw a gain of about 5%. The next doubling saw a gain of 1%. Going from the final setup (with 8 GPUs and 40 CPUs) to a monster of distributed computing with 1,920 CPUs and 280 GPUs saw a performance gain of less than 10%. Of course even that baseline level was generally out of reach not all that long ago, yet that's for playing Go. Chess is simpler and checkers even moreso. Given today's knowledge, these games could potentially have seen computers become dominant much faster than going for the more traditional min-max eval+culling route that was used for decades.
It would be interesting to see the effort required to develop a deep learning system capable of defeating Stockfish (currently the strongest chess program) and what the final product would require in terms of relative performance to remain dominant. E.g. would a deep learning system's final product operating on 1.5 units of compute be able to defeat a stockfish running on 3 units of compute? Alas, I think Google is looking to even grander displays of AI dominance - StarCraft 2 is next on the chopping block.