You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
...which led to the consolidation of the machine learning community behind these highly-effective "winners". Nowadays, the entire deep learning software and hardware ecosystem is micro-optimized for a small set of techniques.
At this point, figuring out a different path would require a multi-year re-engineering of the entire ecosystem.
Great point. Very similar to what happened in molecular biology with CRISPR-Cas systems that have been engineered and optimized around programmable nucleases. Ecosystem is built; no turning back.
We're slowly climbing towards a local maximum. We could be missing out on other maximums. That's why we need increased competition in the hardware and software space.
Seems that the explore/exploit approach wasn’t fully employed. Perhaps due to hype effects ushering in a supply to sudden demand.
On the other hand, there is problem of Universal Priors that this far has no solution for explore/exploit optimal reinforcement agent. So, may be it is not possible in principle.
We need to though, since current large-scale training (ie training large models) is power-inefficient. GPT-3 points to very larger models, and 1 TeraOp/watt, a trillion-parameter NLP model will take 1000s of MWs to complete training.
So, one might say that AI research has reached a local maximum...and not necessarily the best maximum.
Where we are is not as important as where we are goinghttps://mobile.twitter.com/jsalsman/status/1336799268024139776 …
I think @graphcoreai is trying, but to get traction they first have to make existing techniques work on their chips. The real boon will be in techniques specific to the chip though.
Neuromorphic computing is another promising path that is being investigated.
@sarahookr's hardware lottery argument unpacks this exact train of thought I believe http://arxiv.org/abs/2009.06489
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
