Sign Banking Form Alabama Now

Industry sign banking alabama form now

yes give me a second to share i'm mark mitzberg coming to you from the dark of night in berkeley california i am a scientist entrepreneur called upon lately after 35 years in artificial intelligence to advise um governments and organizations but i will be coming to you from a very technical place and uh give you a whirlwind tour of what's under the hood and and my ideas about how this affects finance so i'm at the berkeley ai research lab university of california berkeley one of the great public artificial intelligence laboratories with uh 40 some professors and hundreds of phd students and and i'd like to give the quick thumbnail of why the sides of the buildings in silicon valley uh switched from dot com to dot ai um in in 2010 uh supervised learning on images that is recognizing images uh had a breakthrough and that's called deep learning and uh followed by some variants that can handle time series really sparked this current ai boom and the reason that we are all talking today um in finance this has enabled some leaps in uh as as near mentioned risk assessment actuarial credit scoring fraud detection some sales techniques and a little bit of of alpha generation among other things um i'd like to point out though that that since then there there actually have been a couple of breakthroughs that that followed so this is the the for the quants among you um a little bit of technical background on deep networks but following that um uh four years later um something was developed called word embeddings um and and i i argue that this is really another leap because these embeddings somehow surprised us in finding the math from text and taking as input very large body of text and coming out with what's called a metric a way of measuring the the relative locations of words so you can actually do math with the words and if you take queen and subtract woman and add man the nearest word in that in that space is king and so it allows one to really enable a new level of text-based automation and then finance this allowed better analysis of regulatory filings and it's of course in finance and beyond enabled chat email box and and understanding entities and texts and so forth and this was followed by uh another leap so that again i'm going to just skip right over the deep technical another leap in natural language processing called transformers which add context and allow the parallelization of this natural language processing it internalizes the probabilities of style in a language and this enabled another leap in internet search and in language translation and the dazzling demonstrations of what you might have read about gpt-3 open ai's generative pre-trained transformer that it can create a sonnet or a beatles song or something that looks like a peer-reviewed paper on genetic biology just given a couple of sentences um having again internalized the probabilities of language very very large scale and so in in finance this uh enables better analysis of the regulatory filings and news feeds um and and much more realistic chat and text and and and even generation of reports um and realistic conversations if you use a voice generation system and and it could lead to finding a kind of grammars for market behavior for human behavior and so for the quants a little background uh and and i can't help myself uh and and am uh sharing that there are aspects of this uh context aware neural network that are reflective of the the shape of the cortex in mammals that is there are these these neurons called called pyramidal neurons that have as many as 30 000 synapses that's just one cell and it's connected to 30 000 others and these are really the workhorses of cognition and and they they tend to cross the layers of the cortex which is which is where the higher um cognition occurs so so it seems to be following mother nature's recipe uh at least viewed from a distance and so these are these are some um of the breakthroughs that i think will be following um at the center for human compatible artificial intelligence where i'm the executive director um we we we're working on the problem of control near mentioned the the problem of pursuing um defined objectives and and how there are unattended consequences and um we we believe that this pursuit of fixed but incorrect objectives has already led to a lot of difficulties the the driving people to extremes and um in certain ai systems and and so we're working on a new model of ai that avoids fixed known objectives built into a system um we have a few opinions because this is berkeley and then they include the banning of lethal autonomous weapons more importantly for finance ai impersonation and certain kinds of reinforcement learning and human interaction and we are advocating for regulatory oversight of new algorithms even going so far as as to advocate for trials with human subjects um so these um breakthroughs uh uh from from 2010 have been followed in in my view with more recent breakthroughs embeddings transformers that promise to unlock further leaps and we believe that regulation and finance should track technical developments with an eye to the problem of control thank you very much thank you very much mark we really appreciate uh your contribution and particularly the unpleasant hour that you have to do that that my project dad's in um so we