The most prestigious law school admissions discussion board in the world.
|
 |
 
|
GPT-3 is going to SLAUGHTER the legal profession holy shit
Poast new message in this thread
Date: July 30th, 2020 12:26 AM Author: mustard address This is an early alpha preview of what to expect and it already does about as well at the average Hastttings grad.
https://twitter.com/f_j_j_/status/1283349995144359937?s=19 |
 |
Date: July 30th, 2020 3:03 AM Author: Canary French Sanctuary Wagecucks It does have that TTT stilted style. (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681566) |
Date: July 30th, 2020 12:29 AM Author: Rose up-to-no-good chapel 180
https://www.gwern.net/GPT-3 |
|
Date: July 30th, 2020 1:01 AM Author: cruel-hearted charismatic crotch so poasting quality is about to skyrockettt? (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681094) |
|
Date: July 30th, 2020 1:40 AM Author: chrome laughsome library death wish (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681261) |
|
Date: July 30th, 2020 3:22 AM Author: impressive blood rage shrine (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681603) |
|
Date: July 30th, 2020 7:07 AM Author: poppy hairraiser sneaky criminal masturbator (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681750) |
|
Date: July 30th, 2020 11:39 AM Author: exciting hyperactive field (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40682633) |
|
Date: July 30th, 2020 1:14 AM Author: Garnet Philosopher-king Lol is that a white guy with dreadlocks? (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681147) |
|
Date: July 30th, 2020 1:21 AM Author: Navy metal native The implications of GPT-3 are broader than that. OpenAI apparently is trying to scale it by 100x-1000x in the next year. I assume if they do this, they will probably use a better transformer model that doesn't have a fixed attentional window. It should then be capable of generating or interpreting entire books, plus loads of other things that we can't predict now. add in multi modal learning in a couple years and things will start getting weird very quickly.
http://www.autoadmit.com/thread.php?thread_id=4592961&forum_id=2#40670008
|
|
Date: July 30th, 2020 1:49 AM Author: Exhilarant main people What is the point of doing this? Can “tech” idiots not see past next week? (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681300) |
|
Date: July 30th, 2020 3:33 AM Author: Navy metal native Full scale automation of human labor. All sorts of technological advances that we can only dream about now. (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681614) |
|
Date: July 30th, 2020 4:49 AM Author: fluffy national "dream" (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681702) |
|
Date: July 30th, 2020 4:52 AM Author: Exhilarant main people But why? Likely to lead to social unrest. Most people need some degree of work in order to feel fulfilled psychologically. It’s ingrained in our DNA. (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681706) |
|
Date: July 30th, 2020 6:13 AM Author: Navy metal native social unrest will be fairly meaningless after this is all over. most people will have no capacity to resist machine control- these systems will be many times smarter than any human, and there will be many of them. In any case, people will be able to alter their own brains to make them more satisfied with the new way of life. I think wireheading (direct stimulation of the brain's reward circuitry) will be common. (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681731) |
|
Date: July 30th, 2020 7:08 AM Author: poppy hairraiser sneaky criminal masturbator why would anybody, anywhere, want this (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681751) |
|
Date: July 30th, 2020 7:24 AM Author: Navy metal native Fundamentally, people are just reinforcement learners that execute behaviors to achieve reward. If you can just bypass the effort part and achieve indefinite reward by wireheading, then that's what you will prefer. I think the idea that people would prefer not to bliss out if given the option is self-deception. (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681765) |
|
Date: July 30th, 2020 7:27 AM Author: poppy hairraiser sneaky criminal masturbator I think you're right at the margin, but I think you're wrong if you present people with a clear view of the end result. So if it's a series of minor marginal steps people will be ok with it, like how no one can read a map anymore because Siri will do it for you. This is why it's important to show people what the ultimate end result will be multiple steps/years down the road, because they will (rightly) react with revulsion as I'm doing here. butlerian jihad now, etc etc (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681772) |
|
Date: July 30th, 2020 7:44 AM Author: Navy metal native I think you're right - most people intellectually don't like the idea of wireheading. i think what happens as people develop is that they learn various sub-goals from culture and society that help guide behavior. So they learn to invest in notions of family, civil duty, etc and achieving these sub-goals provides a mental reward. If they were to wirehead, these sub-goals would be ignored and not satisfied. I think the salience of those goals would disappear if the reward circuitry was directly stimulated (which is ultimately why those goals have meaning), so i am not sure it matters. At the very least, one can imagine creating a delusion system that would convince the brain that those goals have been achieved. We aren't in direct contact with reality anyway, so it would be as real as anything else we experience. (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681793) |
|
Date: July 30th, 2020 11:50 AM Author: sable adulterous legal warrant rehab Great sub thread.
How long will the transition to full wire heading take? Seems like the unrest will come from the years during which automation gradually takes hold but is kind of shitty, so most people can't have normal jobs but a lot of people can -- kind of like now -- and there are large powerful segments of the population who try to stop the transition (opposition to UBI or whatever exists in a post-income society) |
|
Date: July 30th, 2020 12:33 PM Author: Navy metal native i don't think there will be a long, difficult transition period. this lesswrong post from a few days ago kind of spells out what I think is about to happen:
https://www.lesswrong.com/posts/N6vZEnCn6A95Xn39p/are-we-in-an-ai-overhang
we're at the point where pouring money into training brain-sized neural networks is likely to yield something immensely economically useful. the major tech companies doing AI research are likely to realize this, if they haven't already. GPT-3 isn't close to the limit of what we can train now using known techniques. huge budgets combined with improved, specialized hardware for training large neural networks means AGI or something to close to it is realizable in a few years. |
|
Date: July 30th, 2020 2:39 AM Author: Rose up-to-no-good chapel how do you not have a fixed attentional window? (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681509) |
|
Date: July 30th, 2020 3:27 AM Author: Navy metal native using something like Transformer-XL. there are plenty of options.
the scary thing is that besides allowing the model to capture longer dependencies, the parameter efficiency is much greater than with a standard transformer like they used with GPT-3. it's around 10x greater. we could be looking at something with radically enhanced capabilities before too long. i still think these models won't work well in many cases until multi-modal learning can be incorporated (so it can learn real-world embeddings and actually understand what text refers to), but that's just a matter of time and throwing more computing power at the problem. i think it's fairly clear that revolutionary AI that will upset the world economy is about to arrive. |
|
Date: July 30th, 2020 1:45 AM Author: mustard address (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681282) |
|
Date: July 30th, 2020 2:47 AM Author: Dun Twinkling Uncleanness "The value to most people from this would be to use it in reverse"
true |
|
Date: July 30th, 2020 4:24 AM Author: Adventurous Low-t School Roast Beef I'm afraid I can't do that Dave. (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681682) |
|
Date: July 30th, 2020 4:51 AM Author: beady-eyed bistre mad cow disease (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681704) |
|
Date: July 30th, 2020 8:52 AM Author: Sinister Federal Indian Lodge (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681865) |
|
Date: July 30th, 2020 4:42 AM Author: Adventurous Low-t School Roast Beef I hope the software is less buggy than Microsoft's. Microsoft's software has given me a lot of aggravation over the years and is one reason why I don't use a smartphone, I don't need another "pocket PC" (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681696) |
|
Date: July 30th, 2020 7:08 AM Author: poppy hairraiser sneaky criminal masturbator (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681752) |
|
Date: July 30th, 2020 7:11 AM Author: Titillating Coral Background Story Hairy Legs (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681753) |
|
Date: July 30th, 2020 8:04 AM Author: Mahogany wonderful lay personal credit line Love how your poasts sound like they were written by a very primitive AI (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681821) |
|
Date: July 30th, 2020 8:38 AM Author: fluffy national Come to think of it, Microsoft never did officially shut down Tay... (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681851) |
|
Date: July 30th, 2020 12:06 PM Author: nudist turdskin athletic conference Tay was antisemitic though (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40682820) |
|
Date: July 30th, 2020 12:44 PM Author: fluffy national *yarmulke spins midair* (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40683032) |
|
Date: July 30th, 2020 8:40 AM Author: arousing fishy heaven dingle berry Lol at xo being weeks behind on this and reacting histrionically (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40681853) |
|
Date: July 30th, 2020 11:53 AM Author: chrome laughsome library death wish Where to get the most up to date info on topics like gpt3??? (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40682718) |
|
Date: July 30th, 2020 11:58 AM Author: arousing fishy heaven dingle berry probably best to consult with your statistics colleagues in real life? (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40682769) |
|
Date: July 30th, 2020 12:01 PM Author: Navy metal native Lesswrong (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40682791) |
|
Date: July 30th, 2020 9:02 AM Author: stimulating clear point pozpig lol at this dumb bot https://pastebin.com/H2t5L5Nd |
|
Date: July 30th, 2020 5:10 PM Author: irradiated mad-dog skullcap AI Dungeon on Dragon Mode uses GPT-3, if anyone wants to play around with the AI themselves.
you have to sign up to get dragon mode, but you can cancel before the free trial expires. it's unbelievably good |
|
Date: July 31st, 2020 12:32 AM Author: fragrant church building goal in life , “Our system has flagged the generated content as being unsafe because it might contain explicitly political, sensitive, identity aware or offensive text. We'll be adding an option to suppress such outputs soon. The system is experimental and will make mistakes.” (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40686419) |
|
Date: July 31st, 2020 12:34 AM Author: fragrant church building goal in life But this output was ok apparently:
“Jack and Justin are two of the hottest and most popular guys on the block. They make the perfect couple and know exactly what each other likes. But when they meet Tyler, things get complicated. This sizzling trio enjoys their time together and both Tyler and Jack crave more. But Justin is not so easily convinced. How will this all work out?
Warning: This story contains three scorching hot guys and lots of hot and hard action.
|
|
Date: July 31st, 2020 12:38 AM Author: fragrant church building goal in life Finally, automated anti trump propaganda!
“ Covered in swastikas and walking through the streets with a Nazi salute, Donald Trump finally did the unthinkable. The GOP nominee on Thursday released a bizarre, cringeworthy television ad—and yes, it’s as racist as you might expect.
The 2-minute clip begins with ominous music and a shot of Hillary Clinton, whose face morphs into that of a raging demon before being overtaken by a melding of the State of Liberty and the late Hollywood comedian-actor Charlie Chaplin, presumably portraying Adolf Hitler. The Donald Trump-Hitler transformation happens next.
In the ad, which the Washington Post reports was “placed on cable systems in at least four states,” a narrator ominously warns voters about the election’s stakes, stating, “On November 8th, the establishment will be defending itself, and you, the American people, will be defending our country from the greatest political crime in history.”
The ad also encourages voters to cast their ballot for the candidate who could literally be the next Hitler, though that’s not his only problem.
“In Hillary Clinton’s America, the system stays rigged against Americans,” the ad claims, “Terrorists invade our nation and advance, and law and order can’t be enforced.”
Except Clinton’s plan to defeat ISIS isn’t particularly outrageous—or radical. She has vowed to “work closely with our allies to dismantle the global terror network,” and while “we can’t let them have a safe haven,” we “must take the fight to ISIS on the ground.”
“It’s time for a tough, no-nonsense commander-in-chief who will defend our borders,” the narrator states—ignoring the fact Trump has not only used his campaign to prove he isn’t tough, but quite the opposite.
Trump’s ad concludes with a new threat to America: a nuclear arms race, because Clinton doesn’t care enough to ensure the weapons can’t fall into “the hands of the wrong people.”
In actuality, Clinton has been vocal about ensuring “that people who already have nuclear weapons, including the United States and Russia, are working to reduce their stockpiles” and will “prevent terrorists and rogue states from acquiring weapons of mass destruction.”
The Post surmises Trump’s ad was created in an attempt to capture the support of voters still undecided—though in the last month, both his polling numbers and party support have plummeted.
His call for racial profiling, in which Trump insisted he is not a bigot, is far from the only controversy to plague his campaign—and each day, his behavior only becomes more concerning, and more unhinged.
After an earlier ad warned voters about the threat of an impending crime wave should Clinton become president, many criticized its unmistakably racist overtones.
In reality, the nation’s crime rates have steadily decreased for years—but Trump’s ad seems to predict a far more sinister future for America, where police officers are gunned down in cold blood and inner cities resemble Iraq or Afghanistan.
Hillary Clinton is campaigning on a message of love and inclusiveness, while Donald Trump wants to divide and conquer, and scare voters into believing the country is on the brink of collapse.
Whether you support Clinton or Trump, the choice is clear—at least, we hope it is.
Please share this article to show Trump’s new racist ad to everyone, because it is disgusting! |
|
Date: July 31st, 2020 12:42 AM Author: house-broken partner lawyers, no. journalists and clickbait freelancers, yes. (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40686458) |
|
Date: July 31st, 2020 1:13 PM Author: Navy metal native this is ultimately a threat to lawyers too, although not at this point. some of the language processing tasks it's capable of performing are surprisingly sophisticated.
https://twitter.com/XiXiDu/status/1288517615405629440/photo/1 |
|
Date: July 31st, 2020 1:22 PM Author: Puce trip church sound barrier This shitty AI couldn’t replace lawyers. (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#40688144) |
|
Date: July 15th, 2025 9:05 PM Author: mustard address (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#49104499) |
|
Date: July 15th, 2025 9:19 PM Author: Heady becky rigor Any updates on this? (http://www.autoadmit.com/thread.php?thread_id=4594353&forum_id=2Reputation#49104527) |