|
https://www.jcwi.org.uk/news/we-won-home-office-to-stop-using-racist-visa-algorithm The Home Office hosed up here by having and "expert" systems type thing which was obviously racist, rather than some black box ml model which was also obviously racist as it used a history of home office decisions to train on, but you could maybe convince a court otherwise
|
#
?
Aug 5, 2020 06:42
|
|
|
|
| # ? Apr 29, 2024 07:45 |
|
|
in order to proactively address the forecasted scourge of bot-generated content that will sweep the internet some new rules will be introduced for pstoing: - all p0sts must be over 2048 words long, the approximate context window gpt-3 can keep track of. this number will be increased in the future as technology progresses - pists must always end with a logical conclusion based on premises introduced throughout that pot, to further test comprehension of your own text - postos should reference recent events and knowledge, preferably newer than a week, to require continuous retraining of models - pick a word and consistently misspell it throughout the pst but never repeat the same spelling twice, we believe this will be hard for the current models to replicate - if possible have two trusted forums members vouch that they watched you type out your ospt - any self-contradiction will now be bannable as these kind of logic errors are common in neural network generated texts - be sure to attend your local pgp key signing party in prepratation for web of trust which will be mainstream any day now thus pasts should from now have word count above or equal to us covid deaths last weekend, or 75% of the tons of ammonium nitrate exploded in beirut, in order to combat bots and computer generated text thank you for your cooperation in keeping the forums clean and remember to press he report pxst button if you spot any nonhuman behaviour
|
|
#
?
Aug 5, 2020 23:07
|
|
|
you fool! you've created a simple list of instructions for a machine to follow to slip through your net!
|
|
#
?
Aug 5, 2020 23:17
|
|
|
what we really need is a machine learning model that can be used to distinguish human posts from machine-generated posts
|
|
#
?
Aug 6, 2020 01:06
|
|
|
eschaton posted:what we really need is a machine learning model that can be used to distinguish human posts from machine-generated posts the halting turing dilemma
|
|
#
?
Aug 6, 2020 07:04
|
|
|
eschaton posted:what we really need is a machine learning model that can be used to distinguish human posts from machine-generated posts you cant just delete all amberpos posters
|
|
#
?
Aug 6, 2020 17:27
|
|
|
I for one welcome
|
|
#
?
Aug 6, 2020 17:54
|
|
|
https://twitter.com/ramencult/status/1291527621222600704
|
|
#
?
Aug 7, 2020 03:08
|
|
|
I joined the ML group in a non-ML driven org (doing things that aren't ML) and the product people keep asking me for ideas on what we can use ML to solve. We're overfitting ML to problems. 
|
|
#
?
Aug 7, 2020 17:50
|
|
|
leper khan posted:I joined the ML group in a non-ML driven org (doing things that aren't ML) and the product people keep asking me for ideas on what we can use ML to solve. this is the entire industry, our whole product group got a "machine learning" mbo for the year we do financial software, the one thing our clients don't want is unpredictable results, like they pay us millions of dollars to get only exactly what they want
|
|
#
?
Aug 7, 2020 18:41
|
|
|
leper khan posted:I joined the ML group in a non-ML driven org (doing things that aren't ML) and the product people keep asking me for ideas on what we can use ML to solve. make the ultimate poker bot
|
|
#
?
Aug 7, 2020 18:58
|
|
|
qirex posted:this is the entire industry, our whole product group got a "machine learning" mbo for the year just fit a l1 regularized logistic regression model to whatever they want, thats machine learning and is far more likely to give accurate results than whatever deep learning neural network stuff they probalby have in mind
|
|
#
?
Aug 7, 2020 22:30
|
|
|
leper khan posted:I joined the ML group in a non-ML driven org (doing things that aren't ML) and the product people keep asking me for ideas on what we can use ML to solve. Sounds like you need to hire an ideas man.
|
|
#
?
Aug 7, 2020 22:42
|
|
|
leper khan posted:I joined the ML group in a non-ML driven org (doing things that aren't ML) and the product people keep asking me for ideas on what we can use ML to solve. make a webpage that says "gently caress you" louder the more recently you visited it see how long it takes for them to catch on
|
|
#
?
Aug 7, 2020 22:45
|
|
|
|
|
#
?
Aug 7, 2020 23:05
|
|
|
bragging that you only were 50% confident the cat was covid lol
|
|
#
?
Aug 7, 2020 23:09
|
|
|
50.1% is pretty much fully confident. that cat is a shapeshifting 50/50 herpes/covid-19 mimic prokaryote elon voice: "i say we nuke it from orbit" *ch-chunk whirrrrr*
|
|
#
?
Aug 7, 2020 23:16
|
|
|
Does catte has COVID? 
|
|
#
?
Aug 8, 2020 00:26
|
|
|
stop posting
|
|
#
?
Aug 8, 2020 00:40
|
|
|
It seems like nasal swab would be v.difficult to get from catte?
|
|
#
?
Aug 8, 2020 00:41
|
|
|
Carthag Tuek posted:50.1% is pretty much fully confident. that cat is a shapeshifting 50/50 herpes/covid-19 mimic prokaryote the whole "logits correspond to confidence levels" thing is pretty suspect like you train the model to output 1 for the correct label and 0 for the other ones, usually with some sort of mean squared error. that's extremely not the same as outputting actual probabilities
|
|
#
?
Aug 8, 2020 00:42
|
|
|
Since the thread seems to mostly be talking about industry, what's ml been like in academia? Genuinely interested.
|
|
#
?
Aug 8, 2020 01:18
|
|
|
Winner's Curse? On Pace, Progress, and Empirical Rigor Troubling Trends in Machine Learning Scholarship
|
|
#
?
Aug 8, 2020 01:22
|
|
|
animist posted:the whole "logits correspond to confidence levels" thing is pretty suspect nah that's a fine interpretation for a logit - you're explicitly modelling Pr(Y = 1 | X).
|
|
#
?
Aug 8, 2020 01:40
|
|
|
BrokenGameboy posted:Since the thread seems to mostly be talking about industry, what's ml been like in academia? Genuinely interested. https://www.ml.cmu.edu/ is doing fantastically well
|
|
#
?
Aug 8, 2020 02:32
|
|
|
Pinterest Mom posted:nah that's a fine interpretation for a logit - you're explicitly modelling Pr(Y = 1 | X). But is your model any good? Just cause you predict probabilities doesn't mean that your predictions aren't terrible.
|
|
#
?
Aug 8, 2020 02:47
|
|
|
ofc, but "is your model appropriate for the situation you're applying it to" and "assuming the model is appropriate and correctly specified, how do you interpret its output" are two separate questions.
|
|
#
?
Aug 8, 2020 03:04
|
|
|
can we train machines on songs ? https://open.spotify.com/track/57u4RUJ7PG5V7es1BRvv1L?si=--mq5wANRx63XAYBG86S_A
|
|
#
?
Aug 8, 2020 03:35
|
|
|
pretty sure you are training the machine just by clicking play, spotify is going to inject farts into your "made for you" playlists forever now
|
|
#
?
Aug 8, 2020 05:49
|
|
|
i tried to hit play on that track many times on my gfs spotify during 2019 so her most played list would have two hours of farts over 2000!!!
|
|
#
?
Aug 8, 2020 07:54
|
|
|
i failed. probably for the best
|
|
#
?
Aug 8, 2020 07:55
|
|
|
BrokenGameboy posted:Since the thread seems to mostly be talking about industry, what's ml been like in academia? Genuinely interested. academia doesn’t do machine learning, that’s an industry thing academia does artificial intelligence
|
|
#
?
Aug 8, 2020 09:36
|
|
|
eschaton posted:academia doesn’t do machine learning, that’s an industry thing idk how common this is but a few years ago i wrote some ml poo poo for a dry lab in college cuz i needed some money. i could see how research groups in a bunch of different fields would put a fancy buzzword on their grant apps to get more funding
|
|
#
?
Aug 8, 2020 14:35
|
|
|
eschaton posted:academia doesn’t do machine learning, that’s an industry thing I mean, I get the joke, but has there been anything interesting regarding AI - - broad definition - - that hasn't been machine learning lately?
|
|
#
?
Aug 8, 2020 20:38
|
|
|
what like expert systems
|
|
#
?
Aug 9, 2020 01:19
|
|
|
I think there's still a lot going on in the planning and decision making categories, but a lot of that stuff isn't branded so much as AI. I don't really follow specifics but I think that recent advances in algebraic topology have had a big impact on route planning and similar problems.
|
|
#
?
Aug 9, 2020 01:24
|
|
|
it is such an open-ended question, but i have in the past commented on how ml has opened up nlp research a lot.
|
|
#
?
Aug 9, 2020 08:18
|
|
|
lets train our ai on reddit whats the https://twitter.com/abidlabs/status/1291165311329341440 oh
|
|
#
?
Aug 9, 2020 23:37
|
|
|
it's just the training data, silly, and besides there's a disclaimer *5 minutes later* check this out i wrote a sentencing predictor by feeding a 5-word prompt into gpt-3. is this the future of governance?
|
|
#
?
Aug 9, 2020 23:40
|
|
|
|
| # ? Apr 29, 2024 07:45 |
|
|
https://twitter.com/paulnovosad/status/1292963252096376839
|
|
#
?
Aug 12, 2020 21:05
|
|