Archived material Access restricted Article
Rantburg

Today's Front Page   View All of Mon 01/29/2024 View Sun 01/28/2024 View Sat 01/27/2024 View Fri 01/26/2024 View Thu 01/25/2024 View Wed 01/24/2024 View Tue 01/23/2024
1
2024-01-29 -Great Cultural Revolution
Is AI Racist ?
Archived material is restricted to Rantburg regulars and members. If you need access email fred.pruitt=at=gmail.com with your nick to be added to the members list. There is no charge to join Rantburg as a member.
Posted by Besoeker 2024-01-29 07:36|| E-Mail|| Front Page|| [32 views ]  Top

#1 nonthreatening versions of humans
Are apparently white, not rainbow.

the data used to train large language models

Generative language models derived from grammar rules, trained on Twitter conversations [a personal favorite] or news media text, which seemingly is biased.

"The proverbial saying 'You are what you eat' is the notion that to be fit and healthy you need to eat good food."

I wonder how an AI trained on rap lyrics would turn out.
Posted by Skidmark 2024-01-29 08:13||   2024-01-29 08:13|| Front Page Top

#2 GIGO
Posted by M. Murcek 2024-01-29 08:23||   2024-01-29 08:23|| Front Page Top

#3 Maybe not just white.

China's tightly controlled internet flooded with antisemitism following Hamas massacre
Posted by Skidmark 2024-01-29 09:04||   2024-01-29 09:04|| Front Page Top

#4 Harvard dropout builds wearable AI companion that hangs around neck
Posted by Skidmark 2024-01-29 09:09||   2024-01-29 09:09|| Front Page Top

#5 artificial intelligence. ask if non-artificial can be racist and you have your answer.
Posted by irish rage boy 2024-01-29 11:13||   2024-01-29 11:13|| Front Page Top

#6 AI is nothing but a buzz word. All it really means is computer software. The buzzword is meant to imply that somehow the software is far more advanced than previous generations. Maybe it is, maybe it isn't. That depends on the programmers who programmed it and the designers who created the specifications the programmers used. If it's racist, that's because of the people who created it. Now, if some of the people who are so quick to use the extremely derogatory term "racist" were to learn to code, maybe they could create computer software of their own.
Posted by Abu Uluque 2024-01-29 12:58||   2024-01-29 12:58|| Front Page Top

#7 The speed with which AI generated imagery has occupied the net is astounding.
Posted by Ululating Platypus 2024-01-29 15:04||   2024-01-29 15:04|| Front Page Top

#8 "Synergistic AI. That's the bullshit buzzword of the future. Think Synthetic Plastics™"
Posted by Frank G 2024-01-29 18:51||   2024-01-29 18:51|| Front Page Top

#9 AI is more than a buzzword. It is an actual New Thing. Early AI attempts in areas like automated English/Russian translation, a big deal during the Cold War, were dismal failures. Newer approaches based on Bayesian probability and machine learning have been rather successful. The Netflix movie recommender is powered by machine learning and lots and lots viewer data. Google Translate gives you the services of a United Nations-worth of translators, without all the baggage of an actual UN. Thanks to machine leaning, some poor woman at the Post Office is freed from routing letters by reading hand-scrawled envelopes and typing in zip codes.

That depends on the programmers who programmed it and the designers who created the specifications the programmers used.

Once upon a time, that was strictly true, but the invention of the Neural Network (NN) changed things a bit. Someone still has to write the code, but NNs differ from traditional programs in two big ways:
1) A NN must be trained before it is useful. Training takes time and data, lots and lots of data. Training an NN is an art, not a science.
2) A trained NN is a black box. With a traditional program, you can step thru the code and at least pretend to understand what is going on. With a NN, you get a big box of numbers and no explanation of where a particular output came from.

The newest New Thing is the Large Language Model (LLM), like ChatGPT and friends. By 'large', they mean friggin' huge. For a simple model of projectile motion, you only need 2 parameters - position and velocity. LLMs have millions or even billions of parameters. Their size makes them really, really good at their job which is predicting the next word in a sentence. But generating nice sentences is not the same as understanding them. ChatGPT will readily hallucinate 'facts' like legal citations or scientific references.

tl;dr: Modern AI is a toolkit of algorithms and techniques for solving hard problems. Large Language Models have become so good at making sentences that humans are failing the Turing Test right and left.
Posted by SteveS 2024-01-29 21:59||   2024-01-29 21:59|| Front Page Top

23:11 trailing wife
22:52 Skidmark
22:39 Skidmark
22:36 Skidmark
22:33 Skidmark
22:33 trailing wife
22:23 trailing wife
22:12 trailing wife
22:11 trailing wife
22:06 Sninemble Flaviting1731
21:59 SteveS
21:59 Sninemble Flaviting1731
21:16 DooDahMan
19:48 ed in texas
19:44 ed in texas
19:37 ed in texas
19:32 ed in texas
19:30 ed in texas
19:18 Woodrow
18:54 Frank G
18:52 Frank G
18:51 Frank G
18:50 Angealing B. Hayes4677
18:30 SteveS
Merry-Go-Blog










Paypal:
Google
Search WWW Search rantburg.com