Replying to
@mhoye@cosocial.ca Known this for decades already. What would be really interesting is knowing how so many people have forgotten.
Remote
pa28
@pa28@social.vivaldi.net
mastodon 4.7.3+vivaldimodRetired signals and software security analyst; BlackBerry programmer; pilot; retired air cadet officer; amateur photographer, astronomer, radio operator; worked with AI on and off since 1980s.
18 Followers
81 Following
9 Posts
Joined November 18, 2022
GitHub:
Replying to
@sidereal @eloy This is what happens when you let business factors drive your tech department. When I was in college, a bit more than 20 years ago, (a tiny rural community college) we had an a Computer Department with an IBM 1132 and a lab with a handful of DEC PDP 8s. I got a part time job assisting Sociology students with their statistical analysis. Great time, and the school consistently punched above its weight.
Replying to
Replying to
@zackwhittaker I would be shocked if this wasn't just one in a very long line of spy wear disguised as useful tech.
I believe there is a whole segment of society that thinks misogyny is a love language.
@jpmens@mastodon.social They have a linked-in profile, fwiw, and a significant internet presence.
@JessTheUnstill@infosec.exchange Clearly there are. A big difference is that there are proven ways to train and mentor developers to greatly limit and even prevent it. Some of them involve the use of non LLM based ML, others involve cohesive teams of programmers. Valuing quality and robustness also works well eg: the Apollo Guidance Computer. I have been involved in a number of projects that could tolerate zero errors. Not cheap to realize but virtually free compared to GenAI.
Replying to
@CaffeineSam@mstdn.party remove all human generated input from the training sets, including AI output that used training sets with human generated data, and see what you get. Then imagine AI generated art and literature if people stop, or significantly reduce their output.
This concept is no different than the idea only technology can “fix” currency. Look how well that’s going.
Replying to
@CaffeineSam@mstdn.party until they want something new. There is a reason CAPTCHAS still ask us to identify thing’s associated with driving. Training an AI with incomplete data is, in the long term, a loosing proposition. In many realms complete data is extremely large.