I have a new favorite explanation of the limits of the current probably-based #AI tools, which also talks about how the next step needs to be the incorporation of #SymbolicLogic to create neuro-symboic systems.
M
mikemccaffrey@drupal.community
@mikemccaffrey@drupal.community
Posts
-
I have a new favorite explanation of the limits of the current probably-based #AI tools, which also talks about how the next step needs to be the incorporation of #SymbolicLogic to create neuro-symboic systems. -
Well, it no longer feels lonely griping about the #AIBubble, so that is something, I guess.Well, it no longer feels lonely griping about the #AIBubble, so that is something, I guess.
-
Bananas that we now have popular historians rattling off numbers about how much #AI usage is being subsidized, rather than just #EdZitron. -
Boy, I sure felt stupid when forgot what alias I gave to 'sudo du -sh ./* | sort -h' and it turned out to be 'duh'.Boy, I sure felt stupid when forgot what alias I gave to 'sudo du -sh ./* | sort -h' and it turned out to be 'duh'.
-
I suppose that instead of referring to #LLMs as "random word generators", it would be more accurate for me to call them "weighted random word selectors".I suppose that instead of referring to #LLMs as "random word generators", it would be more accurate for me to call them "weighted random word selectors".
-
Not sure what to make of announcements about #traceability from #AI companies, and don't trust that they aren't just generating output from the usual random word generators and then retroactively looking for a source to justify the output.Not sure what to make of announcements about #traceability from #AI companies, and don't trust that they aren't just generating output from the usual random word generators and then retroactively looking for a source to justify the output.
-
My coworker started off a prompt to #Claude with "Here's a random question for you..." and I am now dead. -
POSTGRES ONLY?!?!POSTGRES ONLY?!?! After all that?!
-
"If the task you are trying to do is not well represented—either in the base data or [reinforcement learning] data–there is no force on this planet that can make that LLM solve this problem for you.""If the task you are trying to do is not well represented—either in the base data or [reinforcement learning] data–there is no force on this planet that can make that LLM solve this problem for you."
-
Sometimes I feel like I have don't enough gibberish scrolling across my screen to look like a futuristic hacker, and sometimes a user pastes a base 64 encoded png into a text area and it takes out my local database on import. -
An interesting and accessible look at #OpenClaw and #AIAgents by #HannahFry.An interesting and accessible look at #OpenClaw and #AIAgents by #HannahFry.
I only wish she:
1. Changed her SMTP server to not actually spam everyone.
2. Referred to the agent as "it" instead of "her".
3. Didn't end with "these things are getting better fast" when nothing is being done about the fundamental underlying flaws with this entire methodology.