LLMs as circus tigers - outside the cage

A circus tiger escapes its enclosure and ransacks the local super market. Who is responsible and liable? The tiger, the supermarket or the circus …? Now think about the latest OpenAI - HugginFace “security incident”. open.substack.com/pub/logicmac...

Both the perils and prospects of AI chatbots are brought to life in the first episode of AI Confidential by Hannah Fry, Curious Films and OU/BBC. It was pleasure to be involved in the development of the series as an academic advisor. www.bbc.co.uk/programmes/m...

Anyone can say they are open-minded, but how can you prove it? We show how the 'Ideological Turing Test' can be used to measure mutual understanding across controversial topics (and the result suggest people who disagree may understand each other better than we expect) […]

Large Language Models (LLMs) as stochastic parrots? No, more like circus tigers.

open.substack.com/pub/logicmac...

We already had vibe coding, now we also have vibe accounting www.itpro.com/software/mic...

Why conversational LLMs aren’t speakers, in a nutshell (i.e. an argument map): coding2learn.github.io/Piwek_Are_co...
(and more detail here: oro.open.ac.uk/99476/).

open.substack.com/pub/logicmac...

A paper from the very end of 2024 by a team of researchers, most of them employed by Anthropic at some point, claims that LLMs can “fake” alignment.

But who is deceiving who?

open.substack.com/pub/logicmac...

“Our view on the matter is that thinking can’t be outsourced. Although machines can help people improve their thinking, in the end they should make the decisions and carry the responsibility for them.” www.open.edu/openlearn/mo...

1/2 If (…) the ability to tell right from wrong should turn out to have anything to do with the ability to think, then we must be able to ‘demand’ its exercise from every sane person, no matter how erudite or ignorant, intelligent or stupid, he may happen to be. (Arendt, 1978, p. 13)

Just arrived here having decided to leave the other place.