Well, you’d also be forgiven for thinking ‘how on earth can a social website chatbot be a white supremacist?’ And yet xAI managed to prove that is a legitimate concern.

xAI has a shocking track record of poor decisions when it comes to training and prompting their AIs. If anyone can make a partisan coding assistant, they can. Indeed, given their leadership and past performance, we might expect them to explicitly try.

▲

simianwords 5 days ago | parent | next [-]

What’s their incentive to do this? What do they gain by making a partisan model instead of one that just works well?

▲

pennomi 5 days ago | parent [-]

You really can’t think of ANY advantage to becoming a perfected propaganda machine? Not one?

▲

simianwords 5 days ago | parent [-]

Enlighten me. How would a partisan coding model help.

	▲	ben_w 5 days ago \| parent [-]
		Competence in every field is correlated for LLMs. Better coding probably means more competent rhetoric and more competent Swahili-Latin translation. But only "probably", the causation is being argued about.

▲

dudeinjapan 5 days ago | parent | prev [-]

Perhaps you’ve never heard of Tay?

Microsoft did pioneering work in the Nazi chatbot space.

	▲	arjvik 5 days ago \| parent [-]
		Fwiw Tay was unintentional and was shut down immediately upon realization… very good case study for safety folks!