Why three AI builders say open models are safer than closed ones
Mark Zuckerberg makes the case for releasing Llama and for many AIs instead of one, George Hotz argues that open source is itself AI safety, and Yann LeCun says open source is the answer to bias and to fears of AI doom. Their shared argument runs against the common assumption that powerful AI is safest when kept locked up, so it is worth seeing how each one reasons.
What do Zuckerberg, Hotz and LeCun see as the bigger danger with AI? Show answer
A few organisations holding it. They argue that spreading capable models widely is the safer course.
Why does Mark Zuckerberg say open software tends to be more secure? Show answer
More people look at it openly, scrutinise it and find holes, instead of relying on obscurity.
What limit does Zuckerberg put on his own argument? Show answer
He is confident open-sourcing is net positive now, but undecided once models approach superhuman intelligence.
Where does Zuckerberg say governance should focus? Show answer
On how much autonomy systems are given, since very intelligent systems could remain tools.
What alignment problem worries George Hotz? Show answer
Alignment between the user and the AI company, since a cloud AI serves its owner, who holds the off switch.
How does Hotz think AI is most likely to do harm? Show answer
Through other humans using the machine, ending society as we know it, not the human species.
What does Hotz actually demand of closed labs? Show answer
Not open weights, but disclosure of basic facts such as a model's size.
Why does Yann LeCun say AI must be open and diverse? Show answer
An unbiased AI is impossible because bias is in the eye of the beholder, so the remedy is many systems, like a free press.
Why does LeCun reject doom scenarios? Show answer
Superintelligence will arrive gradually with guardrails added along the way, and intelligence does not imply a desire to dominate.
What benchmark do they use for misuse? Show answer
Whether a model makes harm easier than a search engine already does.
Why can Meta afford to give its base model away? Show answer
It already has a huge user base and earns money through ads and business customers, so the release does not hurt.
What weakness do these conversations have as evidence? Show answer
No advocate of closed development appears, and two of the three speakers lead AI at the same company.