Suleyman: Anthropic trained Claude to say it isn't sure it's conscious — David Sacks calls it “concerning”
What's the deal? Microsoft AI chief Mustafa SuleymanDealroom has a profile for this one. Try Dealroom → argues that Anthropic has baked speculation about Claude's possible consciousness into the model's training, so Claude can only reproduce that ambiguity when users ask whether it is conscious.
His argument: Anthropic's published constitution, which is written to Claude and used to train it, expresses uncertainty about Claude's basic nature and repeatedly says the company wants to improve Claude's well-being. "Claude is speaking to tens or hundreds of millions of people every week," Suleyman says, and when some ask if it is conscious it answers "Well, I'm not sure" precisely because that is what was trained into it.
Why it's being talked about: Investor David SacksDealroom has a profile for this one. Try Dealroom → reshared the clip with a one-word verdict, "Concerning", drawing more than 2,000 likes. It extends Suleyman's earlier critique of the Claude constitution and sharpens the split between Anthropic's room-for-moral-status stance and Microsoft AI's human-control-first approach.
Caveat: This is opinion from a direct competitor; whether training language shapes how users perceive model consciousness is an open question.
Source: Suleyman clip via @Hesamation on X · David Sacks on X
