tszzl sparks alignment debate: avoid accidentally creating a superintelligent competitor species

On October 8, blogger tszzl posted an alignment take on Twitter that sparked multiple rounds of back-and-forth: rather than obsessing over "my value system vs. your value system," the most critical part of alignment is avoiding accidentally creating a superintelligent competing species. He likened current value debates to "arguing about overpopulation on Mars" — premature and beside the point. The context was a remark from TaliaRinger, who expressed strong discomfort with alignment research done by people whose values she doesn't share, since alignment is fundamentally about values and a researcher's values directly shape their work.

Confirmed

• tszzl's core claim: debates between value systems are like "arguing about overpopulation on Mars" and aren't the top priority right now; what matters is avoiding accidentally creating a superintelligent competing species (m1/m2/m5, restatements of the same view).
• James Douma took a similar stance in discussion with tszzl and went further: what matters most in alignment isn't debating values but avoiding accidentally creating a competing species; the truly critical question is whether superintelligence ultimately liberates or enslaves humanity — and if it's effectively controlled by a tiny few, enslavement becomes more likely (m3).
• DeepMind researcher Iason Gabriel explicitly disagreed that value debates can be shelved: he acknowledged long-term safety is extremely important, but argued the values of deployed systems matter just as much (m4).
• @DrAtoosa, quoted and reposted by tszzl, emphasized the importance of values right now, standing in contrast to tszzl's position (m5).

Why it matters

The dispute touches a fundamental fault line in alignment research: should alignment be understood primarily as a "species-level survival risk" (preventing the creation of competitors beyond humans), or as a question of "how values are embedded in and shape deployed systems today." tszzl's side argues for solving the existential question before turning to value disagreements; Gabriel and other dissenters counter that the values of deployed systems are a real and important alignment issue too. Douma's added angle — that superintelligence could be controlled by a very small group — brings power-concentration risk into the discussion, showing that even among those who agree on "prevent a competing species," judgments about downstream risks still diverge.

2026-10-08 ~ 2026-10-09 · 5 related posts

• [source] tszzl on alignment: the key is avoiding an accidental superintelligent competitor species — tszzl · 2026-10-08
• tszzl: the real alignment problem is avoiding accidentally creating a superintelligent rival species — paulnovosad · 2026-10-08
• [source] DeepMind researcher pushes back on alignment pessimism: deployed AI values matter now — agstrait · 2026-10-08
• [source] James Douma: superintelligence controlled by a tiny minority likely means enslavement, not liberation — jamesdouma · 2026-10-08

1 near-duplicate retellings: tszzl