Tobi Lütke sparks debate: why discriminative models can't be universal classifiers
caglar_ee · x · 2026-09-19
Shopify CEO Tobi Lütke quote-posted a former OpenAI employee's remark noting that post-ChatGPT people are baffled that the pre-ChatGPT era failed to build generic universal classifiers.
Engineer caglaree pushed back on the technical merits: he wonders how well Jev actually works and whether any benchmarks exist, arguing that discriminative / decision-boundary models are inherently bad at building generic or universal classifiers — you can't learn a large probabilistic generative model and marginalize out missing parts depending on the task.
The exchange captures the core post-launch debate around Jev: whether the discriminative "decision model" route is even sound in principle.
More from Models
- Steering Vectors as a Softer Way to Limit Reasoning Budgets? — maddie-lovelace · 2026-09-19
- Two Prompts That Expose How ChatGPT Quietly Rewrites Your Claims — KazTheMerc · 2026-09-19
- MMLU Is Riddled With Errors: 6.49% of Questions Wrong, 57% in Virology — PMinervini · 2026-09-19
- Jev's structured-output model impresses, Yegge argues fences beat sandboxes for agents — njyx · 2026-09-19
- GPT Astra 6 hits new BALROG heights with 13% NetHack progression, still "not AGI" — _rockt · 2026-09-19
- Daily AI brief: Grok Voice Transcribe 2.0 lands, Meta Muse opens to developers — testingcatalog · 2026-09-19