The AI has no soul
Juan puts words to why Severo, the language app, kept failing to land. Generative AI is deeply probabilistic, and because of that it has no soul: it's neutral, it won't surprise you, it won't be anything wow. That's why its outputs are recognizable, Gemini's images have a style, Sora's videos have a style, ChatGPT's text has its tells (the em-dashes), because the probability curve never really leaves the center; a prompt nudges it a little, but it stays near the mean, harder to pull out for images than for text. Duolingo going 'AI first' has the same problem. And it's Severo's problem: generative AI can make anything, but giving it soul, the feeling a human teacher creates, is enormously hard (making it arrogant just gets tiring), which is why the app wasn't connecting. The reframe he reaches, aimed at Koby instead, is to stop making the AI the author. Let the content exist already, the books, and make the AI a companion, a Jarvis, Cobalto, who helps you understand what you're reading rather than generating it. AI as assistant to existing content, not creator of new content, is the easier, better way to use it.
// trace: where this idea came from
- ↳ video diary @ 22:52 (la IA es muy probabilística, por ende no tiene alma; es neutral, no te va a sorprender)
- ↳ video diary @ 25:29 (darle alma a la IA es muy difícil; por eso Severo no estaba llegando)
- ↳ video diary @ 27:24 (la reframe, que la IA no cree el contenido sino que sea el asistente del contenido existente (Cobalto, el Jarvis))
- ↳ Entry 266-2: When everyone uses the same AI (el mismo estilo reconocible de una IA compartida, aquí como su límite de fondo)
Juan puts words to why Severo kept failing to land. Generative AI is deeply probabilistic, and because of that it has no soul: it’s neutral, it won’t surprise you, it won’t be anything wow ▸ 22:52. That’s why its outputs are so recognizable, Gemini’s images have a style, Sora’s videos have a style, ChatGPT’s text has its tells, the em-dashes, because the probability curve never really leaves the center; a prompt nudges it a little, but it stays near the mean, and images are even harder to pull off-center than text ▸ 24:57. Duolingo going “AI first”, as its CEO put it, has the same problem.
la IA es probabilística, no tiene alma; mejor asistente del contenido que autor →
And it’s Severo’s problem. Generative AI can make anything, but giving it soul, the feeling a human teacher creates, the thing that connects, is enormously hard; making the tutor arrogant just gets tiring after a while, so the human part can’t really be handed over, which is why the app wasn’t reaching people ▸ 25:29. The reframe he reaches, aimed at Koby instead, is to stop making the AI the author. Let the content already exist, the books, and make the AI a companion, a Jarvis, Cobalto, who helps you understand what you’re reading rather than generating it, pulling from the book as context when you get lost ▸ 27:24. AI as assistant to existing content, not creator of new content, is the easier, better way to use a thing that has no soul of its own…
// continued in
no entry has continued this idea yet: the arc is still open