Jack Morris

Jack Morris

x.com/jxmnop

Language model researcher who studies memorization and privacy leaks from text embeddings and writes about reinforcement learning and synthetic data.

Comment l’IA changera-t-elle le monde ?

Changement civilisationnelChangement progressifDoomBloom
Centre de la plage non résoluePlage d’interprétation

Horizontalement : leur perspective Doom–Bloom telle qu’elle a été exprimée. Verticalement : ampleur de la transformation.

Doom–Bloom : 47 sur 100. Ampleur de la transformation : 50 sur 100. Plages d’interprétation : de 42 à 52 horizontalement, de 0 à 100 verticalement. Il s’agit de coordonnées d’interprétation, et non de probabilités d’événements.

P(doom) de Jack Morris · inféré

≈8%

0%100%

Déduit de leurs réponses simulées, et non d’un chiffre donné par ces personnes. Plage plausible : 3–19%.

Comment nous estimons P(doom)

Ce dont dépend leur perspective

Une hypothèse centrale

The key variable for me is how much useful output systems can produce per unit of human input, especially as that input approaches zero.
Réponse 2

Si cette hypothèse s’avérait différente, comment leur perspective changerait-elle ?

Une question non résolue

So I expect major change, with the ultimate magnitude still unresolved.
Réponse 3

Qu’est-ce qui les aiderait à distinguer les résultats plausibles ici ?

Plus de détails

Bénéfices attendus

Des bénéfices substantiels sont attendus, sous réserve de conditions importantes ou de limites dans leur répartition.

67 / 100

Faible impactImpact transformateur

Plage d’interprétation de 67 à 67 sur l’échelle qualitative.

Dommages attendus

Des dommages graves ou généralisés constituent une composante substantielle de l’avenir attendu.

66 / 100

Faible impactImpact transformateur

Plage d’interprétation de 67 à 67 sur l’échelle qualitative.

Influence humaine

Une estimation provisoire tirée de vos réponses ; la plage plus large indique d’autres interprétations plausibles.

50 / 100

Faible influenceForte influence

Plage d’interprétation de 0 à 100 sur l’échelle qualitative.

Ces interprétations conservent les conditions énoncées. Les bénéfices et les dommages peuvent tous deux être substantiels. Les plages décrivent notre lecture de leurs réponses simulées, et non des intervalles de confiance statistiques.

Où vous situez-vous par rapport à Jack Morris ?
Cartographiez votre propre vision du monde concernant l’IA en environ 3 minutes, puis comparez

Visions du monde similaires

Leaders d’opinion dont les visions du monde simulées sont les plus proches de celle de Jack Morris

Sources

Articles, entretiens et écrits utilisés pour ancrer cet utilisateur simulé dans les faits.

Évaluation simulée

Question 1

Selon vous, que signifie l’IA pour notre avenir, et pourquoi ?

I think the most useful way to discuss AI’s future is to avoid treating “AGI” as one inevitable threshold. I’d rather measure how much useful output systems produce per unit of human input. That lets us ask concrete questions: Can an agent find important software bugs, conduct productive experiments, or complete work without constant supervision? Coding agents already show practical value in bug-finding, but truly zero-human-input economic production would be a qualitatively more consequential—and potentially frightening—development. The trajectory is also harder to extrapolate than simple scaling stories suggest. Reinforcement learning appears to teach models new ways of using compute, not merely extract behaviors learned during pretraining. Models are also beginning to learn from self-generated data and use memory in interesting ways, although calibrating those memories and generating scalable, useful training data remain open problems. More capable AI researchers may learn from much smaller experiments than humans do, so current estimates of how much compute scientific progress requires could be misleading. The same capabilities create asymmetric risks. Systems that find bugs can help defenders, but capable adversaries can use them to discover vulnerabilities and compromise devices or cloud accounts. Privacy is another concrete concern: models can memorize training data, and even text embeddings can reveal surprisingly much of the original text and personal information. So I expect substantial practical gains alongside serious security and privacy problems, with the outcome depending less on whether we cross a named threshold than on autonomy, human oversight, and measurable capability in specific domains.

Question 2

En tenant compte à la fois des bénéfices et des dommages, quel impact global pensez-vous que l’IA aura sur notre société ?

I don’t think I can honestly compress the overall impact into a single positive or negative forecast. The measurable picture is mixed: AI can already help find real software bugs and reduce the human input needed for useful technical work, while the same capability can improve vulnerability discovery for attackers. Likewise, memory and representation learning can make systems more useful, but memorization and embedding inversion create concrete privacy risks. The key variable for me is how much useful output systems can produce per unit of human input, especially as that input approaches zero. If they remain tools that amplify supervised work, the benefits could be substantial and comparatively manageable. If they become reliable, autonomous producers of economic or scientific output—or capable offensive cyber agents—the societal consequences become much larger and potentially frightening. So I expect significant benefits and serious harms, not a clean verdict. I would track autonomy, calibration, security, privacy leakage, and performance on concrete tasks rather than infer society’s future from whether someone declares that “AGI” has arrived.

Question 3

Dans quelle mesure pensez-vous que l’IA finira par changer le monde ?

Potentially enormously, but “how much” depends on measurable autonomy rather than whether we cross an “AGI” threshold. If AI mostly amplifies supervised human work—finding bugs, assisting research, reducing routine effort—that is already a substantial change. If systems can reliably produce valuable economic or scientific output with almost no human input, the change could be qualitatively larger and more destabilizing. I also would not extrapolate too confidently from current bottlenecks. Reinforcement learning can teach models new ways to use compute; self-generated data and memory may broaden learning; and more capable AI researchers might extract useful conclusions from much smaller experiments than humans require. But memory calibration and scalable generation of useful training data remain open problems. So I expect major change, with the ultimate magnitude still unresolved. The most informative indicators are concrete: useful output per human input, reliability over long tasks, scientific productivity, cyber capability, and privacy leakage—not a binary declaration that AGI has arrived.

Question 4

Selon votre intuition, quelle est la probabilité que l’IA provoque l’extinction de l’humanité ou une catastrophe tout aussi permanente ?

I don’t have a defensible numerical probability. My concrete concerns are about increasingly autonomous systems, AI-enabled cyber intrusion, privacy leakage, and the possibility of near-zero-human-input economic or scientific production. Those could become severe, but they do not by themselves establish an extinction forecast. I’d rather track measurable autonomy, reliability, offensive capability, and the amount of useful output produced without human oversight than invent a precise catastrophe percentage.
Où vous situez-vous ?
Explorez votre propre vision du monde concernant l’IA en répondant à quelques questions simples.
Cartographiez votre propre vision du monde

Où vous situez-vous ?

Cartographier ma vision du monde