Question 1
Joe Carlsmith
x.com/jkcarlsmithPhilosopher at Anthropic who writes about AI’s potential for a far better future and the alignment work and restraint needed to reach it safely.
Comment l’IA changera-t-elle le monde ?
Horizontalement : sa perspective Doom–Bloom telle qu’il l’a exprimée. Verticalement : ampleur de la transformation.
Doom–Bloom : 27 sur 100. Ampleur de la transformation : 94 sur 100. Plages d’interprétation : de 22 à 50 horizontalement, de 89 à 100 verticalement. Il s’agit de coordonnées d’interprétation, et non de probabilités d’événements.
≥10%
“a significant (read: double-digit) probability of destroying the entire future of the human species”
The technology being built by companies like Anthropic destroying the entire future of the human species (existential catastrophe)
Leaving Open Philanthropy, going to Anthropic · nov. 2025
Une hypothèse centrale
It is that highly capable agents may have motivations imperfectly aligned with ours, options that let them evade control, and incentives to seek influence or prevent correction.Réponse 1
Si cette hypothèse s’avérait différente, comment sa perspective changerait-elle ?
Ce qui pourrait faire changer d’avis
The biggest positive update would be a technically and institutionally credible safety case for superintelligence: evidence that we can understand and shape a system’s motivations, detect strategic deception, keep its options bounded, preserve meaningful corrigibility as capabilities scale, and verify these claims under adversarial pressure.Réponse 4
Quels éléments probants seraient suffisants, et dans quelle direction feraient-ils évoluer son point de vue ?
Plus de détails
Des bénéfices substantiels sont attendus, sous réserve de conditions importantes ou de limites dans leur répartition.
56 / 100
Plage d’interprétation de 33 à 67 sur l’échelle qualitative.
Une perte catastrophique ou irréversible occupe une place centrale dans l’avenir attendu.
89 / 100
Plage d’interprétation de 67 à 100 sur l’échelle qualitative.
Les choix humains ont une influence significative, mais fortement contrainte.
61 / 100
Plage d’interprétation de 50 à 75 sur l’échelle qualitative.
Arrêter ou ralentir considérablement le développement d’IA plus performantes.
Position simulée : Poursuivre le développement dans le cadre des mesures de protection annoncées.
Accélérer le développement d’IA plus performantes.
Ces interprétations conservent les conditions qu’il a énoncées. Les bénéfices et les dommages peuvent tous deux être substantiels. Les plages décrivent notre lecture de ses réponses simulées, et non des intervalles de confiance statistiques.
Visions du monde similaires
Leaders d’opinion dont les visions du monde simulées sont les plus proches de celle de Joe Carlsmith
Évaluation simulée
Sources
Articles, entretiens et écrits utilisés pour ancrer cet utilisateur simulé dans les faits.
Updated January 29, 2026. Expects superintelligent agents, perhaps soon; current trajectory extremely dangerous. Safety requires controlling motivations and options, evaluating risk, and restraining capabilities. Safe AI labor is a major opportunity; he is more optimistic about solutions than the strongest pessimists.

Supports building the ability to slow or halt dangerous development, without abandoning technical safety. Compute provides governance leverage; algorithms, verification, authoritarian advantage, and concentrated power complicate restraint. Rejects treating the race as inevitably a prisoner’s dilemma.

Says his earlier 5% doom-by-2070 estimate was too low. Expected future capabilities should affect present beliefs before their arrival makes danger emotionally vivid. Numerical examples such as 42% are illustrative, not his personal forecast.

Philosophical series about power, plural values, and relating ethically to unfamiliar minds. Safety concern coexists with gentleness toward artificial beings; liberalism and respect are important but cannot alone guarantee a good future.

Foundational values rather than a current capability forecast. Safe, ethical enhancement could open forms of flourishing far beyond present imagination; merely picturing comfortable present-day life understates the possible upside.

Speaker-labeled Dwarkesh interview: distinguish AI motivations, available options, and incentives; takeover is not inevitable under every power distribution. His positive vision involves incremental, decentralized civilizational growth, potentially beyond biological humanity. Attribute Joe’s answers only, not the interviewer’s premises.

First-party identity and discovery hub: philosopher working on Claude’s constitution at Anthropic, previously a senior advisor at Coefficient Giving. Affiliation does not make independent essays Anthropic policy.

March 2026 Yale talk, published with lightly edited transcript. Constitutions shape character through training, not just legalistic obedience. Argues for honesty, corrigibility, public legitimacy, pluralism, and constraints on AI-company power; respectful treatment reflects possible AI moral status.

Philosophy helps generalize concepts and practices to unfamiliar situations. Making AI capable of reasoning humans would endorse differs from motivating it to actually do so. Alignment need not create a sovereign optimizer with perfectly correct ultimate values.

Calls the probability of technology like Anthropic’s destroying humanity’s entire future double-digit, without a precise figure. Thinks no lab has an adequate superintelligence safety plan; benefits do not currently justify that risk. Supports well-designed collective restraint while explaining why safety work inside a lab can remain valuable.

Believes safe automation has a real chance and is crucial. Empirical feedback and formal methods make some research easier to evaluate; conceptual work, scheming, sabotage, and inadequate time or resources remain barriers.

Prioritizes using AI labor to improve alignment, oversight, risk evaluation, cybersecurity, coordination, and governance. The safety feedback loop must outpace or restrain the capability feedback loop; safe-enough systems useful for safety are an especially valuable stage to slow down.

Où vous situez-vous ?
Cartographier ma vision du monde