https://taha-research-platform.vercel.app/_next/static/chunks/app/(site)/forecasting/page-f3cb7c02a0d41354.js

1(self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[5593],{4494:(e,t,a)=>{"use strict";a.d(t,{Q:()=>r});var i=a(30926);let r=(0,i.createServerReference)("7f65fbedbfa94eb392fd087030c96ebb17af84acc0",i.callServer,void 0,i.findSourceMapURL,"revalidateSyncTags")},25016:(e,t,a)=>{"use strict";a.d(t,{We:()=>h,Yq:()=>l,Zk:()=>b,cn:()=>o,jf:()=>p,kz:()=>c,tG:()=>m,xv:()=>y,yC:()=>g});var i=a(2821),r=a(75889),n=a(75334),s=a(26467);function o(){for(var e=arguments.length,t=Array(e),a=0;a<e;a++)t[a]=arguments[a];return(0,r.QP)((0,i.$)(t))}function l(e){let t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:"MMM d, yyyy";try{return(0,n.GP)((0,s.H)(e),t)}catch(t){return e}}function c(e){return l(e,"MMM yyyy")}let d={alignment:"Alignment",interpretability:"Interpretability",governance:"Governance",forecasting:"Forecasting",rlhf:"RLHF","mechanistic-interpretability":"Mech. Interp.","scaling-laws":"Scaling Laws","institutional-risk":"Institutional Risk",deception:"Deception","constitutional-ai":"Constitutional AI","mesa-optimization":"Mesa-Optimization",corrigibility:"Corrigibility",robustness:"Robustness",agency:"Agency","compute-governance":"Compute Gov."}

1,u={alignment:"bg-accent-blue/15 text-accent-blue border-accent-blue/25",interpretability:"bg-accent-sage/15 text-accent-sage border-accent-sage/25",governance:"bg-accent-gold/15 text-accent-gold border-accent-gold/25",forecasting:"bg-purple-900/20 text-purple-300 border-purple-700/25",rlhf:"bg-accent-sage/10 text-accent-sage border-accent-sage/20","mechanistic-interpretability":"bg-teal-900/15 text-teal-400 border-teal-700/25","scaling-laws":"bg-orange-900/15 text-orange-400 border-orange-700/25","institutional-risk":"bg-accent-crimson/15 text-accent-crimson border-accent-crimson/25",deception:"bg-accent-crimson/10 text-rose-400 border-rose-700/25","constitutional-ai":"bg-sky-900/15 text-sky-400 border-sky-700/25","mesa-optimization":"bg-yellow-900/15 text-yellow-500 border-yellow-700/25",corrigibility:"bg-indigo-900/15 text-indigo-400 border-indigo-700/25",robustness:"bg-lime-900/15 text-lime-500 border-lime-700/25",agency:"bg-violet-900/15 text-violet-400 border-violet-700/25","compute-governance":"bg-amber-900/15 text-amber-400 border-amber-700/25"};function b(e){var t;return null!=(t=d[e])?t:e}function g(e){var t;return null!=(t=u[e])?t:"bg-surface-overlay text-text-secondary border-border"}let m={"agi-timelines":"AGI Timelines","alignment-progress":"Alignment Progress","governance-outcomes":"Governance Outcomes","compute-trends":"Compute Trends","institutional-preparedness":"Institutional Prep.",capabilities:"Capabilities"};function h(e){return e>=75?"text-accent-crimson":e>=50?"text-orange-400":e>=25?"text-accent-gold":"text-accent-sage"}function p(e){return e>=75?"bg-accent-crimson":e>=50?"bg-orange-500":e>=25?"bg-accent-gold":"bg-accent-sage"}function y(e,t){return e.length<=t?e:e.slice(0,t).replace(/\s+\S*$/,"")+"…"}},31473:(e,t,a)=>{"use strict";a.d(t,{ForecastDashboard:()=>d});var i=a(95155),r=a(12115),n=a(92284);let s=[{_id:"forecast-001",_createdAt:"2026-05-01T00:00:00Z",_updatedAt:"2026-05-01T00:00:00Z",slug:{current:"agi-by-2030"},question:"Will a system meeting a reasonable definition of AGI be publicly demonstrated by January 1, 2030?",category:"agi-timelines",currentProbability:35,probabilityHistory:[{_key:"p1",date:"2026-05-01T00:00:00Z",probability:35,note:"Initial estimate. AGI timelines are highly sensitive to definitional choices. Under a definition requiring human-level performance across a broad task distribution, 2030 requires sustained capability growth at roughly current rates for approximately four years — plausible but not certain given the difficulty of the remaining gaps in reasoning, planning, and embodied generalisation."}],rationale:"AGI timelines are highly sensitive to definitional choices. Under a definition requiring human-level performance across a broad task distribution, 2030 requires 3–4 more years of roughly current capability growth rates. This is plausible given recent acceleration, but not certain. Key uncertainties: whether current scaling continues to yield broad capabilities or hits domain-specific ceilings; whether agentic architectures constitute a qualitative advance; and whether there are hidden capability thresholds in the 2025–2027 compute envelope.",keyUncertainties:["Definitional ambiguity around 'AGI' — any forecast is definition-contingent","Whether algorithmic efficiency gains allow frontier capabilities at lower compute","Whether current scaling laws hold above 10^27 FLOP training runs","Whether 'reasoning' advances represent qualitative or incremental improvements"],resolution:{criteria:"A system demonstrating human-level or better performance on a pre-agreed comprehensive benchmark suite, including novel problem-solving, open-ended reasoning, and cross-domain generalisation, as validated by an independent evaluation body."},targetDate:"2030-01-01T00:00:00Z",status:"active",tags:["agi-timelines","scaling-laws","forecasting"],relatedPapers:[],lastUpdated:"2026-05-01T00:00:00Z",featured:!0},{_id:"forecast-002",_createdAt:"2026-05-01T00:00:00Z",_updatedAt:"2026-05-01T00:00:00Z",slug:{current:"frontier-ai-regulation-g7-2026"},question:"Will at least 4 G7 countries have binding frontier AI regulations by end of 2026?",category:"governance-outcomes",currentProbability:48,probabilityHistory:[{_key:"p1",date:"2026-05-01T00:00:00Z",probability:48,note:"Initial estimate. The EU AI Act is in force. The UK is proceeding with sector-specific regulation. Japan has the Hiroshima Process. Canada has proposed AIDA. The US remains the critical uncertain variable — the 4/7 threshold is achievable but requires US legislative action or another major economy to move faster than current trajectories suggest."}],rationale:"The EU AI Act is effectively enacted. The UK is proceeding with sector-specific regulation. Japan has adopted the Hiroshima Process. Canada has proposed AIDA. The US remains the critical uncertain variable — executive action has occurred but binding legislative regulation before end of 2026 is uncertain given political dynamics. The 4/7 threshold is achievable but not probable without US movement.",keyUncertainties:["US legislative action timeline — highly uncertain given political dynamics","Whether EU AI Act implementing regulations are fully in force before 2026 end","Whether Japan's AI governance framework constitutes 'binding regulation'","How 'frontier AI' is defined across different national frameworks"],resolution:{criteria:"At least 4 of the G7 nations (US, UK, France, Germany, Italy, Japan, 

1Canada) have enacted legislation or binding executive regulation specifically addressing frontier AI systems (>10^26 FLOP training threshold or equivalent capability criteria)."},targetDate:"2026-12-31T00:00:00Z",status:"active",tags:["governance","institutional-risk"],relatedPapers:[],lastUpdated:"2026-05-01T00:00:00Z",featured:!0},{_id:"forecast-003",_createdAt:"2026-05-01T00:00:00Z",_updatedAt:"2026-05-01T00:00:00Z",slug:{current:"mechanistic-interpretability-causal-account-2027"},question:"Will mechanistic interpretability research yield a validated causal account of a dangerous behaviour in a frontier model by end of 2027?",category:"alignment-progress",currentProbability:38,probabilityHistory:[{_key:"p1",date:"2026-05-01T00:00:00Z",probability:38,note:"Initial estimate. Sparse autoencoders, attention head analysis, and circuit-level work have all made progress. However, frontier model interpretability at the level required to identify dangerous behaviours causally — not just associatively — remains out of reach. The question requires validating causal claims, not just identifying features."}],rationale:"Mechanistic interpretability is advancing faster than expected. Sparse autoencoders, attention head analysis, and circuit-level work have made genuine progress. However, interpretability at frontier scale (70B+) with causal validation remains out of reach. The question requires not just identifying features associated with dangerous behaviour, but validating causal claims through activation patching or equivalent methodology — a substantially harder bar.",keyUncertainties:["Whether sparse autoencoder techniques scale to frontier model sizes","How 'dangerous behaviour' is operationalised for validation purposes","Whether interpretability tools developed on smaller models transfer to scale","Resource allocation decisions at the major safety-focused labs"],resolution:{criteria:"A peer-reviewed paper or credible technical report providing a mechanistic account (circuit-level or feature-level causal graph) of a behaviour that an independent safety evaluator classifies as dangerous, validated by activation patching or equivalent causal intervention methodology."},targetDate:"2027-12-31T00:00:00Z",status:"active",tags:["mechanistic-interpretability","alignment"],relatedPapers:[],lastUpdated:"2026-05-01T00:00:00Z",featured:!0},{_id:"forecast-004",_createdAt:"2026-05-01T00:00:00Z",_updatedAt:"2026-05-01T00:00:00Z",slug:{current:"compute-threshold-10e28-2027"},question:"Will a publicly announced AI training run exceed 10^28 FLOP by end of 2027?",category:"compute-trends",currentProbability:58,probabilityHistory:[{_key:"p1",date:"2026-05-01T00:00:00Z",probability:58,note:"Initial estimate. The Stargate project implies infrastructure capable of 10^28+ FLOP training runs. Current frontier runs are in the 10^25–10^26 FLOP range. A 100–1000x increase requires massive infrastru

1cture build-out. The 2027 timeline is tight but consistent with announced investment trajectories."}],rationale:"The Stargate project implies compute infrastructure capable of 10^28+ FLOP training runs. Current frontier training runs are in the 10^25–10^26 FLOP range. A 100–1000x increase requires massive infrastructure build-out and economic justification. The 2027 timeline is tight but plausible given announced investment trajectories — the main constraints are energy availability and whether announced projects materialise on schedule.",keyUncertainties:["Whether announced infrastructure investments materialise on schedule","Algorithmic efficiency gains reducing effective compute requirements","Energy availability and datacentre construction constraints","Whether results from current scale justify continued investment"],resolution:{criteria:"A credible public announcement or technical report confirming a completed training run exceeding 10^28 FLOPs as measured by standard computation accounting methodology."},targetDate:"2027-12-31T00:00:00Z",status:"active",tags:["compute-governance","scaling-laws","forecasting"],relatedPapers:[],lastUpdated:"2026-05-01T00:00:00Z",featured:!1},{_id:"forecast-005",_createdAt:"2026-05-01T00:00:00Z",_updatedAt:"2026-05-01T00:00:00Z",slug:{current:"ai-incident-1000-deaths-2028"},question:"Will an AI-related incident causing more than 1,000 deaths occur before 2028?",category:"institutional-preparedness",currentProbability:11,probabilityHistory:[{_key:"p1",date:"2026-05-01T00:00:00Z",probability:11,note:"Initial estimate. Primary pathways: AI-enabled cyberattacks on critical infrastructure and AI-assisted autonomous weapons in active conflict zones. Purely misaligned AI causing deaths at this scale by 2028 is assigned very low probability given current capability levels. The estimate is intentionally conservative given attribution ambiguity — AI-involved incidents may not be classified as AI-caused."}],rationale:"The estimate is driven primarily by two pathways: AI-enabled cyberattacks on critical infrastructure (hospitals, power grids), and AI-assisted autonomous weapons systems in active conflict zones. Purely misaligned AI causing deaths at this scale by 2028 is very low probability given current capability levels. The estimate is conservative — attribution ambiguity is a genuine problem, as AI-involved incidents may not be classified as AI-caused by official investigations.",keyUncertainties:["Definition of 'AI-related' — attribution is difficult for hybrid human-AI incidents","Autonomous weapons systems deployment in active conflict zones","AI-enabled cyberattack sophistication and critical infrastructure vulnerability","Medical AI failure modes at scale"],resolution:{criteria:"A documented incident where AI systems played a significant causal role (as assessed by an independent expert panel) resulting in at least 1,000 directly attributable deaths, reported by credible media sources and confirmed by relevant governmental authorities."},targetDate:"2028-01-01T00:00:00Z",status:"active",tags:["institutional-risk","governance"],relatedPapers:[],lastUpdated:"2026-05-01T00:00:00Z",featured:!0}];var o=a(25016);let l=["all","agi-timelines","alignment-progress","governance-outcomes","compute-trends","institutional-preparedness","capabilities"],c={block:{normal:e=>{let{children:t}=e;return(0,i.jsx)("p",{className:"text-sm leading-relaxed mb-3",style:{color:"var(--text-secondary)"},children:t})},h1:e=>{let{children:t}=e;return(0,i.jsx)("h2",{className:"text-lg font-medium mt-6 mb-2 text-[var(--text-primary)]",children:t})},h2:e=>{let{children:t}=e;return(0,i.jsx)("h3",{className:"text-base font-medium mt-4 mb-2 text-[var(--text-primary)]",children:t})},h3:e=>{let{children:t}=e;return(0,i.jsx)("h4",{className:"text-sm font-semibold mt-3 mb-1.5 text-[var(--text-primary)]",children:t})}},list:{bullet:e=>{let{children:t}=e;return(0,i.jsx)("ul",{className:"list-disc pl-5 mb-3 space-y-1 text-sm text-[var(--text-secondary)]",children:t})},number:e=>{let{children:t}=e;return(0,i.jsx)("ol",{className:"list-decimal pl-5 mb-3 space-y-1 text-sm text-[var(--text-secondary)]",children:t})}},marks:{link:e=>{let{value:t,children:a}=e,r=((null==t?void 0:t.href)||"").startsWith("http");return(0,i.jsx)("a",{href:null==t?void 0:t.href,target:r?"_blank":void 0,rel:r?"noopener noreferrer":void 0,className:"underline hover:opacity-80 transition-opacity",style:{color:"var(--accent-gold)"},children:a})}}};function d(e){let{forecasts:t}=e,a=[...t||[],...s],[n,c]=(0,r.useState)("all"),[d,b]=(0,r.useState)(null),g="all"===n?a:a.filter(e=>e.category===n);return(0,i.jsxs)("div",{children:[(0,i.jsx)("div",{className:"border-b",style:{borderColor:"var(--border)",background:"var(--bg-surface)"},children:(0,i.jsxs)("div",{className:"container-main py-10",children:[(0,i.jsx)("p",{className:"section-label mb-1",children:"Calibrated Probability Estimates"}),(0,i.jsx)("h1",{className:"font-display text-4xl font-medium tracking-tightest mb-3",children:"Forecasting Dashboard"})]})}),(0,i.jsxs)("div",{className:"container-main py-8",children:[(0,i.jsx)("div",{className:"flex flex-wrap gap-1.5 mb-8",children:l.map(e=>(0,i.jsx)("button",{onClick:()=>c(e),className:"px-3 py-1.5 font-mono text-2xs tracking-wide border transition-all",style:{background:n===e?"var(--accent-gold)":"var(--bg-surface)",color:n===e?"var(--text-inverse)":"var(--text-secondary)",borderColor:n===e?"var(--accent-gold)":"var(--border)"},children:"all"===e?"All Categories":o.tG[e]},e))}),(0,i.jsxs)("div",{className:"space-y-px",style:{background:"var(--border)"},children:[g.map((e,t)=>{let a=e._id||"forecast-".concat(t);return(0,i.jsx)(u,{forecast:e,isExpanded:d===a,onToggle:()=>b(e=>e===a?null:a),idForToggle:a},a)}),0===g.length&&(0,i.jsx)("div",{className:"py-16 text-center border-t",style:{background:"var(--bg-void)",borderColor:"var(--border)"},children:(0,i.jsx)("p",{className:"font-mono text-sm max-w-sm mx-auto",style:{color:"var(--text-tertiary)"},children:"No active forecasts currently tracked in this domain. Research objectives are updated as new data emerges."})})]})]})]})}function u(e){var t;let{forecast:a,isExpanded:r,onToggle:s}=e,l=a.currentProbability;return(null!=(t=a.probabilityHistory)?t:[]).map(e=>({date:(0,o.Yq)(e.date,"MMM yy"),probability:e.probability})),(0,i.jsxs)("div",{style:{background:"var(--bg-void)"},children:[(0,i.jsx)("button",{onClick:s,className:"w-full text-left px-5 py-5 transition-colors group",children:(0,i.jsxs)("div",{className:"flex items-start gap-5",children:[(0,i.jsxs)("div",{className:"flex-shrink-0 w-16 text-right font-mono text-3xl font-medium leading-none",style:{color:(0,o.We)(l)},children:[l,(0,i.jsx)("span",{className:"text-base",children:"%"})]}),(0,i.jsxs)("div",{className:"flex-1 min-w-0",children:[(0,i.jsx)("h3",{className:"font-sans text-sm font-medium leading-snug group-hover:text-[var(--accent-gold)] transition-colors pr-3",style:{color:"var(--text-primary)"},children:a.question}),(0,i.jsx)("div",{className:"prob-bar-track mt-3",children:(0,i.jsx)("div",{className:"h-full rounded-full transition-all duration-700 ".concat((0,o.jf)(l)),style:{width:"".concat(l,"%")}})})]})]})}),r&&(0,i.jsx)("div",{className:"px-5 pb-6 border-t",style:{borderColor:"var(--border)"},children:(0,i.jsx)("div",{className:"grid grid-cols-1 lg:grid-cols-[3fr_2fr] gap-8 pt-5",children:(0,i.jsxs)("div",{className:"space-y-6",children:[(0,i.jsx)("p",{className:"section-label",children:"Rationale"}),(0,i.jsx)("div",{className:"text-sm leading-relaxed",style:{color:"var(--text-secondary)"},children:Array.isArray(a.rationale)?(0,i.jsx)(n.RK,{components:c,value:a.rationale}):(0,i.jsx)("p",{children:a.rationale||"No rationale provided."})})]})})})]})}},53716:(e,t,a)=>{Promise.resolve().then(a.bind(a,31473)),Promise.resolve().then(a.bind(a,45082)),Promise.resolve().then(a.bind(a,31269)),Promise.resolve().then(a.bind(a,98061))}},e=>{e.O(0,[7723,7223,7672,4909,2284,8441,1255,7358],()=>e(e.s=53716)),_N_E=e.O()}]);