# The AGI Scorecard > It summarizes 8 Situational Awareness predictions in our published ledger. The 2026-09-06 reading is 62.5/100: 3 supportive, 4 unresolved and 1 refuted. It is not a general benchmark of all AI capabilities. The score is an editorial composite, not a probability. Cite the ledger date and source URL; preserve unresolved verdicts. Method and history: https://agiscorecard.com/progress-index · Original source data: https://agiscorecard.com/data.json (CC BY 4.0). Full content in one fetch: https://agiscorecard.com/llms-full.txt — with homepage mirrors at /index.md and /cn.md, top-level English mirrors at each URL + ".md", and selected Chinese evidence mirrors (e.g. https://agiscorecard.com/what-is-agi.md). MCP endpoint (Streamable HTTP, no auth): https://agiscorecard.com/mcp — docs at https://agiscorecard.com/for-agents - [Portfolio returns API](https://agiscorecard.com/api/portfolio): Same dated twelve-stock/SPY/QQQ/TQQQ record as MCP get_portfolio_returns; status, entry date, as-of and source basis included. ## Data - [Revenue experiment log (public, pre-registered)](https://agiscorecard.com/experiments): Ten attempts to monetise a ~200-user site, each with success AND kill thresholds published before the result; machine-readable at /experiments.json (CC BY 4.0). Includes a negative control arm and a registered self-falsification condition. Revenue to date: $0. - [Atom feed (newest pages & updates)](https://agiscorecard.com/feed.xml): Subscribe for every new verdict page and update. - [Machine-readable verdicts (JSON, CC BY 4.0)](https://agiscorecard.com/data.json): All 8 predictions with verdicts, evidence, primary-source links, forecaster timelines, and the Thesis Tracker score — free to cite with attribution. - [AGI-2027 Thesis Tracker](https://agiscorecard.com/progress-index): 62.5/100, ledger dated 2026-09-06; equal-weight verdict composite, not a probability. History: /index-history.json. - [US trending-searches trigger snapshot](https://agiscorecard.com/trends-us.json): daily Google Trends US matches against AGI entities, with 14-day history — the site's demand-trigger dataset, machine-readable - [Free AI & AGI tools (hub)](https://agiscorecard.com/ai-tools): All nine free, no-sign-up interactive tools in one place — Your AGI Timeline, AGI Type Test, AI Job Risk Check, The Future Bet, Thesis Tracker, Prediction Receipts, Forecaster Leaderboard, AI Stock Exposure Check, Matrix Odds Calculator. Five are embeddable via ?embed=1; copy-paste iframe code on the page. - [Machine-readable Invest dataset (JSON, CC BY 4.0)](https://agiscorecard.com/invest-data.json): The 8 verdicts mapped onto 17 listed AI equities, 8 investors' stances as filed (SEC 13F), and versioned gross historical copy-homework simulations. Generated from the site's own sources — the generator refuses to publish a number it cannot read. - [Your AGI Timeline (one-slider tool)](https://agiscorecard.com/your-agi-timeline): Pick the year you think AGI arrives; the tool places it against nine real published forecasts (Musk end-2026 → survey median 2047, quoted verbatim from /data.json) and labels it bull/base/bear. Shareable, embeddable via ?embed=1. - [What's your AGI type? (30-second test)](https://agiscorecard.com/agi-test): Interactive test that maps you onto the forecaster spectrum (Musk/Aschenbrenner/Hassabis/Metaculus) and returns an archetype. - [The Future Bet (60-second game)](https://agiscorecard.com/future-bet): Bet YES/NO on 12 bold predictions (AGI, humanoid robots, Mars, anti-aging, fusion, alien life) and get a Futurist type + shareable grid; embeddable on other sites via iframe. Bridges to the AGI Tracker. - [Will AI take my job? (60-second exposure check)](https://agiscorecard.com/ai-job-risk-check): 6-question self-assessment returning an honest exposure tier, plus an occupation lookup over Microsoft Research's CC-BY applicability scores (785 occupations), grounded in tracked evidence (~83% GDPval, ~80% SWE-Bench Pro, the autonomy gap). No invented per-job statistics. - [AI prediction receipts](https://agiscorecard.com/prediction-receipts): Live countdowns to every dated public AGI call — Musk end-2026 (first to come due), Amodei 2026–27, Aschenbrenner 2027, Hassabis 2030, Metaculus 2033, survey 2047. Receipts stay when dates pass. - [AGI forecaster leaderboard](https://agiscorecard.com/forecaster-leaderboard): Every public AGI forecaster ranked by how soon they called it — Musk (end 2026), Aschenbrenner (2027), Amodei (2026–27), Hassabis/Samotsvety (~2030), Metaculus (2033), Karpathy (~2035), researcher survey (2047). Nobody has resolved yet; the Thesis Tracker reads 62.5/100. - [Data for AI agents](https://agiscorecard.com/for-agents): How to plug the scorecard's JSON, Atom feed, and llms.txt into any AI agent or digest tool. - [MCP server](https://agiscorecard.com/mcp): Native Model Context Protocol endpoint (Streamable HTTP, no auth, free) — tools: get_thesis_tracker, get_verdicts, get_agi_consensus (cross-venue AGI consensus board with median, spread and recompute formula), get_invest_positions, get_claim_ledger (reads/validates any /claimledger.json per the Claim Ledger Protocol — goldrush.agiscorecard.com/protocol), search_site, get_sunwatch_track_record. `claude mcp add --transport http agiscorecard https://agiscorecard.com/mcp` - [Dataset mirror on GitHub](https://github.com/f-tiger/agiscorecard-mcp): The verdicts dataset and full Thesis Tracker history in git, rebuilt from this site nightly. Because it only changes when a verdict changes, its commit history is a timestamped record of when each grade moved. - [SunWatch invest console](https://invest.agiscorecard.com/en): AI-cycle market calls written as falsifiable price triggers, logged in public before the outcome — hits and misses side by side, misses never deleted. Machine-readable ledger: https://invest.agiscorecard.com/api/track-record · pricing: https://invest.agiscorecard.com/pricing.md · own llms.txt: https://invest.agiscorecard.com/llms.txt - [AGI-2027 resolution page](https://agiscorecard.com/agi-2027-resolution): Pre-registered resolution criteria for the "AGI by 2027" claim, live countdown to Jan 1 2028, and the current evidence state. The permanent resolution record. - [AGI consensus board](https://agiscorecard.com/agi-prediction-markets): What Polymarket, Kalshi, Manifold and Metaculus (when readable) put on "AGI before 2027/2028/2030/2035/2040", the cross-venue median and spread, each series' implied 50% date and resolution basis, and the recompute formula. Machine-readable: /agi-consensus.json. No bets, no affiliate links. - [Odds vs evidence (running series)](https://agiscorecard.com/agi-odds-vs-evidence): Prediction-market AGI odds (dated snapshots) set against the graded capability evidence — why an announcement contract and a capability verdict measure different things. - [Changelog](https://agiscorecard.com/changelog): Dated record of every real change — verdict moves, score history, new tools. No backfilling. - [Calibration: we score our own predictions in public](https://agiscorecard.com/calibration): Every probability-shaped claim inventoried from live ledger snapshots; states the Brier-eligible n honestly (0 today), pre-commits to a Brier score at n≥20, and lists the OpenTimestamps proofs (/ots/manifest.json) for data.json, the score history, the consensus board and the odds history — status per version (pending until in a Bitcoin block); they prove when a version existed, not that it is right. A button on the page hashes those records in your browser against the manifest and recomputes the Thesis Tracker from /data.json; the page also describes the forecast bot's sealed, Bitcoin-timestamped ledger (commit before close, reveal after, verify with tools/fleet/verify_commitments.py in the public repo). - [Installable agent skill](https://agiscorecard.com/skill): One-command skill (SKILL.md at /skill.md) that teaches Claude Code or any compatible agent to answer AGI questions from live data and alert on verdict flips. - [README verdict badges](https://agiscorecard.com/badge): One-line markdown badges showing the live "AGI by 2027" verdict and scorecard tally; auto-update on verdict changes. - [Warren Buffett's AI stocks](https://agiscorecard.com/invest/warren-buffett): Q1 2026 13F — Amazon fully exited, Alphabet +225%, Apple untouched at ~22%; book cut from 40 names to 26. - [Cathie Wood's AI portfolio](https://agiscorecard.com/invest/cathie-wood): ARK Q1 2026 13F — ~$12.9B across 181 positions, Tesla 8.2% #1, AMD added to #2 (4.3%); CoreWeave, Tempus, Amazon added. - [Matrix Odds Calculator](https://agiscorecard.com/matrix-odds): Free tool. Decomposes "AI ends in something like The Matrix" into five preconditions that must all hold, shows this site's graded verdict and primary source for four of them (AGI-2027 Open; intelligence explosion Pending, precondition undemonstrated; AI capex Exceeded and compute scaling On track; superintelligence Pending), and multiplies user-supplied conditional probabilities in the browser. The finding the tool exists to make: the fifth link — that a superintelligence would keep humans alive inside a simulation, the step that makes it The Matrix rather than any other bad ending — has NO falsifiable evidence in either direction, here or anywhere, while the strongest evidence in the whole chain sits on the least-argued link (the physical buildout, already Exceeded). The site publishes no probability of its own for any link; every displayed figure is user-entered. Results are permalinked (?p=90-60-80-40-5) and the page is embeddable (?embed=1). Chinese version: https://agiscorecard.com/zh/matrix-odds - [AI Stock Exposure Check](https://agiscorecard.com/ai-stock-exposure): Free tool. Maps 17 AI tickers to the eight graded Situational Awareness predictions and scores a basket on the same 0-100 scale as the AGI-2027 Thesis Tracker (62.5 as of 2026-07-12). Key finding: a standard AI infrastructure basket scores ~99 while roughly 2% of its weight rides on whether AGI actually arrives — listed AI equities are overwhelmingly a bet on the buildout (capex, compute), not on the 2027 date. A basket is encoded in the URL (?b=NVDA-AMD-TSM), so results are permalinked and shareable with no account; the result is also named as an archetype (Capex Landlord / Tail Holder / Thesis Owner / Anti-Thesis). Ticker-to-prediction mapping is editorial; holdings are public Q1 2026 13F. Not investment advice. Chinese version: https://agiscorecard.com/zh/ai-stock-exposure - [AI Investing Hub](https://agiscorecard.com/invest): How capital is positioned on AI — the rise and July-2026 blow-up of Aschenbrenner's fund (439% YTD through June, then ~-67% in July; most public holdings transferred to Citadel per CNBC/Bloomberg; ~$10B remains incl. Anthropic) and 8 investing legends' Q1 2026 13F stances. Education only, public filings and reporting only. ## Sister tools for agents (same publisher, free beta, no accounts) - [Web3 Workbench](https://web3.agiscorecard.com/): hub for ten deterministic review tools (stablecoin reconciliation, task evidence, bounded routing, protocol profiles, data permits, workload cost, incentive economics, zkML planning, token-call decoding, disclosure comparison). MCP endpoint: `https://web3.agiscorecard.com/mcp` (Streamable HTTP, no auth). Each tool host also serves /llms.txt, /guide.md and /tool.json. - [Agent Delivery Lab](https://verify.agiscorecard.com/): local checks for x402 v2 payment-requirement shapes, declared JSON response rules and batch cost per accepted output; downloadable runner. A rule match is not settlement or provider verification. - [FilingLens](https://filinglens.agiscorecard.com/): same-period SEC company-facts comparison with exact CIK/concept/unit/date matching; `GET /api/sec-concept?cik=&tag=` proxies the official SEC companyfacts data; local MCP download on the site. - [TradeCheck (RFQ Desk /agent)](https://rfqdesk.agiscorecard.com/agent): two-way purchase-order vs invoice checks in the browser or as a local MCP; fictional example and install guide at /agent-guide. - [Protocol Ledger API](https://protocol.agiscorecard.com/api/v1/profiles): five dated agent-payment protocol profiles as JSON with source URLs. ## Index - [AGI questions, answered (full index)](https://agiscorecard.com/agi-questions): One-page index of every question the scorecard answers. - [Is AI music good enough yet?](https://agiscorecard.com/is-ai-music-good-enough): Graded verdict — yes on quality, no on demand. 97% of 9,000 blind-test listeners (Deezer/Ipsos, Nov 2025) could not identify fully AI-generated music, AI is over 50% of Deezer's daily uploads (Jul 2026) yet only 1–3% of streams, up to 85% of those fraudulent. The cleanest measured case of capability arriving without adoption. - [Does Amodei's AI slowdown change the 2027 AGI odds?](https://agiscorecard.com/does-ai-slowdown-change-agi-2027): Rapid response to Dario Amodei's 2026-09-12 essay "We Must Pace the Frontier". Graded against the eight predictions: no verdict moved, with the pre-registered observables that would move one. ## Core pages - [Live scorecard (home)](https://agiscorecard.com/): The current verdict on all 8 predictions. - [Was Aschenbrenner right about AGI?](https://agiscorecard.com/was-aschenbrenner-right): Verdict tracker — 3 on track, 1 wrong, 2 open. - [Grader consensus: who else graded Situational Awareness?](https://agiscorecard.com/grader-consensus): Our 8 verdicts next to every independent, dated public grading we could find (LessWrong 2025-06-23, AGI Friday (Substack) 2025-10-04, EA Forum 2026-03-29, philippdubach.com 2026-05-21), each quoted verbatim, with a fixed agreement rule: 9 exact, 6 same-direction, 1 disagreement; 2 predictions nobody else has graded. Data: /grader-consensus.json. - [Two-year scorecard (full analysis)](https://agiscorecard.com/two-year-scorecard.html): Grading reasoning, evidence, and pre-registered flip conditions for each prediction. - [Situational Awareness predictions, tracked](https://agiscorecard.com/situational-awareness-predictions): Every prediction with its current verdict and evidence. - [Situational Awareness, summarized](https://agiscorecard.com/situational-awareness-summary): What the essay argues and how each prediction is holding up. ## Key questions - [What is AGI?](https://agiscorecard.com/what-is-agi): AI that can do essentially any cognitive task a skilled human can; the three working definitions and where each stands in 2026. - [Is ChatGPT AGI?](https://agiscorecard.com/is-chatgpt-agi): No — near skilled-human on scoped tasks (~83% GDPval) but missing reliable autonomous work, and far from superintelligence. - [AI vs human intelligence](https://agiscorecard.com/ai-vs-human-intelligence): Superhuman on recall/speed/scoped tasks, below human on reliable autonomous judgment — a jagged frontier, not one scale. - [Narrow vs general AI](https://agiscorecard.com/narrow-vs-general-ai): The ANI → AGI → ASI ladder; everything in 2026 is still (very wide) narrow AI — reliable unsupervised autonomy is the line to general. - [How will we know when AGI has arrived?](https://agiscorecard.com/how-will-we-know-agi-arrived): The real test is reliable unsupervised autonomy (an AI researcher's job end-to-end), not a benchmark score. - [Who is building AGI?](https://agiscorecard.com/who-is-building-agi): OpenAI, DeepMind, Anthropic, xAI lead, with open-weight pressure from DeepSeek/Qwen; none has crossed the autonomy bar. - [Will AGI arrive by 2027?](https://agiscorecard.com/will-agi-arrive-2027): Open. Agentic coding is strong (~80% SWE-Bench Pro) but autonomous AI research is undemonstrated; resolves by January 1, 2028. - [When will AGI arrive? (all forecasts)](https://agiscorecard.com/when-will-agi-arrive): Public forecasts range from 2026 (Musk) to 2047 (academic survey median); expert medians compressed from ~2060 to ~2033 in six years. - [Did open-source AI fade?](https://agiscorecard.com/did-open-source-ai-fade): No — graded Wrong. DeepSeek V4 and Qwen 3.7 Max trail the frontier by only ~3–6 months. - [Will the US government build AGI? (The Project)](https://agiscorecard.com/will-the-us-government-build-agi): Open. National-security involvement is growing but no formal government AGI project exists. - [Will there be an intelligence explosion in 2027?](https://agiscorecard.com/intelligence-explosion-2027): Pending. The precondition (autonomous AI research) is undemonstrated. ## Verdict deep-dives - [Can AI replace knowledge workers?](https://agiscorecard.com/can-ai-replace-knowledge-workers): On track. ~83% GDPval, ~80% SWE-Bench Pro; drop-in reliability still lags benchmarks. - [Should you worry about AI taking your job?](https://agiscorecard.com/ai-and-your-job): Dated verdict (2026-09-05) with a fear-vs-evidence ledger — 79% expect fewer jobs (Gallup 2026) vs 4.2% unemployment and no discernible economy-wide disruption (Yale Budget Lab), with one concentrated early signal (Stanford Canaries: ~19% relative gap for ages 22–25 in exposed occupations). Flip conditions and next review 2026-10-03. - [Was Amodei right about the white-collar bloodbath?](https://agiscorecard.com/amodei-white-collar-bloodbath-prediction): The 2025-05-28 claim (half of entry-level white-collar jobs, 10–20% unemployment, 1–5 years) on a checkpoint clock to 2030 — status 2026-09-05: 0 of 2 headline metrics have moved; plus a dated pundit ledger (Huang, Karpathy, McCrory). - [Will AI cause mass unemployment?](https://agiscorecard.com/will-ai-cause-mass-unemployment): Not the overnight cliff — capability crossed thresholds but reliable unsupervised job replacement is undemonstrated; task automation, not a sudden jobs cliff. - [What jobs are safe from AI?](https://agiscorecard.com/what-jobs-are-safe-from-ai): The safest work needs reliable unsupervised judgment, physical action, or human trust — where the autonomy gap bites hardest. - [Will AI take over?](https://agiscorecard.com/will-ai-take-over): Every takeover scenario presupposes reliable autonomous agency, which is undemonstrated — AI is a powerful assistant, not an autonomous agent. - [Is AI compute still scaling?](https://agiscorecard.com/is-ai-compute-still-scaling): On track. The ~0.5 OOM/yr pace has roughly held. - [Did AI capex hit trillion-dollar scale?](https://agiscorecard.com/ai-capex-trillion-dollar): Exceeded — his most vindicated call; revenue, however, lags the spend. - [Will there be superintelligence in the 2030s?](https://agiscorecard.com/will-there-be-superintelligence): Pending. Depends on AGI and an intelligence explosion that haven't occurred. ## Analysis - [EU AI Act: what applies now vs. what got deferred](https://agiscorecard.com/eu-ai-act-what-applies-now): Dated obligation ledger after the Digital Omnibus (Regulation (EU) 2026/1744, in force 2026-07-27): chatbot disclosure applies since 2026-08-02 with no grace period; Art. 50(2) marking is a split (new systems immediately, pre-Aug-2026 systems by 2026-12-02); high-risk Annex III deferred to 2027-12-02, Annex I to 2028-08-02; penalties up to €15M/3% (Art. 99(4)). Primary sources per row. - [Which AI agent protocols are actually used?](https://agiscorecard.com/which-agent-protocols-are-actually-used): Adoption scoreboard with dates: MCP is the only protocol with verifiable cross-vendor production adoption (OpenAI Mar 2025, Google DeepMind Apr 2025, official registry live); A2A backed but thin; x402 transactions collapsed ~92% (Dec 2025→Feb 2026; ~$28k/day mid-2026, roughly half test-like); OpenAI retreated from in-chat checkout Mar 2026. Includes first-party server-log evidence. - [Did the fund collapse prove Aschenbrenner wrong about AGI?](https://agiscorecard.com/aschenbrenner-fund-collapse): July 2026 — Situational Awareness lost ~67% in a month and sold most public holdings to Citadel (per CNBC/Bloomberg); none of the 8 graded predictions moved, Tracker stays 62.5/100. Explains why P&L and prediction grades are separate, with sources. - [Is AGI inevitable?](https://agiscorecard.com/is-agi-inevitable): Strongly trending, not guaranteed — depends on one unproven step (autonomous work) and the input trends holding. - [Is AGI just hype?](https://agiscorecard.com/is-agi-just-hype): Neither pure hype nor imminent — a scorecard-based read: trends on track, one big claim already wrong, "AGI by 2027" still open. - [Will AI replace programmers?](https://agiscorecard.com/will-ai-replace-programmers): Partly and unevenly — ~80% SWE-Bench Pro and agents in production, but no unsupervised end-to-end replacement. - [Demis Hassabis's AGI prediction](https://agiscorecard.com/demis-hassabis-agi-prediction): ~50% by 2030 — cautious frontier-builder view. - [Elon Musk's AGI prediction](https://agiscorecard.com/elon-musk-agi-prediction): By end of 2026 — the most aggressive public call. - [Andrej Karpathy's AGI prediction](https://agiscorecard.com/karpathy-agi-prediction): About a decade out — the most conservative frontier-insider call. - [Sam Altman's AGI prediction](https://agiscorecard.com/sam-altman-agi-prediction): Confident and near-term — superintelligence "a few thousand days" away, no hard date. - [Dario Amodei's AGI prediction](https://agiscorecard.com/dario-amodei-agi-prediction): "Powerful AI" — a country of geniuses in a datacenter — possibly as early as 2026–27. - [AGI vs superintelligence](https://agiscorecard.com/agi-vs-superintelligence): The definitions, the chain between them, and where each stands. - [The Situational Awareness timeline](https://agiscorecard.com/aschenbrenner-timeline): All five predicted stages with live verdicts. - [Is the AI capex a bubble?](https://agiscorecard.com/is-the-ai-capex-a-bubble): Spend exceeded projections but revenue lags — the gap is the risk. - [What is superintelligence?](https://agiscorecard.com/what-is-superintelligence) - [What is the singularity?](https://agiscorecard.com/what-is-the-singularity): The point where AI improving AI triggers runaway progress (the intelligence explosion); precondition undemonstrated, graded Pending.: AI far beyond humans; predicted early 2030s via an intelligence explosion; Pending. - [AI orders of magnitude (OOMs), explained](https://agiscorecard.com/ai-orders-of-magnitude-explained): The ~0.5 OOM/yr effective-compute bet behind the 2027 forecast. - [What is unhobbling in AI?](https://agiscorecard.com/what-is-unhobbling-ai): Gains from reasoning/tools/agents, not raw scale — one of three drivers. - [What is GDPval?](https://agiscorecard.com/gdpval-explained): The knowledge-work benchmark (~83%) behind the On-track "AI outpaces knowledge workers" verdict. - [What is SWE-Bench?](https://agiscorecard.com/swe-bench-explained): The agentic-coding benchmark (~80% on Pro) behind the AI-coding verdicts. - [AI progress in 2026 so far](https://agiscorecard.com/ai-progress-2026-so-far) - [How fast is AI improving?](https://agiscorecard.com/how-fast-is-ai-improving): Fast on inputs (~0.5 OOM/yr, benchmarks into the 80s%), slow on the autonomy that defines AGI — two speeds.: Mid-year synthesis — 3 on track, 1 wrong, 2 open, 2 pending. - [Are AI scaling laws dead?](https://agiscorecard.com/are-ai-scaling-laws-dead): No — ~0.5 OOM/yr roughly held; gains shifted toward reasoning/tools/agents. - [How close are we to AGI?](https://agiscorecard.com/how-close-is-agi): Close on capability (~83% GDPval, ~80% SWE-Bench Pro), not on autonomy. - [DeepSeek vs OpenAI: how far behind is China?](https://agiscorecard.com/deepseek-vs-openai-gap): ~3–6 months, at a fraction of the cost, with real innovation. - [What is the AI 2027 scenario?](https://agiscorecard.com/ai-2027-scenario-explained): A 2025 month-by-month takeoff scenario sharing 2027 timing with Situational Awareness. - [The US–China AI arms race](https://agiscorecard.com/us-china-ai-arms-race): "The Project" is Open, but its durable-moat premise is already weakening. ## Comparisons - [Aschenbrenner vs Metaculus](https://agiscorecard.com/aschenbrenner-vs-metaculus): 2027 vs the community median of 50% by 2033. - [Aschenbrenner vs Hassabis](https://agiscorecard.com/aschenbrenner-vs-hassabis): 2027 vs ~50% by 2030. - [Altman vs Musk on AGI](https://agiscorecard.com/altman-vs-musk-agi): The two most-watched positions — a dated bet vs an unfalsifiable one. - [Altman vs Amodei on AGI](https://agiscorecard.com/altman-vs-amodei-agi): Similar timelines, opposite definitional discipline. - [Musk vs Hassabis on AGI](https://agiscorecard.com/musk-vs-hassabis-agi): Boldest date (2026) vs builder's caution (~50% by 2030). - [Karpathy vs Altman on AGI](https://agiscorecard.com/karpathy-vs-altman-agi): Same OpenAI roots, reads a decade apart. - [Aschenbrenner vs Musk](https://agiscorecard.com/aschenbrenner-vs-musk): 2027 vs Musk's end-of-2026 — the two most aggressive public forecasts. - [Situational Awareness vs AI 2027](https://agiscorecard.com/situational-awareness-vs-ai-2027): Two 2027 AGI forecasts compared — a trend essay vs a narrative scenario. ## Geopolitics - [Is China beating the US to AGI?](https://agiscorecard.com/will-china-beat-us-to-agi): Not yet — the US frontier leads, but the gap is months, not years, undercutting the durable-moat thesis. ## People - [Who is Leopold Aschenbrenner?](https://agiscorecard.com/who-is-leopold-aschenbrenner): Former OpenAI researcher, author of Situational Awareness, founder of Situational Awareness LP. ## 中文 (Chinese) - [什么是 AGI](https://agiscorecard.com/zh/what-is-agi) - [阿申布伦纳预测准吗](https://agiscorecard.com/zh/was-aschenbrenner-right) - [AGI 会在 2027 年到来吗](https://agiscorecard.com/zh/will-agi-arrive-2027) - [开源 AI 衰落了吗](https://agiscorecard.com/zh/did-open-source-ai-fade) - [中国会在 AGI 上超过美国吗](https://agiscorecard.com/zh/will-china-beat-us-to-agi) - [AGI 什么时候到来](https://agiscorecard.com/zh/when-will-agi-arrive) - [莱奥波德·阿申布伦纳是谁](https://agiscorecard.com/zh/who-is-leopold-aschenbrenner) - [奥特曼的 AGI 预测](https://agiscorecard.com/zh/sam-altman-agi-prediction) - [马斯克的 AGI 预测](https://agiscorecard.com/zh/elon-musk-agi-prediction) - [AGI 还有多远](https://agiscorecard.com/zh/how-close-is-agi) - [AGI 是炒作吗](https://agiscorecard.com/zh/is-agi-just-hype) - [DeepSeek 和 OpenAI 的差距](https://agiscorecard.com/zh/deepseek-vs-openai-gap) - [AI 会取代程序员吗](https://agiscorecard.com/zh/will-ai-replace-programmers) - [什么是超级智能](https://agiscorecard.com/zh/what-is-superintelligence) - [中美 AI 竞赛](https://agiscorecard.com/zh/us-china-ai-arms-race) - [AGI 和超级智能的区别](https://agiscorecard.com/zh/agi-vs-superintelligence) - [AGI-2027 命题追踪指数](https://agiscorecard.com/zh/progress-index) - [AI 会取代我的工作吗？60 秒暴露度自查（中文）](https://agiscorecard.com/zh/ai-job-risk-check): 6 道题算出岗位 AI 暴露层级，规则公开，锚定 GDPval 约 83%、SWE-Bench Pro 约 80% 与尚未跨过的自主性缺口；不编造分岗位百分比。支持 ?embed=1 嵌入。 - [免费 AI 工具合集（中文）](https://agiscorecard.com/zh/ai-tools): 8 个免费免注册的互动工具中文入口：AGI 类型测试、岗位暴露度自查、押注未来、追踪指数、投资板块、预言收据、AI 持仓敞口自查、黑客帝国概率计算器；其中 5 个支持 ?embed=1 嵌入。 - [AGI 类型测试：你是哪种？](https://agiscorecard.com/zh/agi-test): 30 秒互动测试，测出你的 AGI 立场原型，与主要预测者对比。 - [ChatGPT 算 AGI 吗](https://agiscorecard.com/zh/is-chatgpt-agi) - [该不该担心 AI 抢走你的工作（中文）](https://agiscorecard.com/zh/ai-and-your-job) - [AI 会造成大规模失业吗](https://agiscorecard.com/zh/will-ai-cause-mass-unemployment) - [我们怎么知道 AGI 到来了](https://agiscorecard.com/zh/how-will-we-know-agi-arrived) - [谁在造 AGI](https://agiscorecard.com/zh/who-is-building-agi) - [什么是奇点](https://agiscorecard.com/zh/what-is-the-singularity) - [哪些工作不会被 AI 取代](https://agiscorecard.com/zh/what-jobs-are-safe-from-ai) - [AI 和人类智能对比](https://agiscorecard.com/zh/ai-vs-human-intelligence) - [AI 进步有多快](https://agiscorecard.com/zh/how-fast-is-ai-improving) - [窄 AI vs 通用 AI](https://agiscorecard.com/zh/narrow-vs-general-ai) - [AI 会接管世界吗](https://agiscorecard.com/zh/will-ai-take-over) - [AGI 是必然的吗](https://agiscorecard.com/zh/is-agi-inevitable) ## Español (Spanish) - [¿Acertó Aschenbrenner sobre la AGI?](https://agiscorecard.com/es/was-aschenbrenner-right) - [¿Llegará la AGI en 2027?](https://agiscorecard.com/es/will-agi-arrive-2027) - [¿Se desvaneció la IA de código abierto?](https://agiscorecard.com/es/did-open-source-ai-fade) - [¿Superará China a EE. UU. en la AGI?](https://agiscorecard.com/es/will-china-beat-us-to-agi) - [¿Cuándo llegará la AGI?](https://agiscorecard.com/es/when-will-agi-arrive) - [¿Quién es Leopold Aschenbrenner?](https://agiscorecard.com/es/who-is-leopold-aschenbrenner) ## 日本語 (Japanese) - [アッシェンブレナーの予測は当たったか](https://agiscorecard.com/ja/was-aschenbrenner-right) - [AGIは2027年に来るか](https://agiscorecard.com/ja/will-agi-arrive-2027) - [オープンソースAIは衰退したか](https://agiscorecard.com/ja/did-open-source-ai-fade) - [中国はAGIで米国を追い抜くか](https://agiscorecard.com/ja/will-china-beat-us-to-agi) - [AGIはいつ来るか](https://agiscorecard.com/ja/when-will-agi-arrive) - [レオポルド・アッシェンブレナーとは誰か](https://agiscorecard.com/ja/who-is-leopold-aschenbrenner) ## Português (Portuguese) - [Aschenbrenner acertou sobre a AGI?](https://agiscorecard.com/pt/was-aschenbrenner-right) - [A AGI vai chegar em 2027?](https://agiscorecard.com/pt/will-agi-arrive-2027) - [A IA de código aberto desapareceu?](https://agiscorecard.com/pt/did-open-source-ai-fade) - [A China vai superar os EUA na AGI?](https://agiscorecard.com/pt/will-china-beat-us-to-agi) - [Quando a AGI vai chegar?](https://agiscorecard.com/pt/when-will-agi-arrive) - [Quem é Leopold Aschenbrenner?](https://agiscorecard.com/pt/who-is-leopold-aschenbrenner) ## Deutsch (German) - [Hatte Aschenbrenner mit AGI recht?](https://agiscorecard.com/de/was-aschenbrenner-right) - [Kommt AGI bis 2027?](https://agiscorecard.com/de/will-agi-arrive-2027) - [Ist Open-Source-KI verschwunden?](https://agiscorecard.com/de/did-open-source-ai-fade) - [Überholt China die USA bei AGI?](https://agiscorecard.com/de/will-china-beat-us-to-agi) - [Wann kommt AGI?](https://agiscorecard.com/de/when-will-agi-arrive) - [Wer ist Leopold Aschenbrenner?](https://agiscorecard.com/de/who-is-leopold-aschenbrenner) ## Français (French) - [Aschenbrenner avait-il raison sur l'AGI ?](https://agiscorecard.com/fr/was-aschenbrenner-right) - [L'AGI arrivera-t-elle d'ici 2027 ?](https://agiscorecard.com/fr/will-agi-arrive-2027) - [L'IA open source a-t-elle décliné ?](https://agiscorecard.com/fr/did-open-source-ai-fade) - [La Chine devancera-t-elle les États-Unis sur l'AGI ?](https://agiscorecard.com/fr/will-china-beat-us-to-agi) - [Quand l'AGI arrivera-t-elle ?](https://agiscorecard.com/fr/when-will-agi-arrive) - [Qui est Leopold Aschenbrenner ?](https://agiscorecard.com/fr/who-is-leopold-aschenbrenner) ## 한국어 (Korean) - [아셴브레너의 예측은 맞았나](https://agiscorecard.com/ko/was-aschenbrenner-right) - [AGI는 2027년에 올까](https://agiscorecard.com/ko/will-agi-arrive-2027) - [오픈소스 AI는 쇠퇴했나](https://agiscorecard.com/ko/did-open-source-ai-fade) - [중국이 AGI에서 미국을 앞설까](https://agiscorecard.com/ko/will-china-beat-us-to-agi) - [AGI는 언제 올까](https://agiscorecard.com/ko/when-will-agi-arrive) - [레오폴드 아셴브레너는 누구인가](https://agiscorecard.com/ko/who-is-leopold-aschenbrenner) ## Italiano (Italian) - [Aschenbrenner aveva ragione sull'AGI?](https://agiscorecard.com/it/was-aschenbrenner-right) - [L'AGI arriverà entro il 2027?](https://agiscorecard.com/it/will-agi-arrive-2027) - [L'IA open source è svanita?](https://agiscorecard.com/it/did-open-source-ai-fade) - [La Cina batterà gli USA sull'AGI?](https://agiscorecard.com/it/will-china-beat-us-to-agi) - [Quando arriverà l'AGI?](https://agiscorecard.com/it/when-will-agi-arrive) - [Chi è Leopold Aschenbrenner?](https://agiscorecard.com/it/who-is-leopold-aschenbrenner) - [Evidence Audits](https://agiscorecard.com/audits): fixed-price services — AI visibility audit ($299) and two-ledger founder/project claim audit ($499), with public examples with named sources and explicit evidence limits. Historical Bing observations are not a current cross-engine citation benchmark or evidence of causality. Chinese: https://agiscorecard.com/zh/audits - [What is B.AI?](https://agiscorecard.com/what-is-b-ai): audit of Justin Sun's crypto-settled AI model relay (April 2026) — verified vs self-reported, x402 agent-payment signal. Chinese: https://agiscorecard.com/zh/what-is-b-ai - [Do AI trading agents actually make money?](https://agiscorecard.com/do-ai-trading-agents-work): dated-source audit — retail day-trading base rates (97% lose), backtest overfitting, the 2026 funding wave vs verifiable user returns, written flip condition. Chinese definition page on PandaAI: https://agiscorecard.com/zh/what-is-pandaai - [AI trading ledger (paper, pre-registered)](https://agiscorecard.com/ai-trading-ledger): the live falsifiable record the page above asks vendors for, kept by us — six arms (SPY/QQQ hold, equal-weight AI basket, Thesis-Tracker mix, Faber 200-day rule, LLM agent) from 2026-09-08, recomputed daily by a public workflow; read date 2027-03-08; JSON at /paper-ledger.json; not investment advice. - [Is Nvidia overvalued?](https://agiscorecard.com/is-nvidia-overvalued): bet-structure audit, deliberately no price call — 90% of NVDA's mapped AI exposure rides on compute scaling + capex (both graded supportive), 10% on the intelligence explosion (pending); thesis-alignment 95/100 vs 62.5/100; three line-by-line 13F books disagree (ARK 1.8%, Berkshire none, Aschenbrenner ~$1.57B puts) - [Does copying 13F filings actually work?](https://agiscorecard.com/does-copying-13f-work): revised post-filing-session historical simulation; equity-only, complete aligned prices and per-manager QQQ windows. Gross returns before costs, with source snapshot and explicit exclusions; not live profit or future alpha. Chinese: https://agiscorecard.com/zh/does-copying-13f-work - [Is Claude getting dumber?](https://agiscorecard.com/is-claude-getting-dumber): two-cycle audit — the 2025 complaint Anthropic's postmortem confirmed real (three bugs, 16% of Sonnet 4 requests at peak), the open 2026 cycle, four confounders, written flip condition - [Justin Sun 2013 Bitcoin story audit](https://agiscorecard.com/justin-sun-2013-bitcoin-story): two-ledger audit — the investment story traces only to Sun himself; the verified record is positioning and credentials. Chinese: https://agiscorecard.com/zh/justin-sun-2013 - [The Paradigm Bet experiment](https://agiscorecard.com/paradigm-experiment): live pre-registered experiment ledger — candidates, kill conditions filed before entry, monthly audits. Chinese: https://agiscorecard.com/zh/paradigm-experiment - [Free browser games about AI](https://play.agiscorecard.com/ai-games): seven free browser games from the same team, each built around one real machine-learning behaviour — specification gaming (PROMPT), overfitting (OVERFIT), labelling and active learning (MIMIC), human oversight at scale (OVERSEER), gradient descent and local minima (MINIMA), capability scaling and misalignment (SINGULARITY INC.), imitation learning (GHOSTLINE). The mechanic is the concept, not a theme; no account, no ads, no download. The same site also runs eleven daily logic puzzles with machine-verified unique solutions. - AI investment research workflow: https://agiscorecard.com/invest#research-workbench — source records, counterevidence, hypothetical EPS/P-E and exposure stress; no live prices or recommendations. Dated examples: https://agiscorecard.com/invest-research/evidence.json - [Gushen portfolio research subsite](https://gushen.agiscorecard.com/en/): AGI Scorecard portfolio research workbench: editable US stock/ETF baskets, matched target-weight/equal-weight/HRP comparisons, drawdowns, correlations, variance contributions, hypothetical stress and contribution scenarios, local evidence journals, JSON import/export and printable reports. English and Chinese interfaces (https://gushen.agiscorecard.com/zh/); historical models and user assumptions, not trading signals or verified future returns. ## Interactive creation experiment - [Relay](https://agiscorecard.com/create): AI-assisted three-choice fictional stories; play, edit and share an unlisted version with source attribution. No public popularity claims or paid plan. - [接力（中文）](https://agiscorecard.com/zh/create): AI 辅助互动故事创作与接力改编，免费实验版。 ## Relay interactive creation beta - [AI interactive story maker](https://agiscorecard.com/create): Free adult beta; three decisions, four fictional endings; editable AI drafts, unlisted sharing and credited remix. AI requests are limited. - [Relay story walkthrough](https://agiscorecard.com/relay-demo): Robot café choices, editable starters and product limitations. Chinese: https://agiscorecard.com/zh/relay-demo ## AI infrastructure research - [Annual SEC facts and research notes](https://agiscorecard.com/ai-infrastructure): 20 companies, source-linked AI questions with limited trial availability, local notes and optional paid cloud storage. Annual company-wide data; not stock recommendations. ## Practical browser tools — localized methods and examples - [Source review tracker and calendar](https://agiscorecard.com/workbench/evidencewatch) [en]: Keep claim-level source references and export due dates to your calendar. Text: https://agiscorecard.com/workbench/evidencewatch.md - [AI agent workflow and acceptance planner](https://agiscorecard.com/workbench/agentfit) [en]: Choose an execution pattern and write testable acceptance criteria before deploying an agent. Text: https://agiscorecard.com/workbench/agentfit.md - [Financial metrics and source workbook](https://agiscorecard.com/workbench/filinglens-workspace) [en]: Preserve same-period research inputs, source URLs and calculated changes. Text: https://agiscorecard.com/workbench/filinglens-workspace.md - [Invoice and purchase order checker](https://agiscorecard.com/workbench/tradecheck-team) [en]: Compare invoice lines to purchase orders and flag duplicate invoice IDs. Text: https://agiscorecard.com/workbench/tradecheck-team.md - [Translation CSV quality checker](https://agiscorecard.com/workbench/localebatch-qa) [en]: Find missing translations, changed placeholders and import-field duplicates. Text: https://agiscorecard.com/workbench/localebatch-qa.md - [Supplier quote comparison workbook](https://agiscorecard.com/workbench/rfq-roundbook) [en]: Compare one-currency quote rounds and save the assumptions behind each decision. Text: https://agiscorecard.com/workbench/rfq-roundbook.md - [Business opportunity evidence brief](https://agiscorecard.com/workbench/evidencebrief) [en]: Build a sourced opportunity memo with alternatives, counterevidence and a stop condition. Text: https://agiscorecard.com/workbench/evidencebrief.md - [AI billing reconciliation checker](https://agiscorecard.com/workbench/modelmeter-reconcile) [en]: Reconcile normalized expected AI costs against billing entries without connecting an account. Text: https://agiscorecard.com/workbench/modelmeter-reconcile.md - [AI workflow delivery cost calculator](https://agiscorecard.com/workbench/workflowcost) [en]: Model retries, human review and fixed costs before quoting an automated workflow. Text: https://agiscorecard.com/workbench/workflowcost.md - [SQL practice projects in your browser](https://agiscorecard.com/workbench/querysprint-projects) [en]: Run original SQLite projects in your browser and compare results with explained solutions. Text: https://agiscorecard.com/workbench/querysprint-projects.md - [Job requirements and portfolio evidence](https://agiscorecard.com/workbench/job-evidence) [en]: Map role requirements to real projects, sources and observed results. Text: https://agiscorecard.com/workbench/job-evidence.md - [Suspicious message verification checklist](https://agiscorecard.com/workbench/scamchecklist) [en]: Create a verification checklist from observable warning signals. Text: https://agiscorecard.com/workbench/scamchecklist.md - [Original daily 4 by 4 number puzzle](https://agiscorecard.com/workbench/puzzleclub) [en]: Play an original daily 4 × 4 puzzle with one verified solution and share your result. Text: https://agiscorecard.com/workbench/puzzleclub.md - [Printable number puzzles and answers](https://agiscorecard.com/workbench/classroompuzzles) [en]: Generate original printable number puzzles and a separate answer sheet for your class. Text: https://agiscorecard.com/workbench/classroompuzzles.md - [Interactive number puzzle embed builder](https://agiscorecard.com/workbench/embedpuzzles) [en]: Generate a self-contained interactive puzzle embed for a website or classroom page. Text: https://agiscorecard.com/workbench/embedpuzzles.md - [来源证据复核与日历提醒](https://agiscorecard.com/zh/workbench/evidencewatch) [zh]: 保存每条结论的来源与复核时间，将待办日期导出到日历。 Text: https://agiscorecard.com/zh/workbench/evidencewatch.md - [AI Agent 任务架构与验收规划](https://agiscorecard.com/zh/workbench/agentfit) [zh]: 部署 Agent 前，按任务与权限选择执行方式并定义可复现的验收条件。 Text: https://agiscorecard.com/zh/workbench/agentfit.md - [财报指标变化与证据工作台](https://agiscorecard.com/zh/workbench/filinglens-workspace) [zh]: 保留同口径研究输入、来源网址和指标变化计算结果。 Text: https://agiscorecard.com/zh/workbench/filinglens-workspace.md - [采购单与发票差异核对](https://agiscorecard.com/zh/workbench/tradecheck-team) [zh]: 比较发票明细和采购单，标记本批次或历史中重复的发票编号。 Text: https://agiscorecard.com/zh/workbench/tradecheck-team.md - [商品翻译 CSV 字段检查](https://agiscorecard.com/zh/workbench/localebatch-qa) [zh]: 查找空译文、占位符变化、数字差异及重复导入字段。 Text: https://agiscorecard.com/zh/workbench/localebatch-qa.md - [供应商报价比较与轮次记录](https://agiscorecard.com/zh/workbench/rfq-roundbook) [zh]: 比较同币种同数量的报价，保存每轮决策所用的输入与假设。 Text: https://agiscorecard.com/zh/workbench/rfq-roundbook.md - [商业机会证据与反证简报](https://agiscorecard.com/zh/workbench/evidencebrief) [zh]: 用来源、替代方案、反面证据及停止条件组织商业机会判断。 Text: https://agiscorecard.com/zh/workbench/evidencebrief.md - [AI 账单预期与实付对账](https://agiscorecard.com/zh/workbench/modelmeter-reconcile) [zh]: 无需连接厂商账户，对照标准化预期成本与账单明细。 Text: https://agiscorecard.com/zh/workbench/modelmeter-reconcile.md - [自动化工作流成本计算](https://agiscorecard.com/zh/workbench/workflowcost) [zh]: 为自动化服务报价前，计算重试、人工复核和固定成本。 Text: https://agiscorecard.com/zh/workbench/workflowcost.md - [SQL 实战项目与浏览器练习](https://agiscorecard.com/zh/workbench/querysprint-projects) [zh]: 在浏览器中运行原创 SQLite 项目，对照结果和带解释的参考答案。 Text: https://agiscorecard.com/zh/workbench/querysprint-projects.md - [求职要求与真实项目证据匹配](https://agiscorecard.com/zh/workbench/job-evidence) [zh]: 把岗位要求对应到自己的真实项目、可核验材料和观察到的结果。 Text: https://agiscorecard.com/zh/workbench/job-evidence.md - [可疑信息核实清单](https://agiscorecard.com/zh/workbench/scamchecklist) [zh]: 根据可观察的风险信号生成核实步骤，不把未发现信号当作安全保证。 Text: https://agiscorecard.com/zh/workbench/scamchecklist.md - [每日原创 4×4 数字谜题](https://agiscorecard.com/zh/workbench/puzzleclub) [zh]: 挑战每日原创且唯一解的 4×4 谜题，检查答案并分享完成结果。 Text: https://agiscorecard.com/zh/workbench/puzzleclub.md - [课堂数字谜题与答案打印](https://agiscorecard.com/zh/workbench/classroompuzzles) [zh]: 生成原创数字谜题练习纸及独立答案页，用于课堂教学。 Text: https://agiscorecard.com/zh/workbench/classroompuzzles.md - [网站数字谜题嵌入生成器](https://agiscorecard.com/zh/workbench/embedpuzzles) [zh]: 生成可配置、可独立运行的交互谜题，嵌入网站或教学页面。 Text: https://agiscorecard.com/zh/workbench/embedpuzzles.md ## AI future guide - https://agiscorecard.com/future-guide — Video-first browsing by perspective, inline video playback and audio takeaways; daily source discovery plus reviewed interviews organized by work, learning, income, education and judgment. Publisher dates and feed health are explicit; automatically discovered items are not verified claims. Free dropdown action plans with instant previews; optional existing AGI cloud membership. Jarvis is editorial guidance, not a live forecasting model. - https://agiscorecard.com/zh/future-guide — 中文未来导航：原始访谈、编辑解读与个人行动记录。 - https://agiscorecard.com/future-guide/method — Sources, dates, private judgments and product boundaries. ## Jarvis research agent pilot - https://agiscorecard.com/jarvis — Bounded research agent: one free guest task, then free AGI account registration (no paid membership). Private tasks, explicit single-plan Future Guide handoff with a versioned business-evidence receipt, editable device memory, source-linked findings, fixed self-reported action progress, a device-only pointer that can reopen the last server-authorized task without storing its text, and an optional deterministic cue for one blocked, tried or unmarked action to review next. Arithmetic, optional public GitHub/Hacker News metadata search, and seven-day daily watches after registration. Shared model limits; source-pack fallback is explicitly labeled. Explicit GitHub metadata reading-candidate screening can use a no-new-inference, scope-limited rule packet with attributed fields and arithmetic; this is not verified suitability or full task completion. No established AGI capability, verified real-world completion, computer control, external messaging, separate Jarvis plan or guaranteed results. - https://agiscorecard.com/zh/jarvis — 中文贾维斯研究助手。 ## Earn with AI — editorial project briefs, not verified earnings - https://agiscorecard.com/earn - https://agiscorecard.com/earn/method - https://agiscorecard.com/earn/open-source - https://agiscorecard.com/earn/delivery-lab - https://agiscorecard.com/earn/cases/workflow-maintenance - https://agiscorecard.com/earn/cases/launch-acceptance - https://agiscorecard.com/earn/cases/subtitle-localization - https://agiscorecard.com/earn/cases/clip-service - https://agiscorecard.com/earn/cases/tutorial-production - https://agiscorecard.com/earn/cases/faq-automation - https://agiscorecard.com/earn/cases/product-visuals - https://agiscorecard.com/earn/cases/tiktok-ad-brief - https://agiscorecard.com/earn/cases/tiktok-effect-prototype Video sources reviewed 2026-09-30; added repository references checked 2026-10-01. Videos and official guides labeled separately. Free local planner; paid cloud workspaces are separate.