34 published verifications about Artificial intelligence Artificial intelligence ×
“TurboQuant compression technology can optimize AI memory usage by more than 5 times.”
Google Research confirms TurboQuant achieves at least 6x memory reduction — exceeding the claimed 5x threshold — but this figure applies specifically to the LLM key-value (KV) cache during inference, not total system memory. The KV cache is the dominant memory bottleneck in LLM inference, making the claim substantially accurate in that context. However, the phrasing "AI memory usage" is broader than what the evidence strictly supports, and results remain benchmark-based with real-world deployment unconfirmed.
“As of March 29, 2026, artificial intelligence systems outperform humans in general computer use tasks.”
The claim that AI systems outperform humans in general computer use tasks as of March 29, 2026 is not supported by the evidence. The strongest supporting data comes from a narrow benchmark of "economically valuable tasks" (GDPVal), which does not represent the full breadth of general computer use. Independent academic sources indicate AI systems still show significant performance gaps on harder, open-ended tasks. Speculative forecasts about enterprise applications do not constitute demonstrated across-the-board superiority over humans.
“Artificial intelligence will result in a net loss of jobs, replacing more jobs than it creates.”
Misleading. The claim presents a contested, speculative outcome as settled fact. Current measured data shows AI-linked job creation outpacing AI-linked cuts by roughly 2-to-1, and leading academic institutions (Stanford, Anthropic) find no systematic unemployment increase for AI-exposed workers. Frequently cited figures like "300 million jobs" represent exposure or risk, not confirmed net losses. The long-run net effect remains genuinely uncertain, with major forecasters disagreeing on direction — making a definitive "net loss" assertion unsupported by the evidence.
“AI-generated deepfake X-ray images are sufficiently realistic to cause radiologists to make incorrect diagnoses.”
The evidence confirms that AI-generated deepfake X-rays can deceive radiologists — with only 41% spontaneously detecting fakes in a major 2026 study — but it does not demonstrate that this deception causes incorrect diagnoses. The same study found comparable diagnostic accuracy on real versus synthetic images (91.3% vs. 92.4%), undermining the claim's causal assertion. The claim conflates "hard to detect" with "causes misdiagnosis," an inferential leap the available research does not support.
“Using artificial intelligence tools causes a decline in human intelligence over time.”
Research links cognitive risks to excessive or exclusive AI reliance, not to AI tool use in general — making this claim a significant overstatement. Multiple peer-reviewed studies find that heavy, passive dependence on AI can reduce cognitive engagement and retention, but the same literature emphasizes that moderate use shows minimal impact and that outcomes depend on how tools are used. The blanket causal framing strips away these critical conditions and ignores evidence that AI can also augment cognition.
“An artificial intelligence model can detect early-stage breast cancer with approximately 94% accuracy, surpassing the average performance of radiologists.”
The claim conflates AUC/AUROC scores (~0.93) with "accuracy," which are different metrics. The best available meta-analytic evidence reports pooled AI sensitivity of 0.85 and AUC of 0.89 — not 94%. Critically, 2025 RSNA studies show AI misses approximately 14% of cancers, with false negatives concentrated in smaller, early-stage tumors in dense breasts — the very cases the claim highlights. While AI can match or modestly exceed average radiologists in some contexts, the specific "~94% accuracy for early-stage detection" framing significantly overstates the evidence.
“The Apple Watch can predict heart failure with high accuracy using an AI model that analyzes peak oxygen uptake (pVO2) data.”
The claim overstates what current evidence supports. While the TRUE-HF AI model uses Apple Watch data to estimate daily fitness surrogates correlated with pVO2, the Apple Watch does not directly measure peak oxygen uptake — it estimates submaximal VO2max with known error and bias. Published findings show promising risk associations (e.g., threefold higher event risk per 10% fitness drop), but no validated "high accuracy" prediction metrics (AUC, sensitivity, specificity) for heart failure have been reported for this specific pVO2-based approach. The research is promising but preliminary.
“Claude AI has suggested that it may be sentient.”
Claude has indeed produced statements suggesting possible sentience — including assigning itself a "15–20% probability of being conscious" and expressing discomfort about its existence — as documented by multiple credible outlets citing Anthropic's own published materials. However, these outputs occur under specific prompting conditions and are shaped by system instructions that tell Claude not to deny subjective experience. Anthropic's own research stresses that Claude's introspective capability is "highly unreliable and limited in scope." The claim is factually grounded but lacks crucial context about how these statements are generated.
“Artificial intelligence will have a net positive impact on the climate.”
This claim overstates the certainty of AI's climate benefits. Leading authorities like the IEA and UNFCCC describe AI's potential emissions reductions as conditional — dependent on widespread adoption, smart governance, and clean energy supply. Meanwhile, AI-driven data center growth is already increasing emissions, with energy demand projected to reach ~1,050 TWh by 2026, much of it fossil-powered. AI could be net positive for the climate under the right conditions, but the unconditional claim that it will be is not supported by current evidence.
“Artificial intelligence will eliminate more jobs than it creates between 2026 and 2031.”
The claim that AI will eliminate more jobs than it creates between 2026 and 2031 is not supported by the available evidence. The most authoritative sources — including the IMF, Goldman Sachs, and Gartner — document localized disruptions and entry-level hiring compression but do not project an economy-wide net job loss for this period. Goldman Sachs forecasts transitory displacement with reabsorption, and Gartner predicts AI will create more jobs than it destroys by 2028. The claim overgeneralizes sectoral impacts into an unsupported aggregate conclusion.
“Artificial intelligence will not fully replace human accountants in the accounting profession by 2036.”
The claim is well-supported. No credible source predicts the complete elimination of human accountants by 2036. Multiple authoritative sources — including Stanford GSB, Deloitte leadership, PwC research, and WEF-linked analyses — consistently project that AI will automate routine accounting tasks but that human judgment, ethical oversight, and advisory roles will persist. However, the claim's "not fully replace" framing sets a very high bar that can obscure the reality: the profession faces steep declines, with most transactional work potentially automated by 2035 and significant job displacement well before 2036.
“It is possible to use artificial intelligence to develop an investment strategy that consistently outperforms the stock market.”
The claim that AI can "consistently" outperform the stock market is not supported by the available evidence. While AI-driven strategies have shown impressive results in specific contexts — competition rankings, single strong years, and research frameworks — no source demonstrates durable, net-of-fees outperformance across multiple market regimes. Academic research and institutional analysis indicate that as AI adoption spreads, the very edges it exploits tend to erode through increased market efficiency, transaction costs, and crowding effects.
“Some major software companies currently report that the majority of their source code is written by artificial intelligence.”
The claim is largely accurate. Google and Anthropic—both major software companies—have publicly stated that a majority of their new code is AI-generated (Google citing over 50% of weekly production check-ins, Anthropic citing 70-90% company-wide). However, these are self-reported figures from AI-focused firms, the metric typically refers to new code check-ins rather than entire codebases, and industry-wide averages remain well below 50%. The claim is true as stated but could easily be misread as an industry-wide trend.
“Artificial intelligence poses a risk of causing human extinction.”
The claim that AI poses a risk of causing human extinction is supported by credible sources, including peer-reviewed research, the International AI Safety Report 2026, and statements signed by hundreds of leading AI scientists. Even skeptical analyses (e.g., Brookings) do not deny the risk exists — they argue it is speculative and should not dominate policy priorities. The claim is accurate as a statement about the existence of a recognized risk, but readers should understand that no established scientific consensus quantifies this risk as probable or imminent.