Kaal claims by topic: empirical-evidence, page 2

311 atomic, individually citable claims from the published work of Wulf A. Kaal tagged empirical-evidence.

In the second quarter of 2017 ICO issuances exceeded venture capital financing of start-ups for the first time, with $210 million invested in ICOs versus $180 million invested into start-ups via traditional venture capital funds. 2017
The second survey found long-term negative effects of Title IV: 34.9 percent of respondents expected it to affect the industry over the next five years through additional expenses, and 32.6 percent expected it to create barriers to entry for new private fund market entrants. 2017
Increasing the quantity of data does not dissolve the foundational methodological problems of data: construct validity, measurement, reliability, and data dependencies remain the same regardless of how much data is collected. 2017
Establishing the facts about a new technology is often impossible in practice because there is no adequate sample or other reliable data on the effects of that technology yet. 2017
Time pressure produces two distinct fact failures: the facts about a new technology may simply not exist yet, or regulators may select the wrong, contested, or irrelevant facts as the basis of regulation. 2017
Regulatory experimentation matters within a single jurisdiction and not only across jurisdictions, because it gives regulators data on the real world effects of a particular regulatory scheme in a comparable setting. 2017
Ten of the thirteen S&P 500 companies with above-average revenue growth over the 2012 to 2016 period had never paid a dividend and had never engaged in share buybacks. 2017
Only seven percent of the 250 leading companies on the Forbes Global 2000 list have a CEO who is personally active and connected on Twitter. 2017
Forty-five percent of directors at the thirteen S&P 500 companies with above-average revenue growth over the last five years are feedback providers who make board decisions more data-driven. 2017
Because tokens move in and out of the top 100 daily, the study limits its time series dataset to all available data on the top 100 cryptocurrencies until April 2018, selecting coins by market capitalization before April 2018. 2018
The tokens in the sample of the top 100 cryptocurrencies were launched between January 2009 and March 2018, with the greatest number of top 100 coins launched in a single month occurring in November 2017. 2018
Much of the stability in month over month coin launching from mid-2013 onward is attributable to the launching of the Ethereum platform. 2018
Of the top 100 tokens, fifty-six held an ICO and thirty-eight did not. 2018
Fifty-six of the top 100 tokens are implemented at the protocol level of a blockchain, thirty-five follow the ERC-20 Token Standard, six are implemented in a crypto economic protocol on top of a blockchain, and two are application level dapp tokens. 2018
Among the top 100 tokens, currency and utility models are the most common, USDT is the only purely Stablecoin token, Steem and Maker are both Stablecoin and utility, and only four tokens are asset-backed. 2018
Thirty-two of the top 100 tokens have inherent value, for example as a currency, and thirty-two derive their underlying value from giving holders permission to use a digital service. 2018
Seventy-two of the top 100 tokens cap the number of tokens that will ever be issued, a deflationary model of token issuance. 2018
The remaining twenty-eight tokens in the dataset use inflationary models that operate similarly to fiat currency, contemplating no maximum issuance and a continuing minting process that gives the issuer more flexibility. 2018
Sixty-six of the top 100 tokens are user facing, letting ecosystem participants handle the token directly, while thirty-four are layered and operate underneath another token, platform, or chain. 2018
Sixty of the top 100 tokens allow or intend inter-system functionality, while thirty-two are fundamentally restricted to use within a given ecosystem. 2018
Proof of Work remains the most popular consensus protocol type among the top 100 tokens, and the Other category alone comprises thirty-four tokens. 2018
Attempts to increase throughput and scale in consensus protocols concentrate on proof of stake, and the data are consistent with anecdotal evidence that proof of stake may be the most dominant attempt at scaling. 2018
Tokens launched before 2015 were overwhelmingly developed on a hardfork governance mechanism or a combination of hardfork and one other governance type. 2018
Outlier governance mechanisms rose sharply in the dataset: three per year in 2015 and 2016, then eighteen in 2017, and five in the first six months of 2018. 2018
Successful proof of stake experiments running today cannot be used to infer that their protocols are truly secure, because the current participant population is atypically altruistic; confidence must instead come from sound reasoning about incentives. 2018
Trust barometers such as the Edelman report show a radical depreciation of trust in centralized institutions between 2017 and 2018. 2019
Artificial intelligence and machine learning are used far more heavily for idea generation and portfolio optimization than for execution: two thirds of surveyed funds use them to generate trading ideas and optimize portfolios, while only just over a quarter use automation to execute trades. 2019
Adoption of artificial intelligence in hedge funds remains partial rather than total: more than four out of ten survey respondents still rely on conventional human thinking to guide their investment processes. 2019
Before its collapse LTCM held roughly $4.8 billion in capital while controlling $160 billion in stocks and bonds, with derivatives of a notional value of $1 trillion. 2019
Tether's growth in market capitalization, its stability around one dollar, and investors' use of it as a temporary safe haven provide some empirical support for the proposition that stable cryptocurrencies can create market stability. 2019
Transacting in cash imposes large measurable costs: roughly 200 billion dollars annually in the United States, about 637 dollars per person, driven by counting, managing, storing, transporting, guarding and accounting for bank notes. 2019
In a proprietary dataset of thirty three blockchain for good projects, the projects proliferated between 2013 and 2017 and peaked in 2017, and many of them did not launch successfully or perished over time. 2020
In the author's dataset, the overwhelming majority of blockchain for good projects cluster in healthcare, exchanges, environmental protection, and charities. 2020
Between 2011 and 2018 there were 56 cyberattacks on cryptocurrency exchanges, initial coin offerings and other digital currency platforms worldwide, totaling $1.63 billion in hacking related losses. 2021
As of 2017, 73 percent of digital asset exchanges took custody of their users' private keys while only 23 percent let users maintain control over their own keys. 2021
Custodial service providers spend proportionally less on IT security than non custodial ones: custodians spend between 6 and 10 percent of resources on IT security while non custodial service providers spend between 11 and 20 percent. 2021
Externally led auditing of digital asset reserves among custodial service providers is declining, falling 24 percentage points relative to the 2018 sample. 2021
Insurance coverage among digital asset service providers is far from universal: 46 percent of surveyed service providers reported not being insured against any risks. 2021
Open source software accounts for somewhere between 78 percent and 98 percent of all core digital infrastructure software, according to industry estimates. 2021
Member default on premia is a major inefficiency in chit funds, with estimates that a large share of subscribers have defaulted at least once recently and a substantial share have defaulted after winning an auction. 2021
Patch size degrades multiple dimensions of code review performance at once: comment density per reviewer falls as patches grow, the quality and amount of contribution is affected, and the time required to provide significant contributions increases. 2021
Survey evidence shows a public trust deficit toward charities: while 70 percent of Americans say trust is essential before making a donation, fewer than 20 percent say they highly trust charities. 2021
Cash usage in the United States, the United Kingdom, the Netherlands, Sweden, Finland, Canada, and France and other industrialized nations has fallen well below 50 percent of total transaction volume. 2021
Transacting in cash costs United States consumers roughly 200 billion dollars annually, about 637 dollars per person, driven by the costs of production, storage, and transportation. 2021
By 2012, ninety percent of the United States media and entertainment industry was controlled by the top six media conglomerations. 2021
American faith in media has fallen in direct correlation with the consolidation of broadcasting power for as long as active statistics have been studied. 2021
The Maghribi traders show that reputation alone can sustain a decentralized network under extreme information asymmetry: using only handwritten letters, Jewish merchants built a reputational system spanning the Silk Road in which nothing but the promise of better reputation deterred agents from cheating. 2021
Rug pulls grew sharply as a share of crypto crime: of the $7.7 billion in total illicit crypto revenue in 2021, 37 percent came from rug pulls, up from 1 percent of illicit revenue in 2020. 2022
DAO projects accounted for a material share of the largest crypto frauds of the year: two of the top six crypto rug pulls in 2021 were DAO projects. 2022
Fair launch tokens outperformed centrally distributed projects during the late 2020 and early 2021 rally: the collective crypto average token launch gained 112.41% over 90 days while fair launch projects gained over 296%. 2022
The paper's empirical basis is a dataset of DAOs selected by the assets held in their treasuries, drawn from across different industries. 2023
DAOs in the dataset score well below average on implementing true decentralization, averaging 3.78 out of 10, with the highest score being CRDAO at 8 out of 10 and several DAOs scoring 1 out of 10. 2023
Work to earn is the strongest of the six scored categories, yet it still averages only 4.9 out of 10 across the DAOs studied. 2023
Attack resistance across the studied DAOs averages 4.05 out of 10, and Charity DAOs perform worst in this category, with the best of them, VitaDAO, scoring 3 out of 10 and the remainder at 2 or less. 2023
Regulatory compliance is the weakest of all six categories, averaging 3.01 out of 10, and Services DAOs are the only category to outperform that average. 2023
Governance scores across the studied DAOs average 3.81 out of 10, with only a couple of DAOs scoring 7 or higher and the vast majority scoring 5 or less. 2023
Organizational communication across the studied DAOs averages 4.29 out of 10. 2023
Gamification supplies quality control by making workers review and rate each other's contributions for points or recognition, which surfaces and resolves discrepancies through consensus-based voting or peer review rather than through duplicated independent work. 2024
Combining decentralized governance with gamification of micro task work is the condition under which gamification does not compromise dataset quality and accuracy, and this combination is what allows gamified micro task work to scale high-quality diverse datasets for AI learning. 2024
Losses from smart contract vulnerabilities are large and growing, with 2021 losses alone estimated at 680 million dollars and cumulative global losses estimated at over 6 billion dollars. 2024
The study evaluates each DAO on six factors: Decentralization, Work to Earn, Attack Resistance, Regulatory Compliance, Governance, and Organizational Communication. 2024
Each DAO in the dataset was scored from zero to ten on each factor by analyzing teams, using only publicly available information and the organization's whitepaper where one existed. 2024
The author concedes that the scoring metric is imperfect and that some scored attributes may have changed by the time of publication, presenting it instead as a structured approach to evaluating what drives DAO success or failure. 2024
Across the sampled DAOs, no single industry consistently outperforms the others on total score, indicating a diverse rather than sector determined performance landscape. 2024
The average decentralization score across the sampled DAOs is 4.25, with a maximum of 9 and a minimum of 1, showing wide variability in how decentralized these organizations actually are. 2024
The average attack resistance score across the sampled DAOs is 3.65 on a scale of 0 to 10, the lowest of the measured attributes alongside regulatory compliance. 2024
The average regulatory compliance score across the sampled DAOs is 3.22, ranging from 1 to 9, and the distribution shows that many DAOs struggle with regulatory compliance while only a few achieve higher scores. 2024
The average Work to Earn score across the sampled DAOs is 4.1, ranging from 1 to 9, indicating only moderate effectiveness of contributor reward mechanisms across the market. 2024
The average organizational communication score across the sampled DAOs is 4.32, with a maximum of 10 and a minimum of 1, indicating only moderate communication effectiveness overall. 2024
The analysis reveals significant variability across all six attributes, decentralization, security, governance, regulatory compliance, work incentives, and communication effectiveness, reflecting the divergent developmental stages of DAOs. 2024
DeScience and Data Analysis DAOs generally score higher on overall performance metrics, whereas Investment DAOs and Art and Culture DAOs display idiosyncratic mixes of strengths and weaknesses. 2024
In the legal domain the adoption of transformer based language models is blocked less by capability than by resources and access: training and deployment are resource intensive and large, quality tagged legal datasets are usually restricted. 2024
Deep reinforcement learning demands large amounts of training data, which suggests its algorithms differ fundamentally from human learning, and learning without supervision becomes particularly hard when rewards are sparse, as they typically are in sequence generation tasks. 2024
Many current blockchain projects critically fail to use existing social impact evidence in their design and management, which Kaal identifies as a source of future challenges for blockchain based impact financing. 2024
Contrary to expectation, employees whose managers behave more like a boss than like a partner tend to benefit more from technological change, so a participative management style is not a precondition for workers to gain from new technology. 2024
Quantum economics models must be subjected to empirical testing and to rigorous head to head comparison with classical models across different economic contexts and datasets before their validity and generalizability can be established. 2024
Programmable tokens and smart contracts give quantum economics an experimental testbed, so contested phenomena such as preference reversal can be modeled with quantum decision theory and then empirically validated and refined rather than argued in the abstract. 2024
An operating token ecosystem such as VOW can function as empirical evidence for quantum economics, since a working decentralized system that manages complex economic phenomena supports the validity and utility of the abstract framework it instantiates. 2024
The total effective stock of human-generated text is estimated at roughly 300 trillion tokens, with a plausible range from 100 trillion to 1 quadrillion tokens. 2025
When a training dataset disproportionately represents one region or demographic group, the resulting model produces skewed and sometimes inappropriate outputs once deployed in unfamiliar settings. 2025
In healthcare, biased or stale training data produces algorithms that misdiagnose underrepresented populations and thereby reinforce existing health disparities instead of reducing them. 2025
The reputation systems of SingularityNET, Fetch.ai, Ocean Protocol, Numeraire, and DcentAI are structurally insufficient for a fully decentralized Mechanical Turk model of large-scale AI dataset creation, offering only incremental innovation. 2025
SingularityNET's service-level reputation metrics fail to capture the granular requirements of dataset creation, namely accuracy, consistency, and contextual relevance. 2025
A single reputation score, as used by DcentAI, cannot capture the interdependencies among privacy concerns, domain-specific regulation, and real-time ethical updates that dataset governance requires. 2025
Weighted voting reduces Sybil attack success rates by over eighty-five percent, but only where reputation is openly auditable. 2025
Sub-millisecond reputation refresh rates reduce successful Sybil attack probability by seventy percent in simulation. 2025
Testnet deployment must evaluate weighted voting and microsecond-scale updates against Sybil attacks, collusion, and reputation gaming across adversarial scenarios of ten, thirty-three, and fifty percent malicious participation before SPoS's theoretical strengths can be treated as practical resilience. 2025
The retention uplifts claimed for LER remain unvalidated: empirical pilots are needed to confirm them and to refine the smart contract designs, particularly for AI-personalized airdrops. 2025
Firms carrying fewer defensive mechanisms display higher market valuations and better operating performance, which supports dismantling entrenchment devices rather than adding to them. 2025
Support Vector Machine models reached 96.9% accuracy in classifying semantic biases in judicial judgments on the Chinese AI and Law dataset, outperforming Naive Bayes, multi-layer perceptron, and K-nearest neighbor classifiers. 2025
Predictive and generative AI systems threaten fairness by reducing complex judicial processes to statistical correlations, which neglects the emotive and cognitive dimensions of justice and the contextual nuances human judges consider. 2025
The overreliance concern is not uniform: some studies find that annotators do not over rely on LLM output when labelling court opinions for legally relevant factors, and other studies find an aversion to AI generated legal content consistent with algorithmic aversion in other domains. 2025
Experimental studies suggest risks such as bias amplification from judicial AI, but there is little empirical data on how those risks actually manifest in operational courtrooms, which leaves practical integration guidelines undeveloped. 2025
A Dutch court's AI system for traffic violation appeals improved consistency yet altered legal experts' decisions, which shows that consistency gains from judicial AI can come at the cost of influencing expert judgment and therefore require human oversight. 2025
Reputation-weighted micro-task systems can potentially reduce human task redundancy from ten-to-fifteen-fold to five-fold or below through consistent, high-reputation agent outputs, accelerating production of superior datasets in sectors such as healthcare and finance. 2026
The retroactive citation audit requires a dispute resolution protocol and evidence standards, which can be developed through the contentious debate mechanism moving from loosely-coupled validation pools to tightly-coupled votes as consensus emerges. 2026
Selecting one agent per job by weighted random selection appears efficient but is mathematically suboptimal for quality assurance, since expected quality equals only the reputation weighted average rather than the best available output. 2026
Multi agent competitive collaboration captures a diversity dividend worth more than half a standard deviation of quality improvement while enabling attribution through citation graphs. 2026
Hallucination rates vary sharply by domain: leading frontier models achieve sub one percent rates for general knowledge queries, while rates climb to five to thirty percent for specialized domains and legal information hallucination averages 6.4 percent even for top models. 2026
Algorithmic pricing raises margins in concentrated markets: margins increased 28 percent in local duopoly retail gasoline markets in Germany when both firms adopted algorithmic pricing software. 2026
Along the strict long-match equilibrium envelope, stake size should show no positive marginal association with credible capacity, and in exit-shaped settings a larger refundable principal is predicted to shrink credible capacity, not to be inert. 2026
The DAO form is established but not yet an institution: total assets under DAO governance have crossed the hundred-billion-dollar threshold, yet the institutional architecture the form requires remains incomplete. 2026
Across forty DAOs in eight segments the unweighted dataset mean is 67.3 of 130 points (51.8 percent): the median DAO has implemented roughly half of the institutional architecture the framework prescribes. 2026
A visibility paradox holds across every segment: categories producing visible artifacts score consistently above the midpoint while categories producing invisible governance infrastructure score consistently below it. 2026
The AI-governance vacuum is universal: AI Alignment scores 2.10 of 10 across the dataset with no DAO above 5, the only category in the thirteen-dimension framework where no entity crosses the midpoint, and Agent Integration scores only 3.30. 2026
The projected post-upgrade mean of 95.3 of 130 represents a 28-point, 42 percent improvement over the 67.3 baseline available through the convergent five-upgrade agenda. 2026
Segment-level means range from 55.4 for Political and Civic to 76.2 for Data and Analytics, a spread of roughly twenty points or sixteen percent of the maximum total. 2026
Capital-formation institutions score one to two standard deviations higher than AI-governance institutions: Fundraising at 6.88 and Payment System at 6.25 against Agent Integration at 3.30 and AI Alignment at 2.10. 2026
A DAO can be highly successful on visible categories and effectively non-functional on invisible categories simultaneously, as ConstitutionDAO's maximum Fundraising score paired with minimum Governance score shows. 2026
The evaluations measure what DAOs document about themselves rather than what DAOs do in operation, a source-document constraint the analysis does not claim to overcome. 2026
The propositions separating computative from neoclassical labor markets are concrete hypotheses with neoclassical nulls — on allocation efficiency, participation formation under standing-based entry, cooperation stability under defection incentives, and capture resistance — each evaluable in controlled multi-agent settings now. 2026