181 Tech claim verifications avg. score 5.9/10 99 rated true or mostly true 63 rated false or mostly false
“Sam Altman said that Loopt had 100 times more users than 500 and said he would provide evidence but did not provide evidence.”
The core account is supported: Sam Altman publicly rejected the reported 500-user figure for Loopt, indicated it was far too low, and said he would provide data. The strongest caveat is that the precise “100 times” wording is better supported as a paraphrase from secondary reporting than as a clearly documented direct quote. No public evidence release matching that promise is evident in the record provided.
“Loopt had only 500 users near the end of its operation.”
The evidence does not support the statement that Loopt had only 500 users near the end. Reliable contemporaneous reporting around the sale put Loopt at about 3 million registered users, while the “500” figure came from an unverified report about daily active users and was explicitly disputed by the company. The claim omits that crucial distinction and therefore states something materially wrong.
“The device marketed as "PowerPro Genius" reduces a typical household's electricity bill by about 50%.”
The evidence does not support this product cutting a typical household electricity bill by about 50%. Independent technical explanations and scam warnings indicate these plug-in devices generally do not reduce the kilowatt-hours homes are billed for, so any savings would be negligible rather than dramatic. The 50% figure also exceeds even the product's own marketing claims.
“Restricting rival countries' access to advanced computer chips slows those countries' progress in artificial intelligence capabilities.”
The evidence supports a real slowing effect, especially for frontier AI systems that depend on large amounts of advanced compute. Export controls have imposed meaningful delays and scaling constraints, most clearly in China. But the effect is not universal or complete: software efficiency, alternative supply channels, and narrower definitions of “AI progress” can reduce or mask the slowdown.
“Building an advanced semiconductor chip industry requires years of research and billions of dollars in investment.”
The evidence strongly supports this statement. Advanced semiconductor capability routinely requires multi-year research, facility buildout, process development, and production ramp-up, while leading-edge fabs and supporting ecosystems cost billions to tens of billions of dollars. The wording is broad but not overstated for an advanced chip industry.
“Only a handful of companies worldwide can design leading-edge artificial-intelligence chips, including Nvidia and Advanced Micro Devices.”
The evidence supports the main point that leading-edge AI-chip design is concentrated in a small global group, and Nvidia and AMD are clearly part of it. But the wording is somewhat too tight: beyond merchant GPU vendors, companies such as Google, Amazon, Intel, Huawei, and others also design advanced AI accelerators. The claim is directionally right, but imprecise about how small the club is.
“Automatic Dependent Surveillance–Broadcast (ADS-B) has become a widely used data source for large-scale aviation situational awareness because it provides cooperative aircraft position, velocity, and status information for traffic surveillance and air-traffic management.”
The claim is well supported by authoritative aviation sources. FAA, ICAO, and EUROCONTROL describe ADS-B as a cooperative system that broadcasts aircraft position, velocity, and related status or identification data for surveillance and air-traffic management, and they show it is integrated into major operational networks. The only notable caveat is that “widely used” is not quantified.
“Ethanol-blended gasoline is harmful for vehicles used in India.”
The evidence does not show that ethanol-blended petrol broadly harms vehicles in India. Authoritative sources indicate E20 is generally safe for E20-compatible vehicles and has not been shown to cause widespread engine damage, though some older or non-compatible vehicles can see modest mileage loss and may need certain parts checked or replaced. The claim contains a narrow truth for a subset of vehicles but overstates it into a general rule.
“For public-road and canyon riding, the Ducati Streetfighter V2 S outperforms the Ducati Streetfighter V4 S on objective measures of rideability, including low-to-midrange usability, controllability, and rider comfort.”
The overall evidence supports the idea that the Streetfighter V2 S is the better road-and-canyon bike for accessible performance, easier control, and day-to-day comfort. Its lower weight, friendlier power delivery, and road-focused setup make that conclusion plausible and widely repeated in independent reviews. However, the claim overstates the evidence by calling these advantages "objective measures," because most support is qualitative rather than instrumented.
“Startups that sell claim verification via an API generally use a single-pass, single-model pipeline.”
The evidence points the other way. The cited examples mostly describe multi-stage claim-verification systems that retrieve evidence, evaluate it, and then issue a judgment, often using multiple components or models. No credible market-level evidence shows that startups selling verification APIs usually rely on a single-pass, single-model design.
“Startups that sell claim verification via an API generally do not offer multi-model adversarial adjudication.”
Available evidence indicates that most cited claim-verification APIs use single-model or linear workflows, while multi-model adversarial adjudication appears mainly in research systems. That supports the claim’s basic direction. However, the market evidence is limited, and some products do compare outputs from multiple models without implementing full adversarial adjudication.
“Startups that sell claim verification via an API generally do not offer a full audit trail or grounded follow-up questioning to interrogate the verdict.”
The claim overreaches the available evidence. Research and commentary do suggest that robust auditability and grounded interactive questioning are not standard strengths of automated fact-checking systems, but the cited sources do not show that API-selling startups generally lack them. Because the market is not systematically surveyed and the key features are undefined, the statement is too broad as written.
“Hackers have distributed malware through Steam Workshop items intended for the Wallpaper Engine app on Steam.”
The evidence shows that attackers did use Steam Workshop items for Wallpaper Engine to distribute malware. Security researchers documented malicious “application wallpapers” carrying credential-stealing and remote-access payloads, and Wallpaper Engine’s developers later confirmed the abuse and tightened restrictions. The important caveat is that this was not every wallpaper type, but a specific executable-capable category.
“Adopting generative AI tools increases employee productivity in companies by at least 10%.”
The evidence does not support a broad claim that adopting generative AI reliably increases employee productivity by at least 10% across companies. Strong studies do show gains above 10% in specific tasks such as customer support, writing, and some coding workflows, especially for less-experienced workers. But effects vary widely by role and implementation, some studies find no gain or even slower performance, and firm-wide productivity improvements are not established at that threshold.
“Artificial intelligence is projected to contribute $15.7 trillion to the global economy by 2030, including about $6.6 trillion from productivity gains and $9.1 trillion from consumption-side effects, representing a 14% increase in global GDP versus a scenario without AI.”
The figures are real and accurately reflect PwC’s widely cited 2017 AI macroeconomic projection: up to $15.7 trillion by 2030, with gains split between productivity and consumption effects. But this is a scenario-based estimate, not a consensus forecast, and later analyses emphasize uncertainty, adoption assumptions, and uneven distribution of benefits.
“An air intake scoop captures high-pressure, cool outside air and funnels it directly into an engine, boosting combustion efficiency and increasing horsepower at higher vehicle speeds.”
The basic mechanism is real, but the wording overstates how it works and how much it helps. A forward-facing scoop can deliver cooler outside air and, at higher speeds, create a small ram-air pressure gain that may increase power. In most street vehicles, the horsepower benefit is modest and highly dependent on scoop design, sealing, and speed, and it does not necessarily improve combustion efficiency.
“Lenz.io is the only tool or platform that provides audit-grade fact-checking for AI products.”
The exclusivity claim is not supported by the evidence. Multiple tools and platforms already offer overlapping fact-checking, verification, grounding, citation, and governance capabilities for AI systems, so describing Lenz.io as the only option is inaccurate. The key term “audit-grade” is also undefined, and no cited primary evidence shows that Lenz.io uniquely meets a clear standard that others do not.
“Last-click attribution models can undercount assisted conversions from social media.”
The evidence shows that last-click attribution can understate social media’s role when social appears earlier in the conversion path. In these models, the final touchpoint receives all credit, so prior social interactions often get none in standard attribution reports. Some analytics tools report assists separately, but that does not change the basic limitation of last-click reporting.
“Google Analytics Multi-Channel Funnels reports Assisted Conversions separately from last-click conversion reports.”
Google’s documentation shows that Multi-Channel Funnels distinguishes Assisted Conversions from Last Click or Direct Conversions instead of folding them into ordinary last-click reporting. In Universal Analytics, assisted metrics are surfaced in the MCF reporting suite and explicitly contrasted with last-click metrics. The main caveat is that this terminology belongs to UA, not GA4’s current reporting model.
“A voice coil motor (VCM) used for smartphone autofocus has no gears and no friction.”
Smartphone autofocus VCMs are generally gearless, but they are not frictionless. Stronger technical sources describe direct electromagnetic motion alongside real mechanical suspensions, guides, springs, or contact surfaces where friction and wear must be managed. The claim overstates a true feature of VCMs and turns it into an inaccurate absolute.