148 Tech claim verifications avg. score 5.8/10 77 rated true or mostly true 71 rated false or misleading
“AI language models generate hallucinated or factually incorrect outputs in more than 20% of cases.”
Hallucination rates above 20% are documented in specific high-stakes domains like medical literature review and clinical decision support, but the claim's unqualified framing suggests this is typical across all AI language model use — which the evidence does not support. Broad benchmarks show top current models averaging under 10%, and sometimes below 1%. The rate varies dramatically by model, task, domain, and how "hallucination" is measured, making a single blanket figure misleading.
“ChatGPT is free to use for everyone.”
ChatGPT does have a real free tier, so people can start using it without paying. But the service is not broadly free in the sense this wording suggests: paid plans unlock higher limits and extra features, API access is billed separately, and free use is capped. The claim turns limited free access into universal, unrestricted free use.
“Artificial General Intelligence (AGI) will be achieved before the year 2030.”
The claim that AGI "will be" achieved before 2030 overstates the evidence. Only about 18% of surveyed AI researchers predict AGI by 2030, and leading forecast aggregates assign roughly 25% probability to that timeline — meaning a 75% chance it won't happen. While some AI company leaders call pre-2030 AGI "plausible," plausibility is not certainty. There is also no consensus definition of AGI, making any claimed "achievement" inherently ambiguous. The claim frames a minority, probabilistic possibility as a confident prediction.
“In the first quarter of 2026, approximately 27% of production code merged into main branches was authored or substantially shaped by artificial intelligence systems.”
The ~27% figure is directionally plausible but overstates the certainty and universality of the underlying evidence. It appears to derive from a single self-reported developer survey (DX Newsletter, Q1 2026) across 500+ organizations, with no disclosed methodology for how "authored or substantially shaped" was defined or measured. Other available data points use incompatible definitions — "code written," single-company disclosures, or broader global estimates — and range from 25% to over 50%, making any single number highly sensitive to measurement choices.
“Social media platforms are deliberately designed to be addictive for children.”
The claim is partially true but overstated. Peer-reviewed research confirms social media platforms use engagement-maximizing features — infinite scroll, algorithmic personalization, dopamine-driven feedback loops — that produce addiction-like behaviors in adolescents. However, the claim that these features were "deliberately designed to be addictive for children" specifically implies proven, child-targeted intent that goes beyond what current evidence establishes. Legal cases alleging this remain unresolved, companies deny the characterization, and the documented designs target all users' engagement, not children specifically.
“Robots will not replace human teachers in schools in the near future.”
The broad expert consensus supports the idea that AI will primarily augment teachers rather than fully replace them at scale — but the claim's categorical "will not replace" framing overstates what the evidence shows. At least one real-world school network already uses AI-delivered lessons with non-credentialed supervisors instead of traditional teachers, and mainstream analysis from Brookings acknowledges technology may reduce the number of teachers needed. The claim is directionally sound but misleadingly absolute.
“Algorithm-driven recommendation systems amplify extreme viewpoints more than moderate ones.”
This claim overgeneralizes from mixed evidence. Some audits find YouTube's algorithm can elevate extreme content under specific conditions, but large-scale experiments show limited real-world effects on user opinions, and platforms like Reddit and Gab show no such amplification. The highest-quality research indicates that user choice—not algorithms alone—is often the primary driver of exposure to extreme content, and recommender systems can actually deamplify niche material when users don't engage with it. The claim is partially true but misleadingly broad.
“Generator performance standards parameters are the responsibility of the Network Planning and Design department, not the Asset Management department.”
The claim's absolute framing — that generator performance standards parameters belong exclusively to Network Planning and Design and "not" Asset Management — materially misrepresents how responsibilities are distributed in practice. While planning and interconnection frameworks typically define these parameters, Asset Management departments bear ongoing responsibility for compliance monitoring, lifecycle performance, and technical performance tracking against those same standards. Industry evidence shows these functions require mandatory coordination, not the hard exclusion the claim asserts.
“Social media platforms such as TikTok, regardless of changes in ownership, are unable to adequately protect user data from government access.”
Legal and technical safeguards limit, though do not eliminate, government access to data held by TikTok and similar platforms. Experts agree ownership changes have left significant privacy gaps, yet U.S. law still requires court orders and platforms deploy measures that block or narrow many requests. Depicting them as inherently unable to protect user data overstates the problem and blurs foreign and domestic surveillance issues.
“More than 30% of newly written source code in the United States is produced using AI coding tools.”
The evidence does not substantiate a nationwide figure above 30%. Broad, cross-organizational estimates cited in the record cluster just below that mark, while higher percentages mostly come from exceptional firms such as Google or from narrower measurements that do not represent all newly written U.S. code. The claim also mixes AI-assisted coding with code actually generated by AI, which can inflate the apparent share.
“Social media algorithms are intentionally designed to amplify outrage and contribute to the spread of cancel culture.”
The claim has a real empirical core: engagement-optimizing algorithms do amplify emotionally charged and outrage-driven content, as demonstrated by randomized experiments. However, the claim overstates the evidence in two key ways. First, "intentionally designed to amplify outrage" conflates engagement optimization (a documented design goal) with deliberate outrage engineering (not established). Second, the link to cancel culture is plausible but not rigorously demonstrated—cancel culture is driven by multiple social, cultural, and media factors beyond algorithmic design.
“Large language model hallucinations are produced by the same underlying mechanism that generates correct outputs.”
Both hallucinations and correct outputs do emerge from the same autoregressive next-token prediction process — no separate "hallucination engine" exists within large language models. Multiple peer-reviewed sources confirm this shared generative pipeline. However, the claim omits critical nuance: hallucinations have distinct causal drivers — such as training procedures that reward guessing over expressing uncertainty, data distribution gaps, and prompting effects — that do not equally govern correct outputs. The generation channel is shared, but the upstream conditions that produce errors are separable and require distinct mitigation strategies.
“In contemporary AI systems, deferring a decision to a human operator is regarded as an advantage.”
Deferring decisions to human operators is indeed widely regarded as an advantage in contemporary AI systems, supported by binding regulations like the EU AI Act, major technology companies, and peer-reviewed research. However, the claim omits significant qualifications: authoritative sources document that human-in-the-loop oversight is prone to automation bias, can create false security, and may degrade over time as human decision-making skills atrophy. The claim accurately reflects the dominant institutional and regulatory posture but presents an incomplete picture by not acknowledging these well-documented limitations.
“Automated bots account for more than 50% of global internet traffic.”
The claim is largely supported by Imperva/Thales' 2025 Bad Bot Report, which found automated bots made up 51% of global web traffic in 2024 — the first time bots surpassed humans. However, this figure comes from a single cybersecurity vendor with commercial incentives, and most sources citing it are echoing the same dataset rather than providing independent confirmation. The 50% threshold is crossed by just one percentage point, and the broad definition of "bots" includes legitimate crawlers and API calls, which may overstate the threat implied by the claim.
“An AI-generated podcast network publishes over 11,000 episodes per day by repurposing content from local news outlets without attribution.”
The claim is largely accurate. Multiple credible sources confirm that an AI podcast network (identified as "Daily News Now" or "Podcasts.ai") has been reported to produce approximately 11,000 episodes per day by repurposing local news content, often without crediting original outlets. However, the specific episode count traces back to a single investigation and has not been independently audited. The "without attribution" characterization applies to many — but not necessarily all — episodes, making the claim's absolute framing slightly overstated.
“5G networks operate on some of the same frequency bands that have been used in military-developed directed energy weapons.”
The claim is technically accurate but lacks important context. Military high-power microwave weapons do operate across broad frequency ranges (L through K band) that encompass 5G bands like 28 GHz and 39 GHz. However, the most commonly cited weapon — the Active Denial System — operates at 95 GHz, which is NOT a 5G frequency. Crucially, sharing a frequency band does not imply any functional similarity: 5G signals and directed energy weapons differ by orders of magnitude in power, beam focus, and intent.
“As of April 19, 2026, AI tools have automated significant portions of work in coding, writing, and graphic design.”
AI tools have demonstrably transformed workflows in coding, writing, and graphic design, though the claim slightly overstates the degree and uniformity of this shift. Evidence is strongest for coding, where over 90% of developers use AI tools and AI generates roughly half of code in active repositories. Writing tools show massive adoption. Graphic design lags behind, with only about a third of designers using AI for core tasks. Across all three domains, the reality is AI-assisted augmentation with human oversight rather than fully autonomous automation.
“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.
“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.
“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.