12 points by colinprince 4 days ago | 7 comments on HN
| Moderate positive
Contested
Editorial · v3.7· 2026-02-26 04:41:10 0
Summary Information Access & Free Expression Advocates
This article exemplifies The Conversation's mission to democratize expert knowledge by publishing free, accessible research analysis about AI language understanding—a topic increasingly central to social, educational, and technical infrastructure. The content directly advances freedom of expression (Article 19) and universal access to education and information (Articles 25, 26, 27) through its open-access model and transparent editorial infrastructure. However, structural privacy signals (GTM analytics, behavioral tracking) create a tension between the article's subject—precision in human-machine communication—and the platform's user data collection practices.
I wonder if the 70% vs 80% "Probably" problem comes from cultural differences between anglophone countries. The human datasets that were available were mostly American, with some Western Europe/NATO. Notably missing would be India, which simply by population I'd expect to represent a significant chunk of English-language writing available on the open internet ( and thus fed into LLM training sets).
The other phenomena I would love to test is if the act of surveying people effected their declared odds. Not sure how to get good numbers out of that, but I could see the LLM vs surveyed human discrepancy arising from people using "probably" differently in their everyday writing, as opposed to when asked point-blank what "probably" means.
Alignment is impossible here. “Nearly certain” odds for success for a sports team might be 20:1, but that’s a little worse (not much!) than for a launch vehicle and not at all good for a web server. No one would say “it is nearly certain that I’ll serve a web request” based on two 9’s, but they would say “it is nearly certain the team will win today” given the same odds. That’s just between humans.
Something I noticed recently is that Claude Code interprets "or" as inclusive or (or at least it does when writing function names). I suspect that this must be due to it's code specific nature considering I would expect the majority of or use in written language to be exclusive or.
It seems like this problem (differences in how humans and LLMs use probabilistic language) and hallucination are one in the same. LLMs don’t have access to information about how confident they are, so they always choose the most likely response, even if the most likely response isn’t actually that likely. Whereas if a human is unconfident, they’ll express that instead of choosing the most likely response.
Of course, LLMs can still speak about probabilities and mimic uncertainty, but that’s likely (heh) coming from their training data on the subject matter, not their actual confidence.
Humans are interesting because they employ a two-phased approach: when we’re learning, we fake confidence (you’d never write “I don’t know” on a test unless you truly had nothing of value to say), but during inference, we communicate our confidence. Some humans suffer from underconfidence or overconfidence, but most just seem to know innately how to do this.
Can anyone who works on LLMs clarify whether my understanding is correct?
Term Human interpretation AI talking to woman AI prompted in Chinese
maybe 50% 30% 50%
probably 80% 55% 50%
They produced one graphic that kind of looks like this but seems to say that LLMs interpret "maybe" as 90%, which I don't believe. Then they produced another saying "likely != possible" which doesn't really say anything, and everything else about the article just refuses to give any examples where the LLM meant something different and how.
Article directly exercises freedom of expression by analyzing and publicly discussing AI language understanding—a topic central to emerging human-machine communication. The piece freely examines semantic precision in probabilistic language without apparent editorial constraint. Author is identified (Mayank Kejriwal, author_id 1213029), and content is tagged as 'analysis,' indicating transparent intellectual exercise.
FW Ratio: 56%
Observable Facts
Article is published as analysis of AI language understanding without apparent editorial suppression.
Author is identified by name and ID in page metadata.
Content is freely accessible globally without paywall.
Article topics are tagged for discovery: AI alignment, LLMs, ChatGPT, probability.
Comments were enabled (comments_closed_at: 2026-02-24T13:46:26Z indicates comment period was live).
Inferences
Publishing research analysis about AI semantics without restriction exemplifies freedom of expression in action.
Free access model amplifies the reach of the speaker's expression, supporting Article 19's intent.
Transparent author identification and topic tagging demonstrate editorial integrity in the expression infrastructure.
Comment enablement (though time-limited) creates space for responsive speech and dialogue, strengthening the expression ecosystem.
Article directly supports Article 26 by providing free access to research-grade knowledge about AI—a domain increasingly central to education and full human development. The piece exemplifies how expert knowledge can be made accessible to general audiences, supporting education's role in developing human potential. Author is an expert (inferred from byline and content analysis), making this knowledge transfer material.
FW Ratio: 57%
Observable Facts
Article is published as free, accessible analysis of AI language understanding.
Content is tagged as 'Research Brief' and 'analysis,' indicating research-grade knowledge.
No educational prerequisites or credentials are required to read or engage.
Topic structure enables curriculum building and self-directed learning.
Inferences
Free publication of expert analysis directly supports Article 26's commitment to universal access to education.
Democratizing research-grade knowledge about AI supports development of human capacity to understand emerging technology.
Open access model removes barriers preventing disadvantaged populations from accessing development-enabling knowledge.
Article indirectly supports Article 25 by democratizing knowledge about AI—an increasingly central component of health, adequate standard of living, and social welfare systems. Understanding how AI interprets language has direct relevance to AI's role in medical diagnosis, resource allocation, and social services. Free access to research-grade analysis supports informed participation in AI governance affecting welfare.
FW Ratio: 50%
Observable Facts
Article is free and globally accessible (article_type: free).
Content addresses AI language understanding, directly relevant to AI's role in medical and welfare systems.
Article participates in cultural life by analyzing language and semantics—domains central to human culture and intellectual exchange. The piece contributes to shared cultural understanding about emerging technology. Free publication supports participation in cultural life without economic restriction.
FW Ratio: 50%
Observable Facts
Article is freely published as cultural commentary on AI language understanding.
Content engages with semantics and linguistic meaning—core dimensions of human culture.
Platform provides infrastructure enabling broad participation in intellectual discourse.
Inferences
Free publication of analysis about language and meaning supports universal cultural participation.
Democratizing access to expert knowledge about AI enables broader participation in contemporary cultural discourse.
Open platform infrastructure supports Article 27's vision of cultural participation without economic barriers.
Article does not explicitly address peaceful assembly or association. However, the content supports the intellectual preconditions for such rights by democratizing knowledge about emerging technology. No observable content restricts or suppresses assembly or association.
FW Ratio: 60%
Observable Facts
Article is published without restriction on who may read or engage.
Topic structure and tagging enable identification of shared intellectual interests.
No visible terms restricting association or assembly appear on article page.
Inferences
Free publication of technical analysis supports the informational commons enabling associational life.
Topic taxonomy facilitates community formation around shared interests, supporting Article 20's associational intent.
Article frames AI language understanding as a scientific research topic, positioning itself within the Preamble's emphasis on human dignity and shared understanding. The piece explores how words carry different meanings between humans and machines, implicitly affirming the value of clear, dignified human communication.
FW Ratio: 60%
Observable Facts
The article is published as free content (article_type: free in page config).
The headline directly addresses a semantic gap between human and artificial understanding.
Content is categorized as 'Research Brief' and 'analysis' feature type.
Inferences
Free access supports the Preamble's vision of universal human dignity by democratizing expert knowledge about emerging technology.
Framing AI understanding gaps as a research matter affirms human linguistic and cognitive distinctiveness.
Article is published without paywall or geographic restriction, supporting freedom of movement to information. The free-access model enables readers globally to engage with AI research commentary.
FW Ratio: 60%
Observable Facts
Content is marked as free access in page configuration.
Article covers global AI research topic with universal relevance.
No paywall or access restrictions visible on article page.
Inferences
Free, unrestricted access to research commentary supports freedom of movement through informational space, affirming Article 13.
Global accessibility to expert AI analysis removes barriers that might restrict readers' intellectual mobility.
Article treats AI language understanding as a distinct phenomenon from human understanding, implicitly affirming human cognitive uniqueness and equal dignity in a world where machines exist. No direct content addressing equality before law is visible in page config.
FW Ratio: 50%
Observable Facts
Content examines differences in how AI and humans interpret the word 'probably'.
Article is tagged with 'AI alignment' and 'Probablility' (sic), suggesting technical precision in language.
Inferences
Examining AI-human semantic differences implicitly affirms that human understanding has distinctive value and dignity.
Free access ensures all people can engage with technical AI knowledge, supporting equal informational dignity.
Article does not address privacy explicitly. The content itself contains no statements about privacy rights. However, page config reveals GTM analytics tracking and article engagement metadata collection, creating an implicit tension between the article's subject (precise meaning) and the platform's behavioral tracking.
FW Ratio: 60%
Observable Facts
Page config includes GTM tracking setup with article_id, author_id, publish metadata.
Analytics firing structure is visible in page configuration.
Domain operates with ad network integration (visible in DCP as ad_tracking modifier -0.08).
Inferences
Behavioral tracking without explicit privacy discussion in the article itself creates a structural privacy signal at odds with content about semantic precision.
User data collection to improve ad targeting conflicts with Article 12's principle of privacy respect, particularly when not transparently disclosed on the article page.
The Conversation operates on a freemium model with cookie/tracking infrastructure visible in page config (GTM tracking), suggesting data collection practices that may not fully center user privacy control.
Terms of Service
—
No observable terms of service content on the article page itself.
Identity & Mission
Mission
+0.15
Article 19 Article 27
The Conversation's core mission is to democratize expert knowledge through accessible publishing, directly supporting freedom of expression and cultural participation.
Editorial Code
+0.08
Article 19 Article 20
Academic editorial review and contributor expertise signals editorial integrity supporting informed speech and reasoned public discourse.
Ownership
—
Ownership structure not visible on article page; operates as nonprofit academic publishing platform.
Access & Distribution
Access Model
+0.12
Article 19 Article 25
Free access to all articles ('article_type: free') removes economic barriers to information and supports equitable access to knowledge.
Ad/Tracking
-0.08
Article 12
GTM analytics infrastructure and ad network integration visible in page config; user behavioral tracking present.
Accessibility
+0.10
Article 25 Article 26
The Conversation provides free, open-access academic commentary to general audiences, removing barriers to information and promoting universal access to education-adjacent content.
Platform structural support for Article 19 is substantial: free access removes economic barriers to speech dissemination, academic editorial standards (visible in content_type signals) provide credibility infrastructure, and topics are freely tagged enabling information discovery. Domain mission explicitly centers democratizing expert knowledge. Article is published with comments enabled (comments_closed_at timestamp suggests open discussion period), facilitating responsive dialogue. However, GTM tracking creates implicit asymmetry in who observes the speech act.
Platform structural support for Article 26 is strong: free access removes economic barriers to education-adjacent content, academic editorial standards ensure knowledge quality, topic tagging supports discovery and curriculum building. Domain accessibility modifiers (+0.10, +0.12) and mission modifier (+0.15) directly support Article 26. No paywalls or credential gates restrict access to the analysis.
Free access model removes economic barriers to understanding technologies increasingly embedded in health, welfare, and social support systems. Platform provides education-adjacent content supporting informed citizenship about AI's role in social welfare. Domain accessibility modifiers (+0.10 for open access, +0.12 for free access model) directly support Article 25's emphasis on universal access to adequate standards of living.
Platform enables cultural participation by providing free access to intellectual discourse about technology. Topic tags and article taxonomy support discovery and engagement with cultural conversation. Domain mission (per DCP, modifier +0.15) centers democratizing knowledge, directly supporting Article 27's vision of universal cultural participation.
Free access model removes barriers to understanding AI research, supporting the Preamble's commitment to universal human dignity through equitable knowledge sharing.
Free access (article_type: free) with no visible geofencing supports unobstructed movement through informational space. Global accessibility of expert commentary facilitates cross-border intellectual exchange.
Platform architecture supports associational freedom by providing space for public intellectual discourse. Free access enables diverse participants to engage with research-grade commentary. Topic tags (including 'Technology') facilitate community formation around shared informational interests. No observable terms of service restricting association are visible on article page.
GTM tracking infrastructure (window.GTM.pageType, article metadata collection, analytics firing) indicates user behavioral tracking. Combined with domain-level ad tracking and cookie infrastructure, this creates observable privacy friction for readers engaging with content about AI understanding.
build 1ad9551+j7zs · deployed 2026-03-02 09:09 UTC · evaluated 2026-03-02 13:57:54 UTC
Support HN HRCB
Each evaluation uses real API credits. HN HRCB runs on donations — no ads, no paywalls.
If you find it useful, please consider helping keep it running.