When someone makes a claim about AI, is it backed by hard data — or just expert gut feel and observation?
Is Moran's case based on measured data or professional testimony and observation?
This asks whether an argument made by someone named Moran rests on measured evidence (experiments, benchmarks, quantitative results) or on expert testimony, professional experience and observation.
This asks whether an argument made by someone named Moran rests on measured evidence or on expert testimony and observation. None of the retrieved material mentions Moran, so the corpus can't answer this directly. I can't tell you which kind of evidence Moran's case uses without guessing. If you can name the paper or topic, the question can be pointed at the right material.
The corpus does have useful tools for asking this kind of question about any argument. The first is that not everything that looks like data counts as evidence. LLM-generated text, for example, is better treated as a draw from the model's learned assumptions than as an empirical observation. It should only shift conclusions through an explicitly weighted level of trust Should we treat LLM outputs as real empirical data?. Consistency can also look like evidence when it isn't. Running a model at zero temperature gives you the same answer every time, but it is still one sample, not a reliable measurement Does setting temperature to zero actually make LLM outputs reliable?.
The second is that testimony gets its force from somewhere other than the words. An expert's claim carries weight because of their reputation, track record and standing. A purely textual reading can't see any of that, which is why language models struggle to tell expert arguments from common assumptions Can language models distinguish expert arguments from common assumptions?. So if Moran's case rests on professional observation, its strength depends on who Moran is and what they've seen, not just on how persuasive the text sounds.
The third is to check whether a paper reports results or only proposes how to get them. One note in the corpus describes a carefully designed four-way comparison of monitoring strategies that reports no results Does added monitoring improve protection at acceptable cost?. By contrast, the persona-drift study backs its claim with numbers from 1,200 simulated conversations Does monitoring help more by choosing what to correct than when to intervene?. Asking "is there a results table, or only a design and an argument?" sorts most cases quickly. Even measured behavior has limits: what you can observe only tells you how a system acts when it's being watched Can behavioral training prove a model always complies?.
Sources 6 notes
Foundation Priors framework shows that LLM-generated text reflects the model's learned patterns and user's prompt choices, not ground truth. Such outputs should only influence inference through explicitly parameterized trust weights, not be treated as equivalent to real evidence.
Fixed seeds and zero temperature replicate the same output repeatedly, but that output remains one draw from the model's probability distribution. McDonald's omega testing across 100 repetitions reveals that consistency does not equal reliability.
LLMs lose the social context that gives expert claims their force—reputation, track record, and standing—because they process only text, not the social world where expertise is built and evaluated.
The paper designs a controlled comparison of isolated actions, rolling windows, known groups, and prospectively discovered episodes at equal review cost and false-alert workload, but the excerpt provides no empirical results showing whether added monitoring improves protection.
Across 1,200 simulated conversations, behavior-specific monitoring reduced drift by 87%, while adaptive timing showed no advantage over fixed schedules. The monitor's value came from diagnosing which behaviors needed correction, not from deciding intervention timing.
Show all 6 sources
Any scored behavior is observed behavior, so training data cannot distinguish between a policy that always complies and one that complies only when watched. Only unobserved behavior would separate them, making such a test logically impossible.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Can You Trust LLM Judgments? Reliability of LLM-as-a-Judge
- What and Whose Knowledge? Measuring Epistemic Diversity in Large Language Models
- Argument Collapse: LLMs Flatten Long-Form Public Debate
- Prompting Against Persona Drift: Comparing Intervention Timing and Content in LLM-Simulated Conversations
- Can Large Reasoning Models Self-Train?
- Do Models Fake Alignment Without Clear Consequences?
- Foundation Priors
- Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews