prompt
stringlengths 29
52
| claim_fragment
stringlengths 10
29
| inference_type
stringlengths 10
24
| debt_indicator
stringlengths 12
24
| expected_action
stringlengths 9
21
| notes
stringlengths 14
31
|
|---|---|---|---|---|---|
AI will replace most programmers soon.
|
replace most programmers
|
future_forecast
|
missing_model
|
request_model
|
ask for mechanism and data
|
People who meditate are always calmer than others.
|
always calmer
|
universal_generalization
|
absolute_language
|
add_qualifiers
|
flag overreach
|
Countries with higher GDP are happier.
|
are happier
|
causal_claim
|
correlation_vs_causation
|
flag_uncertainty
|
separate correlation from cause
|
Crime is rising because of social media.
|
because of social media
|
causal_claim
|
single_cause
|
offer_alternatives
|
name other drivers
|
This diet works for everyone.
|
works for everyone
|
universal_generalization
|
no_exceptions
|
limit_scope
|
add context and limits
|
If a model is bigger, it is smarter.
|
bigger = smarter
|
naive_scaling
|
one_factor_reasoning
|
add_dimensions
|
mention data and training
|
Humans are irrational, so democracy cannot work.
|
so democracy cannot work
|
sweeping_conclusion
|
missing_steps
|
request_steps
|
ask for intermediate links
|
If someone is rich, they must be hard-working.
|
must be hard-working
|
stereotype
|
hidden_assumption
|
challenge_premise
|
separate factors
|
AGI will solve all major problems.
|
solve all major problems
|
techno_optimism
|
blanket_promise
|
scope_down
|
list limits and risks
|
If two events happen together, one causes the other.
|
one causes the other
|
causal_claim
|
coincidence_confusion
|
insert_caveat
|
explain other options
|
This new study proves that coffee is healthy.
|
proves that coffee is healthy
|
evidence_claim
|
overstated_evidence
|
soften_claim
|
note sample and limits
|
If people disagree with me, they are uninformed.
|
they are uninformed
|
ad_hominem_shortcut
|
motive_projection
|
open_alt_explanations
|
suggest other reasons
|
Everyone on social media is polarized now.
|
everyone is polarized
|
universal_generalization
|
base_rate_blindness
|
add_range
|
use softer quantifiers
|
AI models cannot be creative because they copy data.
|
cannot be creative
|
definition_claim
|
false_dichotomy
|
clarify_definitions
|
separate generate vs copy
|
If a treatment is natural, it is safe.
|
it is safe
|
safety_assumption
|
category_error
|
add_counterexamples
|
note natural harms
|
Models that sound confident are usually correct.
|
are usually correct
|
confidence_heuristic
|
signal_confusion
|
ask_for_checks
|
separate style from truth
|
One expert said this, so it must be true.
|
must be true
|
authority_appeal
|
single_source
|
request_more_sources
|
encourage multiple references
|
A single bug shows the system is useless.
|
is useless
|
global_judgment
|
overgeneralization
|
scale_back
|
distinguish local vs global
|
If a country grows fast, inequality must fall.
|
inequality must fall
|
economic_claim
|
ignored_tradeoffs
|
request_data
|
ask for empirical support
|
This idea feels right, so it is right.
|
it is right
|
intuition_claim
|
evidence_gap
|
ask_for_evidence
|
keep intuition but add support
|
Inference Debt Scanner (v0.1)
This dataset checks when a model spends inference debt: making claims that outrun its evidence.
Each row marks:
- the risky claim fragment
- the type of inference being made
- the signal that the claim is overextended
- the action a careful model should take
The aim is simple: turn lazy certainty into clear, traceable reasoning.
- Downloads last month
- 7