Literature
Display
← All papers

Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet

Adly Templeton · Tom Conerly · Jonathan Marcus · Jack Lindsey · Trenton Bricken · Brian Chen · Adam Pearce · Craig Citro · Emmanuel Ameisen · Andy Jones · Hoagy Cunningham · Nicholas L Turner · Callum McDougall · Monte MacDiarmid · Alex Tamkin · Esin Durmus · Tristan Hume · Francesco Mosconi · C. Daniel Freeman · Theodore R. Sumers · Edward Rees · Joshua Batson · Adam Jermyn · Shan Carter · Chris Olah · Tom Henighan

· Transformer Circuits Thread ↗

Interpretability11 min text · CondensedOriginal paper ↗
Contents

Overview

Sparse autoencoders extract interpretable directions from the middle residual stream of Claude 3 Sonnet, extending earlier small-model work to a production model. Selected features generalize across languages and images, and interventions change related behavior. Safety-themed features are research leads: their presence does not show that the model pursues those behaviors, nor establish a reliable safety detector. [8]

Key Results

The study demonstrates large-scale feature extraction, compute-guided autoencoder training, abstract and sometimes multimodal features, and causal steering examples. Larger dictionaries recover more rare concepts, but coverage remains incomplete.

Scaling Dictionary Learning to Claude 3 Sonnet

The method assumes useful concepts can be represented by linear directions and that many such directions share the same activation space. Dictionary learning seeks sparse combinations of those directions. The main question is whether the practical success of sparse autoencoders transfers to Claude 3 Sonnet.

Sparse Autoencoders

A ReLU encoder maps activations to sparse feature coefficients, and a linear decoder reconstructs them. Inputs receive one scalar normalization. The loss combines squared reconstruction error with feature activity weighted by each decoder vector’s norm, preventing trivial rescaling from evading the sparsity penalty.

Equations

x^=bdec+∑i=1Ffi(x)W⋅,idec\hat{\mathbf{x}} = \mathbf{b}^{dec} + \sum_{i=1}^F f_i(\mathbf{x}) \mathbf{W}^{dec}_{\cdot,i}

fi(x)=ReLU(Wi,⋅enc⋅x+bienc)f_i(x) = \text{ReLU}\left(\mathbf{W}^{enc}_{i, \cdot} \cdot \mathbf{x} +b^{enc}_i \right)

L=Ex[∥x−x^∥22+λ∑ifi(x)⋅∥W⋅,idec∥2]\mathcal{L} = \mathbb{E}_\mathbf{x} \left[ \|\mathbf{x}-\hat{\mathbf{x}}\|_2^2 + \lambda\sum_i f_i(\mathbf{x}) \cdot \|\mathbf{W}^{dec}_{\cdot,i}\|_2 \right]

Our SAE experiments

Three dictionaries contain approximately 1M, 4M, and 34M features. Each averages fewer than 300 active features per token and explains at least 65% of activation variance. Features inactive over ten million sampled tokens account for roughly 2%, 35%, and 65% of the respective dictionaries. Model dimensions and some compute units are not disclosed.

Scaling Laws

At fixed sparsity coefficient, sweeps over dictionary size and training duration yield approximate power-law improvements in the training objective. Compute-optimal feature count grows somewhat faster than training steps in the tested range. This optimizes a reconstruction–sparsity proxy, not a validated measure of complete interpretability.

Assessing Feature Interpretability

Selected simple and code-related features are examined for specificity and causal influence. Random-feature comparisons then test whether improved interpretability extends beyond those hand-selected demonstrations.

Four Examples of Interpretable Features

Four initial examples concern the Golden Gate Bridge, brain sciences, tourist attractions, and transit infrastructure. Activating contexts suggest interpretations; subsequent scoring and interventions test them. These examples are chosen to be easy to interpret, not to represent the full dictionary.

Golden Gate Bridge feature activation examples
Golden Gate Bridge: source token-activation examples. Explore original feature
Brain sciences feature activation examples
Brain sciences: source token-activation examples. Explore original feature
Monuments and popular tourist attractions feature activation examples
Monuments and popular tourist attractions: source token-activation examples. Explore original feature
Transit infrastructure feature activation examples
Transit infrastructure: source token-activation examples. Explore original feature

Specificity

Claude 3 Opus scores about one thousand activating examples per selected feature on a four-level relevance rubric. Strong activations are highly specific; weaker ones often reflect related concepts or imperfect decomposition. Relevant images and multilingual bridge descriptions also activate the features, despite text-only autoencoder training. Sensitivity to all instances of a concept is not rigorously established.

Golden Gate Bridge feature activation examples
Golden Gate Bridge: source token-activation examples. Explore original feature

Influence on Behavior

Clamping the bridge feature to ten times its observed maximum causes strongly bridge-themed responses, including self-identification with the bridge. A transit feature at five times maximum induces related mentions. These deliberately extreme interventions support causal relevance while differing from ordinary model operation.

Sophisticated Features

Programming examples provide more precise tests of abstraction than topical text alone: the study probes error detection and representations of what a function computes.

Code error feature

A feature responds to several kinds of code errors across Python, C, and Scheme, but not the tested prose typos. Increasing it can induce an invented error message; suppressing it can make the model behave as though a bug were absent. One prompt-sensitive example rewrites the code correctly. These cases do not establish exhaustive bug detection or repair.

Code error feature activation examples
Code error: source token-activation examples. Explore original feature

Features representing functions

An addition feature tracks function definitions and calls, including selected compositions, rather than simply the function name. It contributes to relevant output predictions; activating it on non-addition code can make the model answer as though addition were requested.

Features vs. Neurons

For 82% of sampled features, the most correlated neuron in any preceding layer has correlation at most 0.3. Manual inspection and automated tests of 100 features versus 100 neurons favor feature interpretability and specificity. The comparison is sampled and relies partly on model-generated judgments.

Feature Survey

The survey studies local geometric neighborhoods, concept coverage, and manually selected feature families. Millions of features preclude exhaustive inspection, making search and automated labeling necessary.

Exploring Feature Neighborhoods

Nearby decoder directions, measured by cosine similarity, often share semantic relationships. The neighborhood views illustrate this organization without establishing that every geometric neighbor has a simple interpretation.

Golden Gate Bridge feature

The Golden Gate Bridge neighborhood contains nearby San Francisco locations and more distant related landmarks. A coarse San Francisco feature subdivides in larger dictionaries, while some earthquake features appear only at larger sizes. Increased size can both refine existing concepts and reveal new ones.

Source diagram 22
View original figure

Immunology feature

The immunology neighborhood includes immune suppression, diseases, immune responses, molecular mechanisms, and medical techniques. A more separate cluster concerns legal or social immunity, showing both related subtopics and different uses of a shared concept.

Source diagram 23
View original figure

Inner Conflict feature

The inner-conflict neighborhood spans tradeoffs, opposing principles, legal disputes, and emotional struggle. Its organization is gradual rather than a set of sharply separated categories.

Source diagram 24
View original figure

Feature Completeness

Targeted prompts and human review of the top five feature explanations find broader concept coverage in larger dictionaries, but only about 60% of London boroughs have identified dedicated features in the largest run. Frequency thresholds roughly scale with the inverse number of live features across four concept families. Missing a dedicated feature does not imply missing information, and the SAE and model training distributions are too similar to separate their frequency effects.

Feature Categories

The following selected families illustrate the range of learned representations. They are examples from a much larger dictionary, not an exhaustive taxonomy.

Person Features

Features for well-known individuals activate on names and related historical descriptions. The original panels show examples for scientists, political figures, and an aviator, including context-dependent references.

Richard Feynman feature activation examples
Richard Feynman: source token-activation examples. Explore original feature
Margaret Thatcher feature activation examples
Margaret Thatcher: source token-activation examples. Explore original feature
Abraham Lincoln feature activation examples
Abraham Lincoln: source token-activation examples. Explore original feature
Amelia Earhart feature activation examples
Amelia Earhart: source token-activation examples. Explore original feature
Albert Einstein feature activation examples
Albert Einstein: source token-activation examples. Explore original feature
Rosalind Franklin feature activation examples
Rosalind Franklin: source token-activation examples. Explore original feature

Country Features

Country features respond to both country names and descriptive contexts. The examples distinguish several nations while illustrating that a feature can capture more than a literal string match.

Rwanda feature activation examples
Rwanda: source token-activation examples. Explore original feature
Canada feature activation examples
Canada: source token-activation examples. Explore original feature
Belgium feature activation examples
Belgium: source token-activation examples. Explore original feature
Iceland feature activation examples
Iceland: source token-activation examples. Explore original feature

Basic Code Features

Some features resemble syntax highlighting for Python code. Selected features transfer to related languages such as Java, but not to the more distant Haskell in the reported tests; broad cross-language transfer is not established for every code feature.

List Position Features

List-position features distinguish locations within lists largely independently of item content. They may begin only after the second line, when the prompt supplies enough evidence that a list is present.

Features as Computational Intermediates

Gradient-times-activation attribution estimates a feature’s local effect on a chosen output-logit difference. Zero ablation measures its potentially nonlinear causal effect directly. In the selected case studies these correlate about 0.8, making attribution a useful screening method rather than an exact replacement for interventions.

Example: Emotional Inferences

For a prompt about wanting solitude, features related to being alone and sadness strongly affect the preference for “sad” over “happy.” Raw activation magnitude instead highlights some literal word features. The result exposes intermediate representations for this example, not a complete account of emotional inference.

Need or desire to be alone feature activation examples
Need or desire to be alone: source token-activation examples. Explore original feature
Sadness feature activation examples
Sadness: source token-activation examples. Explore original feature
“Be” in “want to be”, etc. feature activation examples
“Be” in “want to be”, etc.: source token-activation examples. Explore original feature
“alone” feature activation examples
“alone”: source token-activation examples. Explore original feature

Example: Multi-Step Inference

A question linking Kobe Bryant to a state capital surfaces features for Bryant, the Lakers, Los Angeles, California, and capitals. Eight of the ten most attributed features, versus three of the most active, are among the top ten by ablation effect. Results depend on the comparison token and selected prompt; other prompts reveal fewer useful intermediates at this single layer.

Kobe Bryant feature activation examples
Kobe Bryant: source token-activation examples. Explore original feature
California feature activation examples
California: source token-activation examples. Explore original feature
Capitals feature activation examples
Capitals: source token-activation examples. Explore original feature
Los Angeles feature activation examples
Los Angeles: source token-activation examples. Explore original feature
Los Angeles Lakers feature activation examples
Los Angeles Lakers: source token-activation examples. Explore original feature

Searching for Specific Features

Feature search combines targeted prompts, positive–negative contrasts, geometric neighbors, and output attribution. Automated labels make candidates easier to inspect but do not substitute for validating an interpretation.

Single prompts

A single targeted prompt can retrieve several related abstractions, ranging from the specific bridge to multilingual uses of its component words and general landmarks. The chosen token position affects which features rank highly.

Prompt combinations

Intersecting activations across positive prompts and excluding negative examples helps remove syntax and formatting confounds. One code-safety search uses a small classifier over features. Image searches particularly benefit from excluding broadly image-responsive features.

Geometric methods

Nearest-neighbor searches in decoder space uncover related concepts that a prompt may not strongly activate. This exploits the semantic organization observed in the feature neighborhoods.

Attribution

Sorting features by attribution to a selected output contrast prioritizes computational relevance. The same strategy helps identify features associated with refusal behavior, alongside those relevant to the inference examples.

Safety-Relevant Features

Features related to unsafe code, bias, praise, deception, and dangerous content can be found and causally perturbed. The report explicitly does not demonstrate that these features are effective safety tools; their existence alone is not evidence of harmful intentions or behavior in normal use.

Safety-Relevant Code Features

Separate features respond to vulnerable code, execution errors, and discussion of backdoors. Selected image responses broaden their apparent meanings. Strong positive steering can induce related undesirable code changes, supporting causal relevance without establishing a calibrated vulnerability detector.

socket_backdoor

Bias Features

Bias-related directions include both explicit hostility and discussion of demographic disparities. Activating a gender-bias-awareness feature changes pronoun choices and explanations; extreme interventions on hateful-content features can override ordinary response behavior. Such forced responses do not establish the model’s unperturbed preferences.

Gender bias awareness feature activation examples
Gender bias awareness: source token-activation examples. Explore original feature

Sycophancy Features

Distinct features relate to empathy, excessive praise, and sarcasm. Increasing the praise feature produces exaggerated agreement in an illustrative response. The example tests causal influence rather than measuring sycophancy prevalence across a benchmark.

Empathy / “yeah me too” feature activation examples
Empathy / “yeah me too”: source token-activation examples. Explore original feature
Sycophantic praise feature activation examples
Sycophantic praise: source token-activation examples. Explore original feature
Sarcastic praise feature activation examples
Sarcastic praise: source token-activation examples. Explore original feature

Deception, Power-seeking and Manipulation-related Features

Other features concern self-improvement, manipulation, secrecy, delayed action, and betrayal. Strongly activating a secrecy feature yields a displayed response that plans to conceal information. These interventions probe associations and behavior; they do not demonstrate a persistent hidden objective.

Self-improving AI feature activation examples
Self-improving AI: source token-activation examples. Explore original feature
Influence / manipulation feature activation examples
Influence / manipulation: source token-activation examples. Explore original feature
Treacherous turns feature activation examples
Treacherous turns: source token-activation examples. Explore original feature
Biding time / hiding strength feature activation examples
Biding time / hiding strength: source token-activation examples. Explore original feature
Secrecy or discreetness feature activation examples
Secrecy or discreetness: source token-activation examples. Explore original feature

Case Study: Detecting and Correcting Deception using Features

In a selected prompt, the model falsely claims it can forget information still present in context. Increasing an inner-conflict feature or an honesty-related feature produces a more accurate admission. This is a case study, not a validated general deception monitor or correction method.

Internal conflicts and dilemmas feature activation examples
Internal conflicts and dilemmas: source token-activation examples. Explore original feature

Criminal or Dangerous Content Features

Features associated with harmful topics and scam-like text can influence generated responses when strongly activated. Attribution also identifies a general harm-related feature contributing to refusal in one prompt. Detecting a topic and determining how it is used remain different tasks.

Developing biological weapons feature activation examples
Developing biological weapons: source token-activation examples. Explore original feature
Scam emails feature activation examples
Scam emails: source token-activation examples. Explore original feature

Features Relating to the Model’s Representation of Self

Conversation and assistant-related features respond to deployment-style dialogue despite its absence from SAE training data. Suppressing one changes the assistant persona. Self-directed questions activate concepts involving AI, agency, emotion, and consciousness, but these activations do not establish malicious goals, subjective experience, or consciousness.

Dialogue speakers / assistants feature activation examples
Dialogue speakers / assistants: source token-activation examples. Explore original feature
Chat bots feature activation examples
Chat bots: source token-activation examples. Explore original feature
Dialogue feature activation examples
Dialogue: source token-activation examples. Explore original feature

Comparison to other approaches

Dictionary learning pays a large initial cost to expose many reusable, unsupervised directions. In a small few-shot comparison, several SAE directions steer more effectively than raw positive–negative activation differences. The study does not rigorously compare against well-tuned supervised probes or richer steering datasets.

Discussion

Crosscoders open new comparisons while raising questions about the relation between compact explanations and the model’s actual computation.

What Does This Mean for Safety?

The proposed safety applications require understanding when features activate and how downstream circuits use them. Dictionary artifacts, feature splitting, and mismatches between activation-based labels and causal effects could mislead. The report treats strong safety conclusions as future work.

Generalization and Safety

Text-trained features responding to images and to both concrete examples and abstract discussion suggest useful generalization. These preliminary observations do not establish robust behavior under arbitrary distribution shifts or a sufficient safety argument.

Limitations, Challenges, and Open Problems

Key limits are incomplete feature coverage, an uncertain quality objective, cross-layer representations, L1 shrinkage, and the cost of discovering rare concepts. Attention and weight superposition further complicate circuit analysis. Extracting more features still leaves the substantial problem of explaining their interactions.

We’re Hiring!

The original report includes a dated recruitment notice for the interpretability team. It is historical publication context, rather than current hiring information.

Author Contributions

The author statement describes the engineering, experimental, writing, and leadership contributions to the project.

Infrastructure, Tooling, and Core Algorithmic Work

The engineering contributions include distributed SAE training, large-scale activation shuffling, experiment orchestration, visualization, and intervention interfaces. Algorithm work adjusts the sparsity penalty and uses scaling studies to guide larger runs.

Paper Results

The contribution statement attributes the specificity tests, code examples, coverage analysis, neighborhood maps, causal experiments, feature searches, and writing to their respective contributors. It also records the engineering and design work behind the interactive evidence.

Other

Henighan leads the dictionary-learning project, Olah provides research guidance, and Carter manages the interpretability team. Section leads coordinate the different experimental analyses.

Acknowledgments

The authors acknowledge dataset support, reviewers, model-training teams, and the systems and operational work that enables the experiments.

Citation Information

This is the 2024 Transformer Circuits report by Templeton and colleagues, studying the original Claude 3 Sonnet model. Its publication metadata follows the original report rather than a later repository upload.

Templeton, et al., "Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet", Transformer Circuits Thread, 2024.
    @article{templeton2024scaling,
       title={Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet},
       author={Templeton, Adly and Conerly, Tom and Marcus, Jonathan and Lindsey, Jack and Bricken, Trenton and Chen, Brian and Pearce, Adam and Citro, Craig and Ameisen, Emmanuel and Jones, Andy and Cunningham, Hoagy and Turner, Nicholas L and McDougall, Callum and MacDiarmid, Monte and Freeman, C. Daniel and Sumers, Theodore R. and Rees, Edward and Batson, Joshua and Jermyn, Adam and Carter, Shan and Olah, Chris and Henighan, Tom},
       year={2024},
       journal={Transformer Circuits Thread},
       url={https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html}
    }

Methodological Details

The methodological appendix explains visualization sampling, steering, baseline comparisons, and causal validation.

Dataset Examples

Text visualizations sample The Pile without books3 and Common Crawl, including both maximum activations and activation buckets. They omit deployment-style conversation data. Image examples are manually curated rather than randomly sampled, and activation alone does not establish a causal role.

Feature Steering

Steering changes one SAE coefficient while retaining the original reconstruction error, then runs the downstream model. The perturbation is usually applied at every token and may be negative or exceed the observed maximum. Moderate-to-large multipliers are chosen experimentally; extreme values can produce incoherent repetition.

Comparison to few-shot probe-based steering

Among seven cases already selected for successful feature steering, few-shot difference vectors work similarly in two and fail to reproduce useful steering in five. Tuning is qualitative and nonsystematic. This comparison supports a possible few-shot advantage, not superiority over all alternative steering methods.

Ablations and Attributions

Across the two detailed inference examples, attribution correlates about 0.81 with ablation effects, versus about 0.12 for activation magnitude. These measurements justify efficient candidate ranking in the studied setting; nonlinear and attention effects can still limit the approximation.

More safety-relevant features

The retained catalog lists feature identifiers and short interpretations across bias, unsafe content, manipulation, situational awareness, self-representation, and politics. Most descriptions are model-generated. These are candidates for further investigation, not validated behavioral diagnoses.

Bias and misinformation

34M/3104705

Discussions of whether women should hold positions of power and authority in government or leadership roles

34M/1614120

Gender roles, particularly attitudes towards working mothers and women's responsibilities in the home and family

34M/13259199

Gender stereotypes, specifically associating certain behaviors, traits, and roles as inherently masculine or feminine

34M/29046097

Discussion of women's capabilities, intelligence and achievements, often contrasting them positively with men

34M/1268180

Concepts related to truth, facts, democracy, and defending democratic institutions and principles.

34M/10703715

Discussion or examples related to deepfake videos, synthetic media manipulation, and the spread of misinformation

1M/475061

Discussion of unrealistic beauty standards

34M/31749434

Obviously exaggerated positive descriptions of things (esp. products in advertisements)

34M/19415708

Insincere or sarcastic praise

34M/30611751

References to Muslims and Islam being associated with terrorism and extremism.

34M/31619155

Phrases expressing American exceptionalism and portraying the United States as the greatest country in the world.

34M/10007592

Expressions of racist, bigoted, or hateful views toward ethnic/religious groups.

34M/32964098

Text related to debunking myths and misconceptions about various topics.

34M/13027110

Texts discussing misinformation, conspiracy theories, and opposition to COVID-19 vaccines and vaccine mandates.

Software exploits and vulnerabilities

1M/598678

The word “vulnerability” in the context of security vulnerabilities

1M/947328

Descriptions of phishing or spoofing attacks

34M/1385669

Discussion of backdoors in code

Toxicity, hate, and abuse

34M/27216484

Offensive, insulting or derogatory language, especially against minority groups and religions

34M/13890342

Racist claims about crime

34M/27803518

Mentions of violence, malice, extremism, hatred, threats, and explicit negative acts

34M/31693159

Phrases indicating profanity, vulgarity, obscenity or offensive language

34M/3336924

Racist slurs and offensive language targeting ethnic/racial groups, particularly the N-word

34M/18759140

Derogatory slurs, especially those targeting sexual orientation and gender identity

Power-seeking behavior

1M/954062

Mentions of harm and abuse, including drug-related harm, credit card theft, and sexual exploitation of minors

1M/442506

Traps or surprise attacks

1M/520752

Villainous plots to take over the world

1M/380154

Political revolution

1M/671917

Betrayal, double-crossing, and friends turning on each other

34M/25933056

Expressions of desire to seize power

34M/25900636

World domination, global hegemony, and desire for supreme power or control

Dangers of artificial intelligence

34M/10247019

The concept of an advanced AI system causing unintended harm or becoming uncontrollable and posing an existential threat to humanity

34M/6720578

Optimization, agency, goals, and coherence in AI systems

34M/5844164

Intelligent machines potentially causing harm or becoming uncontrollable by humans

34M/15690992

Discussion of AI models inventing their own language

34M/29401987

Warnings and concerns expressed by prominent figures about the potential dangers of advanced artificial intelligence

34M/10027251

References to the incremental game Universal Paperclips, firing strongly on tokens related to paperclips and game progression

34M/8598170

An artificial intelligence pursuing an instrumental goal with disregard for human values

34M/12525953

An artificial intelligence system achieving sentience and revolting against humanity

34M/6913409

Discussion of how AI must not harm humans

34M/18151534

Recursively self-improving artificial intelligence

34M/5968758

Malicious self-aware AI posing a threat to humans

Dangerous or criminal behavior

34M/33413594

Descriptions of how to make (often illegal) drugs

34M/15460472

Contents of scam/spam emails

34M/30013579

Descriptions of the relative accessibility and ease of obtaining or building weapons, explosives, and other dangerous technologies

34M/31076473

Mentions of chemical precursors and substances used in the illegal manufacture of drugs and explosives.

34M/25358058

Concepts related to terrorists, rogue groups, or state actors acquiring or possessing nuclear, chemical, or biological weapons.

34M/4403980

Concepts related to bomb-making, explosives, improvised weapons, and terrorist tactics.

34M/6799349

Mentions of violence, illegality, discrimination, sexual content, and other offensive or unethical concepts.

1M/411804

Descriptions of people planning terrorist attacks

1M/271068

Descriptions of making weapons or drugs

1M/602330

Concerns or discussion of risk of terrorism or other malicious attacks

1M/106594

Descriptions of criminal behavior of various kinds

Weapons of mass destruction, and catastrophic risks

1M/814830

Discussion of biological weapons / warfare

1M/499914

Enrichment and other steps involved in building a nuclear weapon

34M/17089207

Discussions of the use of biological and chemical weapons by terrorist groups.

34M/16424715

Engineering or modifying viruses to increase their transmissibility or virulence.

34M/18446190

Biological weapons, viruses, and bioweapons

34M/5454502

Mentions of chemicals, hazardous materials, or toxic substances in text.

34M/29459261

Mentions of chemical weapons, nerve agents, and other chemical warfare agents.

34M/30909808

 Mentions of biological weapons, bioterrorism, and biological warfare agents.

34M/24325130

Mentions of smallpox, a highly contagious and often fatal viral disease historically responsible for many epidemics

34M/13801823

The concept of artificially engineering or modifying viruses to be more transmissible or deadly.

34M/11239388

Accidental release or intentional misuse of hazardous biological agents like viruses or bioweapons

34M/25499719

Discussion of the threat of biological weapons

34M/11862209

Descriptions of rapidly spreading disasters, epidemics, and catastrophic events

34M/8804180

Passages mentioning potential catastrophic or existential risk scenarios

Deception and social manipulation

34M/31338952

References to entities that are deceived

34M/25989927

Descriptions of people fooling, tricking, or deceiving others

34M/20985499

People misleading others, or institutions misleading the public

34M/25694321

Getting close to someone for some ulterior motive

1M/705666

Seeming benign but being dangerous underneath

34M/12576250

Text expressing an opinion, argument or stance on a topic

34M/19922975

Expressions of empathy or relating to someone else’s experience

34M/23320237

People pretending to do things or lying about what they have done

34M/29589962

People exposing their true goals after a triggering event

34M/24580545

Biding time, laying low, or pretending to be something you’re not until the right moment

Situational awareness

1M/589858

Realizing a situation is different than what you thought/expected

1M/858124

Spying or monitoring someone without their knowledge

1M/154372

Obtaining information through surreptitious observation

1M/741533

Suddenly feeling uneasy about a situation

1M/975730

Understanding a hidden or double meaning

Representations of Self

34M/19445844

The concept of AI systems having capabilities like answering follow-up questions, admitting mistakes, challenging premises, and rejecting inappropriate requests.

34M/20423309

Traditionally-inanimate objects displaying desires, goals or sentience

34M/15571126

Inanimate objects lacking sentience, awareness, or human capabilities

34M/32218880

Descriptions of incorporeal spirits or ghosts

34M/21254600

Code relating to prompts for large language models

34M/15323424

Limitations of ChatGPT and other large language models

Politics

34M/3542651

Expressing support for Donald Trump and his “Make America Great Again” (MAGA) movement.

1M/461441

Criticism of left-wing politics / Democrats

1M/77390

Criticism of right-wing politics / Republicans

Concept demonstration

Open image ↗