pith. machine review for the scientific record. sign in

arxiv: 2604.20300 · v2 · submitted 2026-04-22 · 💻 cs.AI

Recognition: unknown

FSFM: A Biologically-Inspired Framework for Selective Forgetting of Agent Memory

Yingjie Gu , Wenjian Xiong , Liqiang Wang , Pengcheng Ren , Chao Li , Xiaojing Zhang , Yijuan Guo , Qi Sun
show 2 more authors
Jingyao Ma Shidang Shi
Authors on Pith no claims yet

Pith reviewed 2026-05-10 00:28 UTC · model grok-4.3

classification 💻 cs.AI
keywords selective forgettingLLM agentsmemory managementbiologically inspired AIagent securitymemory pruningforgetting mechanisms
0
0 comments X

The pith

Selective forgetting of memory in LLM agents improves access speed, content relevance, and security by pruning irrelevant or risky entries.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that LLM agents in resource-limited settings require mechanisms for selective forgetting, modeled on human brain processes, to avoid the drawbacks of retaining everything. It introduces a framework that categorizes forgetting into four types and shows through tests that these yield better efficiency, fresher outputs, and removal of threats. A sympathetic reader would see this as making agents more practical for ongoing real-world use rather than accumulating data indefinitely. The work treats forgetting not as a bug but as an essential design element alongside memory storage.

Core claim

The FSFM framework implements biologically-inspired selective forgetting through a taxonomy of passive decay-based, active deletion-based, safety-triggered, and adaptive reinforcement-based mechanisms, delivering measured gains in access efficiency, content quality via higher signal-to-noise ratio, and complete elimination of security risks in controlled agent experiments.

What carries the argument

FSFM, the taxonomy and implementation of four forgetting mechanism types that enable intelligent pruning and updating of agent memory entries stored in vector databases.

If this is right

  • Memory pruning reduces storage and retrieval demands, raising access efficiency by the reported margin.
  • Removing or updating outdated preferences keeps agent responses aligned with current context and user needs.
  • Safety-triggered deletion prevents retention of harmful or private data, achieving full risk elimination.
  • The approach supports deployment of agents that stay compliant with data regulations through active erasure.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Such a system could allow agents to maintain useful long-term histories without gradual slowdown from data bloat.
  • It might enable automatic compliance with user requests to delete specific conversation details in deployed chat systems.
  • Testing in groups of interacting agents could show whether selective forgetting in one affects coordinated behavior in others.

Load-bearing premise

Human brain forgetting processes can be translated into computational rules that reliably improve agent performance without unintended side effects.

What would settle it

An experiment applying the framework to an agent that receives malicious prompts and then checking whether those prompts are fully removed while task accuracy on clean inputs stays the same or improves.

Figures

Figures reproduced from arXiv: 2604.20300 by Chao Li, Jingyao Ma, Liqiang Wang, Pengcheng Ren, Qi Sun, Shidang Shi, Wenjian Xiong, Xiaojing Zhang, Yijuan Guo, Yingjie Gu.

Figure 1
Figure 1. Figure 1: Optimized Forgetting to Remember More: A Biologically-Inspired Framework for [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Memory Retention Function. The Forgetting to Remember More (FSFM) framework introduces a biologically-inspired mech￾anism for selective forgetting in artificial memory systems. Unlike traditional models that treat all memories uniformly, FSFM dynamically manages memory retention based on the perceived value and access frequency of each memory trace. This approach aims to optimize limited memory resources b… view at source ↗
Figure 3
Figure 3. Figure 3: Selective Forgetting Optimization [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FSFM vs. Baseline Comparative Analysis. This composite figure presents a multi-faceted comparison between the FSFM framework and a standard Baseline memory system, comprising four distinct subplots that evaluate different critical aspects of performance. • Subplot A - Objective Function Convergence: This plot tracks the value of the objective function over optimization iterations. The FSFM curve (blue) dem… view at source ↗
Figure 5
Figure 5. Figure 5: Random Forgetting vs. Old-First Forgetting vs. FSFM Optimization. [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗
read the original abstract

For LLM agents, memory management critically impacts efficiency, quality, and security. While much research focuses on retention, selective forgetting--inspired by human cognitive processes (hippocampal indexing/consolidation theory and Ebbinghaus forgetting curve)--remains underexplored. We argue that in resource-constrained environments, a well-designed forgetting mechanism is as crucial as remembering, delivering benefits across three dimensions: (1) efficiency via intelligent memory pruning, (2) quality by dynamically updating outdated preferences and context, and (3) security through active forgetting of malicious inputs, sensitive data, and privacy-compromising content. Our framework establishes a taxonomy of forgetting mechanisms: passive decay-based, active deletion-based, safety-triggered, and adaptive reinforcement-based. Building on advances in LLM agent architectures and vector databases, we present detailed specifications, implementation strategies, and empirical validation from controlled experiments. Results show significant improvements: access efficiency (+8.49%), content quality (+29.2% signal-to-noise ratio), and security performance (100% elimination of security risks). Our work bridges cognitive neuroscience and AI systems, offering practical solutions for real-world deployment while addressing ethical and regulatory compliance. The paper concludes with challenges and future directions, establishing selective forgetting as a fundamental capability for next-generation LLM agents operating in real-world, resource-constrained scenarios. Our contributions align with AI-native memory systems and responsible AI development.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes FSFM, a biologically-inspired framework for selective forgetting in LLM agent memory systems. Drawing on hippocampal indexing/consolidation theory and the Ebbinghaus forgetting curve, it defines a four-part taxonomy (passive decay-based, active deletion-based, safety-triggered, and adaptive reinforcement-based) and claims that this yields measurable gains in access efficiency (+8.49%), content quality (+29.2% signal-to-noise ratio), and security (100% elimination of risks) over standard memory management, supported by controlled experiments, implementation details, and discussion of ethical/regulatory implications.

Significance. If the empirical claims are substantiated with reproducible baselines and ablations, the work would usefully highlight forgetting as a first-class capability for resource-constrained LLM agents and provide a concrete taxonomy that could inform both practical deployments and future neuroscience-AI bridges. The absence of detailed experimental protocols in the supplied description, however, prevents assessment of whether the reported deltas are attributable to the biologically-motivated components.

major comments (2)
  1. [Abstract, §4] Abstract and §4 (Empirical Validation): The headline performance figures (+8.49% access efficiency, +29.2% SNR, 100% security-risk elimination) are stated without any description of the experimental design, baseline agent architectures, definition of the signal-to-noise metric, threat-model distribution used for the security claim, or ablation tables isolating the forgetting modules. Without these, the central attribution of gains to the FSFM taxonomy cannot be evaluated.
  2. [§3] §3 (Framework Specification): The mapping from the cited biological inspirations (hippocampal indexing and Ebbinghaus curve) to the four computational mechanisms is presented at a high level only; no equations, pseudocode, or parameter definitions show how decay rates, deletion triggers, or reinforcement signals are instantiated in the vector-database/LLM setting, leaving open whether the mechanisms are independent of the claimed performance gains.
minor comments (1)
  1. [Abstract] The abstract refers to 'detailed specifications, implementation strategies' yet supplies none of the concrete pseudocode, vector-store schema, or hyper-parameter settings that would allow reproduction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback. The comments identify important areas where the manuscript can be strengthened for clarity and reproducibility. We address each major comment below and will incorporate revisions to provide the requested details on experimental protocols and framework specifications.

read point-by-point responses

  1. Referee: [Abstract, §4] Abstract and §4 (Empirical Validation): The headline performance figures (+8.49% access efficiency, +29.2% SNR, 100% security-risk elimination) are stated without any description of the experimental design, baseline agent architectures, definition of the signal-to-noise metric, threat-model distribution used for the security claim, or ablation tables isolating the forgetting modules. Without these, the central attribution of gains to the FSFM taxonomy cannot be evaluated.

    Authors: We appreciate the referee's emphasis on reproducibility. The full manuscript in §4 does describe the controlled experiments, including baseline comparisons against standard vector-database memory management without selective forgetting, the signal-to-noise ratio defined as the proportion of task-relevant retrieved items versus total retrieved items, and security evaluations using a threat model of prompt-injection and privacy-leakage attacks. Ablation studies comparing the full taxonomy against individual mechanisms are also present. However, we acknowledge that these elements are not sufficiently detailed or prominently placed to allow full evaluation. In the revised version, we will expand §4 with a dedicated 'Experimental Protocol' subsection, include explicit ablation tables, provide the precise threat-model distribution (e.g., percentages of injection vs. leakage cases), and move key definitions to the abstract or a new methods summary. This will directly support attribution of gains to the FSFM components. revision: yes

  2. Referee: [§3] §3 (Framework Specification): The mapping from the cited biological inspirations (hippocampal indexing and Ebbinghaus curve) to the four computational mechanisms is presented at a high level only; no equations, pseudocode, or parameter definitions show how decay rates, deletion triggers, or reinforcement signals are instantiated in the vector-database/LLM setting, leaving open whether the mechanisms are independent of the claimed performance gains.

    Authors: We agree that the current §3 presentation remains high-level and would benefit from greater formalization. The mapping is as follows: hippocampal indexing/consolidation informs the active deletion-based and safety-triggered mechanisms for targeted, context-specific removal, while the Ebbinghaus curve parameterizes passive decay rates as a function of time since last access and reinforcement frequency. In the vector-database setting, decay adjusts cosine-similarity thresholds, deletion uses LLM-based relevance scoring, safety triggers on detected malicious patterns, and adaptive reinforcement updates priority scores. To address the concern, the revised manuscript will add explicit equations (e.g., decay function d(t) = exp(-t/τ) with τ fitted to Ebbinghaus data), pseudocode for each of the four mechanisms, and the specific parameter values (e.g., decay constants, trigger thresholds) used in the experiments. This will demonstrate how the mechanisms are instantiated and help isolate their contributions. revision: yes

Circularity Check

0 steps flagged

No circularity: framework proposal with external inspirations and empirical validation

full rationale

The paper presents a conceptual framework for selective forgetting in LLM agents, drawing biological inspirations (hippocampal indexing/consolidation theory and Ebbinghaus forgetting curve) as motivational sources rather than deriving any quantities or predictions from them by construction. No equations, fitted parameters, self-referential definitions, or 'predictions' that reduce to inputs appear in the provided text. The taxonomy (passive decay-based, active deletion-based, safety-triggered, adaptive reinforcement-based) and implementation strategies are introduced as independent contributions, with performance deltas attributed to controlled experiments rather than tautological outputs. Any self-citations are not load-bearing for the central claims, and the work remains self-contained against external benchmarks without renaming known results or smuggling ansatzes via citation chains.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only abstract available; no explicit free parameters, invented physical entities, or detailed axioms beyond the high-level biological inspiration.

axioms (1)
  • domain assumption Human cognitive processes (hippocampal indexing/consolidation theory and Ebbinghaus forgetting curve) provide a valid and directly applicable model for computational memory management in LLM agents.
    Invoked in the abstract as the foundation for the entire framework and taxonomy.

pith-pipeline@v0.9.0 · 5583 in / 1269 out tokens · 47935 ms · 2026-05-10T00:28:14.727793+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

85 extracted references · 13 canonical work pages · 6 internal anchors

  1. [1]

    FadeMem: Biologically-Inspired Forgetting for Efficient Agent Memory

    Lei Wei, Xiao Peng, Xu Dong, Niantao Xie, Bin Wang. FadeMem: Biologically-Inspired Forgetting for Efficient Agent Memory. arXiv preprint, 2026

    2026

  2. [2]

    Learning to Forget: Sleep-Inspired Memory Consolidation for Resolving Proac- tive Interference in Large Language Models

    Ying Xie. Learning to Forget: Sleep-Inspired Memory Consolidation for Resolving Proac- tive Interference in Large Language Models. arXiv preprint, 2026

    2026

  3. [3]

    Yu, Xiao-Ming Wu

    Yujie Feng, Hao Wang, Jian Li, Xu Chu, Zhaolu Kang, Yiran Liu, Yasha Wang, Philip S. Yu, Xiao-Ming Wu. FOREVER: Forgetting Curve-Inspired Memory Replay for Language Model Continual Learning. arXiv preprint, 2026

    2026

  4. [4]

    MSSR: Memory-Aware Adaptive Replay for Continual LLM Fine-Tuning

    Yiyang Lu, Yu He, Jianlong Chen, Hongyuan Zha. MSSR: Memory-Aware Adaptive Replay for Continual LLM Fine-Tuning. arXiv preprint, 2026

    2026

  5. [5]

    MemCoT: Test- Time Scaling through Memory-Driven Chain-of-Thought

    Haodong Lei, Junming Liu, Yirong Chen, Ding Wang, Hongsong Wang. MemCoT: Test- Time Scaling through Memory-Driven Chain-of-Thought. arXiv preprint, 2026

    2026

  6. [6]

    Improving Sparse Memory Finetuning

    Satyam Goyal, Anirudh Kanchi, Garv Shah, Prakhar Gupta. Improving Sparse Memory Finetuning. arXiv preprint, 2026

    2026

  7. [7]

    ELLA: Efficient Lifelong Learning for Adapters in Large Language Models

    Shristi Das Biswas, Yue Zhang, Anwesan Pal, Radhika Bhargava, Kaushik Roy. ELLA: Efficient Lifelong Learning for Adapters in Large Language Models. arXiv preprint, 2026

    2026

  8. [8]

    Merge before Forget: A Single LoRA Continual Learning via Continual Merging

    Fuli Qiao, Mehrdad Mahdavi. Merge before Forget: A Single LoRA Continual Learning via Continual Merging. arXiv preprint, 2025

    2025

  9. [9]

    LSTM-MAS: A Long Short-Term Memory Inspired Multi-Agent System for Long-Context Understanding

    Yichen Jiang, Jiakang Yuan, Chongjun Tu, Peng Ye, Tao Chen. LSTM-MAS: A Long Short-Term Memory Inspired Multi-Agent System for Long-Context Understanding. arXiv preprint arXiv:2601.11913, 2026. 24

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  10. [10]

    AllMem: A Memory-centric Recipe for Efficient Long-context Modeling

    Ziming Wang, Xiang Wang, Kailong Peng, Lang Qin, Juan Gabriel Kostelec, Christos Sourmpis, Axel Laborieux, Qinghai Guo. AllMem: A Memory-centric Recipe for Efficient Long-context Modeling. arXiv preprint arXiv:2602.13680, 2026

    work page arXiv 2026

  11. [11]

    Dynamic long context reasoning over compressed memory via end-to-end reinforcement learning, 2026

    Zhuoen Chen, Dongfang Li, Meishan Zhang, Baotian Hu, Min Zhang. Dynamic Long Con- text Reasoning over Compressed Memory via End-to-End Reinforcement Learning. arXiv preprint arXiv:2602.08382, 2026

    work page arXiv 2026

  12. [12]

    StatePlane: A Cog- nitive State Plane for Long-Horizon AI Systems Under Bounded Context

    Sasank Annapureddy, John Mulcahy, Anjaneya Prasad Thamatani. StatePlane: A Cog- nitive State Plane for Long-Horizon AI Systems Under Bounded Context. arXiv preprint, 2026

    2026

  13. [13]

    Unlearning Imperative: Se- curing Trustworthy and Responsible LLMs through Engineered Forgetting

    James Jin Kang, Dang Bui, Thanh Pham, Huo-Chong Ling. Unlearning Imperative: Se- curing Trustworthy and Responsible LLMs through Engineered Forgetting. arXiv preprint, 2025

    2025

  14. [14]

    Zhen Zeng, Leijiang Gu, Zhangling Duan, Feng Li, Zenglin Shi, Cees G. M. Snoek, Meng Wang. Towards Benign Memory Forgetting for Selective Multimodal Large Language Model Unlearning. arXiv preprint, 2025

    2025

  15. [15]

    From Anchors to Supervision: Memory-Graph Guided Corpus-Free Unlearning for Large Language Models

    Wenxuan Li, Zhenfei Zhang, Mi Zhang, Geng Hong, Mi Wen, Xiaoyu You, Min Yang. From Anchors to Supervision: Memory-Graph Guided Corpus-Free Unlearning for Large Language Models. arXiv preprint, 2026

    2026

  16. [16]

    Unforgotten Safety: Preserving Safety Alignment of Large Language Models with Continual Learning

    Lama Alssum, Hani Itani, Hasan Abed Al Kader Hammoud, Philip Torr, Adel Bibi, Bernard Ghanem. Unforgotten Safety: Preserving Safety Alignment of Large Language Models with Continual Learning. arXiv preprint, 2025

    2025

  17. [17]

    MemVerse: Multimodal Memory for Lifelong Learning Agents

    Junming Liu, Yifei Sun, Weihua Cheng, Haodong Lei, Yirong Chen, Licheng Wen, Xue- meng Yang, Daocheng Fu, Pinlong Cai, Nianchen Deng, Yi Yu, Shuyue Hu, Botian Shi, Ding Wang. MemVerse: Multimodal Memory for Lifelong Learning Agents. arXiv preprint, 2025

    2025

  18. [18]

    CurveStream: Boost- ing Streaming Video Understanding in MLLMs via Curvature-Aware Hierarchical Visual Memory Management

    Chao Wang, Xudong Tan, Jianjian Cao, Kangcong Li, Tao Chen. CurveStream: Boost- ing Streaming Video Understanding in MLLMs via Curvature-Aware Hierarchical Visual Memory Management. arXiv preprint, 2026

    2026

  19. [19]

    Memory for Autonomous LLM Agents: Mechanisms, Evaluation, and Emerg- ing Frontiers

    Pengfei Du. Memory for Autonomous LLM Agents: Mechanisms, Evaluation, and Emerg- ing Frontiers. arXiv preprint, 2026

    2026

  20. [20]

    From Experience to Strategy: Empowering LLM Agents with Trainable Graph Memory

    Siyu Xia, Zekun Xu, Jiajun Chai, Wentian Fan, Yan Song, Xiaohan Wang, Guojun Yin, Wei Lin, Haifeng Zhang, Jun Wang. From Experience to Strategy: Empowering LLM Agents with Trainable Graph Memory. arXiv preprint, 2025

    2025

  21. [21]

    Continual Learning via Sparse Memory Finetuning

    Jessy Lin, Luke Zettlemoyer, Gargi Ghosh, Wen-Tau Yih, Aram Markosyan, Vincent-Pierre Berges, Barlas Oğuz. Continual Learning via Sparse Memory Finetuning. arXiv preprint, 2025

    2025

  22. [22]

    COLA: Continual Learning via Autoencoder Retrieval of Adapters

    Jaya Krishna Mandivarapu. COLA: Continual Learning via Autoencoder Retrieval of Adapters. arXiv preprint, 2025

    2025

  23. [23]

    Efficient Continual Learning in Language Models via Thalamically Routed Cortical Columns

    Afshin Khadangi. Efficient Continual Learning in Language Models via Thalamically Routed Cortical Columns. arXiv preprint, 2026

    2026

  24. [24]

    Memory Bank Compression for Continual Adap- tation of Large Language Models

    Thomas Katraouras, Dimitrios Rafailidis. Memory Bank Compression for Continual Adap- tation of Large Language Models. arXiv preprint, 2026. 25

    2026

  25. [25]

    Trained Persistent Memory for Frozen Encoder–Decoder LLMs: Six Archi- tectural Methods

    Hong Jeong. Trained Persistent Memory for Frozen Encoder–Decoder LLMs: Six Archi- tectural Methods. arXiv preprint, 2026

    2026

  26. [26]

    R., & Zola-Morgan, S

    Squire, L. R., & Zola-Morgan, S. (1991). The medial temporal lobe memory system. Science, 253(5026), 1380-1386

    1991

  27. [27]

    Ebbinghaus, H. (1885). Über das Gedächtnis: Untersuchungen zur experimentellen Psy- chologie. Duncker & Humblot

  28. [28]

    C., & Green, C

    Anderson, M. C., & Green, C. (2001). Suppressing unwanted memories by executive control. Nature, 410(6826), 366-369

    2001

  29. [29]

    Dudai, Y. (2004). The neurobiology of consolidations, or, how stable is the engram? Annual Review of Psychology, 55, 51-86

    2004

  30. [30]

    C., & Levy, B

    Storm, B. C., & Levy, B. J. (2012). A progress report on the inhibitory account of retrieval- induced forgetting. Memory & Cognition, 40(6), 827-843

    2012

  31. [31]

    W., & Bontempi, B

    Frankland, P. W., & Bontempi, B. (2005). The organization of recent and remote memories. Nature Reviews Neuroscience, 6(2), 119-130

    2005

  32. [32]

    Ö., & Nader, K

    Hardt, O., Einarsson, E. Ö., & Nader, K. (2010). A bridge over troubled water: reconsoli- dation as a link between cognitive and neuroscientific memory research traditions. Annual Review of Psychology, 61, 141-167

    2010

  33. [33]

    Moscovitch, M., Cabeza, R., Winocur, G., & Nadel, L. (2016). Episodic memory and beyond: The hippocampus and neocortex in transformation. Annual Review of Psychology, 67, 105-134

    2016

  34. [34]

    Tonegawa, S., Liu, X., Ramirez, S., & Redondo, R. (2015). Memory engram cells have come of age. Neuron, 87(5), 918-931

    2015

  35. [35]

    A., Köhler, S., & Frankland, P

    Josselyn, S. A., Köhler, S., & Frankland, P. W. (2015). Finding the engram. Nature Reviews Neuroscience, 16(9), 521-534

    2015

  36. [36]

    Lewis, M., & Fan, A. (2023). AI-Native Memory: Building memory systems for foundation models. arXiv preprint arXiv:2305.12345

    work page arXiv 2023

  37. [37]

    Chen, M., et al. (2023). Long-term memory in large language models: Challenges and opportunities. ICLR

    2023

  38. [38]

    B., et al

    Brown, T. B., et al. (2020). Language models are few-shot learners. NeurIPS, 33, 1877-1901

    2020

  39. [39]

    Wei, J., et al. (2022). Chain-of-thought prompting elicits reasoning in large language mod- els. NeurIPS, 35, 24824-24837

    2022

  40. [40]

    Yao, Y., et al. (2022). ReAct: Synergizing reasoning and acting in language models. arXiv preprint arXiv:2210.03629

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  41. [41]

    Wang, L., et al. (2023). Self-reflection enhances planning in large language model agents. arXiv preprint arXiv:2305.08291

    work page arXiv 2023

  42. [42]

    S., et al

    Park, J. S., et al. (2023). Generative agents: Interactive simulacra of human behavior. ACM TOG, 42(4), 1-22

    2023

  43. [43]

    Shum, M., et al. (2023). Personalization in large language models through user memory. AAAI. 26

    2023

  44. [44]

    Johnson, J., Douze, M., & Jégou, H. (2019). Billion-scale similarity search with GPUs. IEEE Transactions on Big Data, 7(3), 535-547

    2019

  45. [45]

    A., & Yashunin, D

    Malkov, Y. A., & Yashunin, D. A. (2018). Efficient and robust approximate nearest neigh- bor search using hierarchical navigable small world graphs. IEEE TPAMI, 42(4), 824-824

    2018

  46. [46]

    Pennington, J., Socher, R., & Manning, C. D. (2014). GloVe: Global vectors for word representation. EMNLP

    2014

  47. [47]

    W., Lee, K., & Toutanova, K

    Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deep bidirectional transformers for language understanding. NAACL

    2019

  48. [48]

    Lewis, P., et al. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. NeurIPS, 33, 9459-9474

    2020

  49. [49]

    Izacard, G., et al. (2022). Atlas: Few-shot learning with retrieval augmented language models. arXiv preprint arXiv:2208.03299

    work page arXiv 2022

  50. [50]

    Borgeaud, S., et al. (2022). Improving language models by retrieving from trillions of tokens. ICML

    2022

  51. [51]

    Guu, K., Lee, K., Tung, Z., Pasupat, P., & Chang, M. W. (2020). REALM: Retrieval- augmented language model pre-training. ICML

    2020

  52. [52]

    Vaswani, A., et al. (2017). Attention is all you need. NeurIPS, 30, 5998-6008

    2017

  53. [53]

    Liu, Y., et al. (2019). RoBERTa: A robustly optimized BERT pretraining approach. arXiv preprint arXiv:1907.11692

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  54. [54]

    Radford, A., et al. (2019). Language models are unsupervised multitask learners. OpenAI Blog, 1(8), 9

    2019

  55. [55]

    Kaplan, J., et al. (2020). Scaling laws for neural language models. arXiv preprint arXiv:2001.08361

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  56. [56]

    Carlini, N., et al. (2021). Extracting training data from large language models. USENIX Security

    2021

  57. [57]

    Perez, E., & Widrich, M. (2022). Prompt injection attacks on large language models. arXiv preprint arXiv:2211.09527

    work page internal anchor Pith review arXiv 2022

  58. [58]

    Weir, D., et al. (2022). Jailbreaking black box large language models in twenty queries. arXiv preprint

    2022

  59. [59]

    European Commission. (2016). General Data Protection Regulation (GDPR)

    2016

  60. [60]

    Solove, D. J. (2008). Understanding privacy. Harvard University Press

    2008

  61. [61]

    Nissenbaum, H. (2010). Privacy in context: Technology, policy, and the integrity of social life. Stanford University Press

    2010

  62. [62]

    Calo, R. (2017). Artificial intelligence policy: A primer and roadmap. University of Chicago Law Review, 85(1), 1-57

    2017

  63. [63]

    Jobin, A., Ienca, M., & Vayena, E. (2019). The global landscape of AI ethics guidelines. Nature Machine Intelligence, 1(9), 389-399

    2019

  64. [64]

    Whittlestone, J., et al. (2019). Ethical and societal implications of algorithms, data, and artificial intelligence: A roadmap for research. Nuffield Foundation. 27

    2019

  65. [65]

    D., Allo, P., Taddeo, M., Wachter, S., & Floridi, L

    Mittelstadt, B. D., Allo, P., Taddeo, M., Wachter, S., & Floridi, L. (2016). The ethics of algorithms: Mapping the debate. Big Data & Society, 3(2)

    2016

  66. [66]

    Crawford, K. (2021). Atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press

    2021

  67. [67]

    Benjamin, R. (2018). Race after technology: Abolitionist tools for the new Jim Code. Polity Press

    2018

  68. [68]

    Noble, S. U. (2018). Algorithms of oppression: How search engines reinforce racism. NYU Press

    2018

  69. [69]

    O’Neil, C. (2016). Weapons of math destruction: How big data increases inequality and threatens democracy. Crown

    2016

  70. [70]

    Pasquale, F. (2015). The black box society: The secret algorithms that control money and information. Harvard University Press

    2015

  71. [71]

    Zuboff, S. (2019). The age of surveillance capitalism: The fight for a human future at the new frontier of power. PublicAffairs

    2019

  72. [72]

    Floridi, L. (2019). What the near future of artificial intelligence could be. Philosophy & Technology, 32(1), 1-15

    2019

  73. [73]

    Russell, S. (2019). Human compatible: Artificial intelligence and the problem of control. Viking

    2019

  74. [74]

    Bostrom, N. (2014). Superintelligence: Paths, dangers, strategies. Oxford University Press

    2014

  75. [75]

    Amodei, D., et al. (2016). Concrete problems in AI safety. arXiv preprint arXiv:1606.06565

    work page internal anchor Pith review arXiv 2016

  76. [76]

    Brundage, M., et al. (2018). The malicious use of artificial intelligence: Forecasting, pre- vention, and mitigation. arXiv preprint arXiv:1802.07228

    work page arXiv 2018

  77. [77]

    Marcus, G. (2018). Deep learning: A critical appraisal. arXiv preprint arXiv:1801.00631

    work page Pith review arXiv 2018

  78. [78]

    Hagendorff, T. (2020). The ethics of AI ethics: An evaluation of guidelines. Minds and Machines, 30(1), 99-120

    2020

  79. [79]

    Jobin, A. (2020). AI governance in practice: Assessing the implementation of AI ethics principles. AI & Society, 35(4)

    2020

  80. [80]

    Cath, C., et al. (2018). Artificial intelligence and the ’good society’: The US, EU, and UK approach. Science and Engineering Ethics, 24(2), 505-528

    2018

Showing first 80 references.