Pith. sign in

REVIEW 4 major objections 7 minor 66 references

Symbiotic Agents: A Novel Paradigm for Trustworthy AGI-driven Networks

T0 review · 4 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read Symbiotic agents pair LLMs with optimizers to cut network-control decision errors fivefold.

desk verdict Type I is a genuine empirical contribution worth reading; Type II's confidence-interval story is substantially circular and needs rework before the trustworthiness claims can stand. read the letter →

arxiv 2507.17695 v2 pith:FWLO2XGL submitted 2025-07-23 cs.AI cs.NI

classification cs.AIcs.NI
keywords symbioticagentslargelanguagemodelssmalltrustworthyAIRANslicingSLAnegotiationconfidenceintervalguard-railAGI-drivennetworks
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that language models alone cannot be trusted for numerically precise, real-time network control, and it proposes a remedy it calls symbiotic agents: an LLM or small language model paired with a deterministic optimizer that handles the arithmetic while the model handles reasoning. It claims this pairing cuts decision errors fivefold relative to standalone LLM agents in two network tasks, RAN slicing and multi-tenant SLA negotiation, on a real 5G testbed with channel fluctuations from moving vehicles. It also claims that small models of 3 to 8 billion parameters match large ones in accuracy while using 99.9% less GPU memory and running in 82 ms loops, making edge deployment practical. If correct, the work offers a path to AGI-driven networks where flexibility comes from language models and guarantees come from optimizers.

What carries the argument

The machinery is the optimizer wrapped around the language model. On the input side, Oin pre-processes the prompt with a bounded uncertainty estimate: for Type II, a gradient-descent side-car runs R = 100 jittered restarts on a strictly concave utility model and returns a 95% confidence interval x* ± 1.96 s / sqrt(R) for the Pareto-optimal SLA, which is injected into every agent's prompt as a numeric guard-rail. On the output side, Oout is a proportional controller, a linear feedback rule that adjusts the resource allocation in proportion to the gap between the intent and the current state, with the LLM asynchronously retuning the gain Kp when a convergence KPI exceeds a threshold. The two-optimizer loop is what converts stochastic text generation into numerically certified actions.

What would settle it

Run the Type II negotiation with a ground-truth Pareto frontier obtained by exhaustive grid search over SLA values for many random tenant intents, and count how often the paper's 95% confidence interval, computed from jittered gradient-descent restarts, contains the true optimum; if the coverage is below 95%, the interval is not a genuine confidence bound and the guard-rail is not providing the claimed trustworthiness.

Watch

Extended reading notes

Core claim

The central discovery is that attaching a tiny optimizer to a language model converts the model's semantic flexibility into numerically trustworthy network decisions. In the Type I design, the LLM acts as a meta-optimizer that periodically retunes the proportional gain Kp of a P-controller allocating Physical Resource Blocks to enforce a throughput intent, so the sub-millisecond control loop stays deterministic while the LLM adapts to channel variability. In the Type II design, a gradient-descent side-car computes a 95% confidence interval for the Pareto-optimal SLA from jittered restarts and injects it into every agent's prompt as a numeric guard-rail, requiring bids to stay inside the interval unless justified. On the testbed this yields a fivefold error reduction over standalone LLM agents, negotiation error below 1.3 Mbps, an 82 ms near-real-time loop for a small model with 99.9% less GPU footprint than a large model, and a 44% reduction in RAN over-utilization in the collaborative demonstration.

Load-bearing premise

The central claim stands or falls on whether the hand-built utility model used to compute the negotiation confidence interval actually captures the true Pareto-optimal SLA; if the weights or the jitter scheme are wrong, the 95% confidence guard-rail bounds nothing real and would steer agents toward a confidently wrong target.

Editorial extensions

If this is right

  • If the central claim holds, standalone LLM agents should not be used for real-time RAN resource allocation; the symbiotic pairing becomes the natural architecture for such loops.
  • Small models of 3 to 8 billion parameters become viable substitutes for large ones in near-real-time network control, opening the door to edge GPUs with roughly 2 GB footprints.
  • The confidence-interval guard-rail improves not only numeric accuracy but also the alignment and fairness scores of negotiation dialogues, across model families and sizes.
  • The architecture enables automatic SLA renegotiation during channel degradation, cutting RAN over-utilization by roughly 44% compared with static SLA enforcement.
  • Because next-token sampling stays stochastic, even improved future LLMs will still need external optimizers to deliver deterministic numeric bounds.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • One could test the guard-rail's statistical claim directly by computing the same interval with different utility weights or a different optimizer and checking whether the true Pareto-optimal SLA falls inside it at the claimed 95% rate; the paper's weights are chosen by hand and the same optimizer defines the evaluation ground truth, so the bound is only as good as the utility model.
  • The Type I pattern generalizes beyond RAN: any domain where an LLM tunes a small set of hyperparameters of a fast, stable inner controller, such as video bitrate adaptation or power management, could inherit the same error reduction and latency guarantees.
  • The 82 ms near-real-time loop depends on a deliberately short memory window; as prompts grow with richer context or larger memory, small-model latency will climb, so prompt compression or cache reuse would be the next practical step.
  • The trustworthiness claim covers numeric decision accuracy and bounded outputs; adversarial robustness of the guard-rail itself, for example a tenant prompting the LLM to ignore the interval, is not evaluated and would be a natural red-team test.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 7 minor

Summary. The paper introduces 'symbiotic agents,' an agentic architecture that pairs LLMs/SLMs with deterministic optimizers to make network control and SLA negotiation more trustworthy. Two agent types are designed: Type I agents let an LLM act as a meta-optimizer that tunes the proportional gain of a P-controller for real-time RAN slicing, and Type II agents inject a confidence interval produced by a gradient-descent optimizer into the prompts of negotiating LLM agents to bound SLA bids. The authors evaluate on a 5G testbed with OpenAirInterface and FlexRIC, using channel traces from 78 moving vehicles, and report up to a 5x reduction in decision error, 82 ms near-real-time loops for SLMs, 99.9% GPU memory savings, and a 44% reduction in RAN over-utilization in an end-to-end AGI-RAN demonstration. The paper also proposes a Next-G Open/AI-RAN architecture and releases a demo video and (upon acceptance) partial open-source code.

Significance. If the central claims hold, the symbiotic-agent paradigm is a concrete and practically useful step toward deploying LLM-based decision-making in near-real-time network control, because it externalizes numeric guarantees to deterministic optimizers while preserving the semantic flexibility of LLMs. The Type I experiment is a strength: the 20 Mbps operator intent is external to the LLM and P-controller, the 78-vehicle trace is concrete, and the RMSE comparison against a hand-tuned P-controller is internally consistent. The paper also provides a useful multi-model benchmark showing that small open-weight SLMs can match large proprietary models, which supports edge deployment. The proposed architecture and live demo are valuable for the community. However, the Type II trustworthiness argument is weakened by a statistically questionable confidence interval and by an evaluation that measures MAE against the same optimizer that produces the injected guard-rail, so the claimed error reduction for SLA negotiation is partly by construction.

major comments (4)
  1. [Sec. 3.3, Eq. (8)] The '95% confidence interval' is computed from R jittered restarts of a deterministic, strictly concave optimization problem (Eqs. (6)-(7)). Because the objective is strictly concave with a unique maximizer, perturbing only the initial point x(0) cannot produce a distribution that reflects statistical uncertainty about the true Pareto-optimal SLA; any spread s in Eq. (8) mostly reflects incomplete convergence or the arbitrary jitter scale. The interval x* ± 1.96 s/sqrt(R) is a confidence interval for the mean of the restart outcomes, not a bound on the true optimum. The paper's statement that this interval 'serves as a statistical estimate of the true optimal SLA' is therefore unjustified and should be revised or replaced with a proper uncertainty quantification (e.g., bootstrapping over data or utility parameters, or a prediction interval).
  2. [Sec. 5.2.3, Table 7] The MAE ground truth is the same optimizer's consensus value that produces the confidence interval injected into every LLM prompt, and the guard-rail instruction in Sec. 3.3.1 forces bids into [L,U]. Consequently, the large MAE reduction for symbiotic agents relative to standalone LLMs is substantially by construction: a model that obeys the guard-rail will have low MAE against the optimizer's own answer. This does not demonstrate that the LLM improves numerical accuracy; it demonstrates compliance with the injected interval. Please provide a control condition in which the LLM receives the interval but is not constrained by it, or evaluate against an independent held-out objective, and then report the marginal contribution of the LLM beyond the guard-rail.
  3. [Sec. 3.3.2 and Eq. (6)] The solution of Eq. (6) is a weighted sum of individual quadratic utilities and a mediator utility with hand-set weights alpha_i, gamma, beta and an unspecified lambda. This is a particular scalarization, not generally the Pareto-optimal SLA of the multi-agent game described in the text. The paper repeatedly refers to 'Pareto-optimal' (e.g., Figures 5, 6, 8 and Sec. 3.3). Please either compute an actual Pareto front (e.g., by varying lambda) or soften the language to 'the optimizer's preferred SLA under the chosen weights.' Also specify the lambda value used in the experiments and the distribution of the jitter in Sec. 3.3.1, which is currently not described.
  4. [Abstract and Sec. 8 (Conclusion)] The claim of 'fivefold' error reduction relative to standalone LLMs is not consistently supported by the tables. In Table 4, gpt-4o standalone RMSE is 12.8 Mbps vs. symbiotic 4.5 Mbps (~2.8x), while mistral-7b shows ~4.7x; in Table 7, gpt-4o shows a much larger reduction. The 'up to 5 times' phrasing in the abstract is technically defensible only if the ratio is computed per model and the best case is reported. Please report the exact per-model ratios and qualify the headline claim accordingly.
minor comments (7)
  1. [Sec. 3.2.1 and Appendix B] Section 3.2.1 states that the LLM freely chooses K_new_p in (0, inf), but Listing 3 in Appendix B instructs the LLM to choose Kp between 0.5 and 1.5 with granularity 0.1. These conflicting descriptions should be reconciled, and the impact of the hard bounds on the claimed adaptivity should be discussed.
  2. [Fig. 4 caption] The caption reads 'Zoom on P-Cotrol' and the text mentions 'base line'; these typos should be corrected.
  3. [Tables 4 and 7] The column headers 'RMSE↑' and 'MAE↑' use an up arrow, which conventionally indicates that higher is better, but for these error metrics lower is better. The arrows should be reversed or removed.
  4. [Tables 4 and 7] The VRAM values for gpt-4o (about 3500 GB) are cited in the same units as local models, but gpt-4o is accessed via API and its exact runtime VRAM is not directly comparable. Please clarify whether these figures are model parameter sizes in GB, estimated deployment footprints, or something else.
  5. [Sec. 5.2.1] In Table 5, the last column header 'Score↓' is odd because higher scores are better; please rename to 'Score' and explain that higher is better in the text.
  6. [Sec. 1 vs. Sec. 2] The introduction says 'We are the first to formalize an agent architecture' while the related work states that Agoran [36] is 'the first work to formally utilize and scale the symbiotic paradigm.' These claims should be aligned to avoid an apparent contradiction.
  7. [Sec. 9 (Data availability)] The open-source repository is promised only upon acceptance; during review, reproducibility would be improved by providing an anonymized or partial version of the simulation framework and agent code, even if the full testbed code cannot be released.

Circularity Check

2 steps flagged · score 6.0 of 10

Type II '95% CI' and MAE ground truth both come from the same deterministic optimizer, so the fivefold SLA-error reduction is partly by construction.

  1. fitted input called prediction [Section 3.3, Eq. (8), and Section 3.3.1 ('Uncertainty Bounding')]
    "The side-car optimizer runs the gradient scheme of Eq. (7) for R = 100 independent restarts, each seeded with a jittered copy of x(0). From the resulting sample distribution it computes a mean x∗ and a 95 % confidence interval C = [L,U] Eq. (8). ... The interval C is appended to the prompt of every LLM agent, prefixed by a short instruction: Numerical guard-rail: Offer an SLA strictly within [L,U]."

    Eq. (8)'s 'uncertainty' comes only from jittering the initial point x(0) of the deterministic GD in Eq. (7), yet the paper states that the objective is 'strictly concave and admits a unique maximizer' (Section 3.3.2). Every restart therefore converges to the same optimizer value up to numerical tolerance; s in Eq. (8) reflects arbitrary jitter and convergence noise, not a statistical confidence bound on the true Pareto-optimal SLA. The same optimizer's output is then injected as [L,U] into every LLM prompt, so the 'uncertainty bound' is not an independent external validation: it is the optimizer's own answer re-labeled as a 95% interval and fed back to the agent.

  2. self definitional [Section 5.2.3, Table 7]
    "Table 7 compares mean-absolute error (MAE) in throughput... Red rows are the optimization baseline (gradient descent, tuned/untuned)... Every symbiotic agent achieves sub-1.3 Mbps MAE—a more than 8 times reduction over its standalone counterpart. gpt-4o drops from 9.0 Mbps to 0.6Mbps."

    The MAE is measured against the 'Pareto target' that is the same gradient-descent consensus (the 'Grad-Descent tuned' row at 0.9 Mbps) used to compute [L,U] in Eq. (8). Because the guard-rail forces every bid into [L,U], a symbiotic agent's error relative to that target is bounded by construction before any LLM reasoning occurs. The reported gpt-4o drop from 9.0 to 0.6 Mbps therefore measures prompt steering toward the optimizer's own answer, not an independent improvement in SLA prediction.

full rationale

Type I evaluation (P-control plus LLM gain tuning) is self-contained: the LLM's Kp choices are evaluated against the actual P-control loop on the testbed, and the comparison against standalone LLMs, PID, Bayesian optimization, and RL baselines is an external benchmark. The NLG scoring in Tables 5-6 is also an independent human/LLM evaluation. However, the central Type II trustworthiness claim is partially circular. The '95% confidence interval' in Eq. (8) is computed by R jittered restarts of a deterministic gradient-descent optimizer on a strictly concave problem with a unique maximizer; the jitter changes only the starting point, so the spread is numerical noise rather than statistical uncertainty about the true Pareto-optimal SLA. This optimizer-derived interval is then injected into every LLM prompt as a guard-rail, and Table 7's MAE is measured against the same optimizer's consensus (the 'Grad-Descent tuned' row). Thus the large symbiotic-versus-standalone MAE reduction is substantially a prompt-steering effect: the model is told the optimizer's answer and graded on proximity to it. The paper's own Section 7 limitation ('single-cell, single-RIC') and the promise to release code only upon acceptance do not by themselves constitute circularity, but they underline that the CI computation cannot currently be independently checked. Overall: partial circularity, score 6, because Type I and the NLG results are independent while the Type II numeric claim reduces in part to construction.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central results depend on several hand-set numeric choices: the utility weights, beta schedule, learning rate, convergence threshold, Kp bounds, and target KPI. None of these receives a sensitivity analysis, and the Type II ground truth is generated by the same utility model that produces the guard-rail interval. The P-control stability assumption and the representativeness of the CQI traces are domain assumptions from prior literature and the testbed setup.

free parameters (5)
  • alpha_i and gamma_i utility weights = 7 for all agents
    Hand-set in Algorithm 1; these weights determine the shape of the Pareto-optimal SLA and therefore the confidence interval. No sensitivity analysis is provided.
  • beta schedule for mediator alignment = initial 0.5, increase 0.01 per iteration
    Hand-set in Algorithm 1; controls the mediator's pull toward the network target and affects the consensus value.
  • Gradient descent learning rate eta = 0.01
    Hand-set in Algorithm 1; affects convergence speed and the spread of jittered restarts used to build the confidence interval.
  • Convergence threshold and iteration cap = 0.5 and 1000 iterations
    Hand-set stopping rules in Algorithm 1; they define when consensus is declared.
  • Type I Kp prompt bounds and target KPI = Kp in [0.5, 1.5] with 0.1 granularity; target average iterations 1.7
    Restricts the LLM's meta-optimizer search in Listing 3 and Listing 4; changing these bounds or the target KPI would change convergence behavior and RMSE.
assumptions (5)
  • domain assumption LLMs are stochastic next-token predictors and cannot provide deterministic numeric error bounds or worst-case latency.
    Core motivation for adding optimizers; asserted in Sections 1 and 3.1 without formal proof, though widely accepted.
  • ad hoc to paper The quadratic utility functions in Eqs. (4) and (5) with the hand-set weights represent real tenant and mediator preferences, so their maximizer is the Pareto-optimal SLA.
    The Type II ground truth is an artifact of these functions; different weights change the optimum and the confidence interval.
  • standard math Gradient descent on the strictly concave objective in Eq. (6) converges geometrically to a unique maximizer.
    Standard convex-optimization result, cited in [42,43].
  • domain assumption The PRB-to-throughput plant is first-order, so proportional control with a suitably tuned gain is stable and sufficient.
    Stated in Section 3.2.2 and motivated by control theory [39]; the linear model is an approximation of the real scheduler and channel.
  • domain assumption CQI traces from 78 moving vehicles and the 3GPP CQI-to-MCS mapping emulate realistic channel fluctuations.
    The testbed is driven by recorded mobility traces rather than a live multi-cell deployment; representativeness is not independently validated beyond the cited datasets.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Symbiotic Agents: A Novel Paradigm for Trustworthy AGI-driven Networks." pith.science (2026). https://pith.science/paper/FWLO2XGL

@misc{pith2026250717695,
  author       = {Pith},
  title        = {Pith review of: Symbiotic Agents: A Novel Paradigm for Trustworthy AGI-driven Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FWLO2XGL}},
  note         = {Machine review of arXiv:2507.17695}
}
read the original abstract

Large Language Model (LLM)-based autonomous agents are expected to play a vital role in the evolution of 6G networks, by empowering real-time decision-making related to management and service provisioning to end-users. This shift facilitates the transition from a specialized intelligence approach, where artificial intelligence (AI) algorithms handle isolated tasks, to artificial general intelligence (AGI)-driven networks, where agents possess broader reasoning capabilities and can manage diverse network functions. In this paper, we introduce a novel agentic paradigm that combines LLMs with real-time optimization algorithms towards Trustworthy AI, defined as symbiotic agents. Optimizers at the LLM's input-level provide bounded uncertainty steering for numerically precise tasks, whereas output-level optimizers supervised by the LLM enable adaptive real-time control. We design and implement two novel agent types including: (i) Radio Access Network optimizers, and (ii) multi-agent negotiators for Service-Level Agreements (SLAs). We further propose an end-to-end architecture for AGI networks and evaluate it on a 5G testbed capturing channel fluctuations from moving vehicles. Results show that symbiotic agents reduce decision errors fivefold compared to standalone LLM-based agents, while smaller language models (SLM) achieve similar accuracy with a 99.9% reduction in GPU resource overhead and in near-real-time loops of 82 ms. A multi-agent demonstration for collaborative RAN on the real-world testbed highlights significant flexibility in service-level agreement and resource allocation, reducing RAN over-utilization by approximately 44%. Drawing on our findings and open-source implementations, we introduce the symbiotic paradigm as the foundation for next-generation, AGI-driven networks-systems designed to remain adaptable, efficient, and trustworthy even as LLMs advance.

Figures

Figures reproduced from arXiv: 2507.17695 by the authors.

Figure 1
Figure 1. LLM Symbiosis Paradigm: Input-level optimizers provide bounded uncertainty steering for numerically precise tasks, whereas output-level optimizers enable adaptive real-time control actions. when future LLMs improve, next-token sampling remains stochastic and can￾not yield deterministic error bounds. Therefore Oin and/or Oout remain in￾dispensable to close the gap towards AGI-grade decision-making and trust￾worthines… view at source ↗
Figure 2
Figure 2. Type I Symbiotic Agent for RAN Control. An LLM ( [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Timeline of a Type I Symbiotic Agent executing [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Throughput under a fluctuating channel (MCS 28 [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: Multi-Tenant Negotiation Topology where multiple agents (tenants) negotiate [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]
Figure 6
Figure 6. Figure 6: Type II Symbiotic Agent for SLA Negotiations: Intent Prompt is merged with [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Three negotiation Games using a pre-negotiation pipeline that bounds numerical [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: Multi-Agent negotiations employing SLA confidence intervals for reducing the [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]
Figure 9
Figure 9. Figure 9: Testbed for Next-G Open and AI-RAN Architectures. Type II symbiotic agents [PITH_FULL_IMAGE:figures/full_fig_p022_9.png]
Figure 10
Figure 10. Figure 10: Sequence Diagram presents the communication of the agentic components. Two [PITH_FULL_IMAGE:figures/full_fig_p023_10.png]
Figure 11
Figure 11. Figure 11: Snapshot of Multi-Agent Negotiations. Three agents belonging to different [PITH_FULL_IMAGE:figures/full_fig_p024_11.png]
Figure 12
Figure 12. Figure 12: Snapshots of RAN throughput using different agentic designs to enforce an [PITH_FULL_IMAGE:figures/full_fig_p026_12.png]
Figure 13
Figure 13. Figure 13: Box-plot of the different methods on enforcement of 20 Mbps intent with 5 [PITH_FULL_IMAGE:figures/full_fig_p026_13.png]
Figure 14
Figure 14. Figure 14: Comparison with traditional controllers, contextualizes the decision for a Type [PITH_FULL_IMAGE:figures/full_fig_p027_14.png]
Figure 15
Figure 15. Figure 15: Large scale experimentation of LLM negotiations with different models, number [PITH_FULL_IMAGE:figures/full_fig_p034_15.png]
Figure 16
Figure 16. Figure 16: Collaborative AGI-RAN: Multi-Tenant SLA Consensus and Resource Allo [PITH_FULL_IMAGE:figures/full_fig_p036_16.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 39 canonical work pages

  1. [1]

    rep., GSMA (Feb

    GSMA Intelligence, The mobile economy 2025, Tech. rep., GSMA (Feb. 2025). URL https://www.gsma.com/solutions-and-impact/ connectivity-for-good/mobile-economy/wp-content/uploads/ 2025/02/030325-The-Mobile-Economy-2025.pdf

  2. [2]

    Z. Cui, P. Zhang, S. Pollin, 6g wireless communications in 7-24 ghz band: Opportunities, techniques, and challenges, arXiv preprint arXiv:2310.06425 (2023)

  3. [3]

    C.-X. Wang, X. You, X. Gao, X. Zhu, Z. Li, C. Zhang, H. Wang, Y. Huang, Y. Chen, H. Haas, et al., On the road to 6g: Visions, require- ments, key technologies, and testbeds, IEEE Communications Surveys & Tutorials 25 (2) (2023) 905–974

  4. [4]

    Samdanis, X

    K. Samdanis, X. Costa-Perez, V. Sciancalepore, From network sharing to multi-tenancy: The 5g network slice broker, IEEE Communications Magazine 54 (7) (2016) 32–39

  5. [5]

    Leivadeas, M

    A. Leivadeas, M. Falkner, A survey on intent-based networking, IEEE Communications Surveys & Tutorials 25 (1) (2022) 625–655

  6. [6]

    URL https://www.o-ran.org/

    O-RAN Alliance, O-RAN: Transforming Radio Access Networks To- wards Open, Intelligent, Virtualized and Fully Interoperable RAN, ac- cessed: March 13, 2025 (2025). URL https://www.o-ran.org/

  7. [7]

    URL https://ai-ran.org/

    AI-RAN Alliance, AI-RAN Alliance: Advancing AI-Native Radio Access Networks, accessed: March 13, 2025 (2025). URL https://ai-ran.org/

  8. [8]

    Bariah, Q

    L. Bariah, Q. Zhao, H. Zou, Y. Tian, F. Bader, M. Debbah, Large gener- ative ai models for telecom: The next big thing?, IEEE Communications Magazine 62 (11) (2024) 84–90

Show all 66 references
  1. [9]

    F. Dou, J. Ye, G. Yuan, Q. Lu, W. Niu, H. Sun, L. Guan, G. Lu, G. Mai, N. Liu, et al., Towards artificial general intelligence (agi) in the internet of things (iot): Opportunities and challenges, arXiv preprint arXiv:2309.07438 (2023). 38

  2. [10]

    W. Saad, O. Hashash, C. K. Thomas, C. Chaccour, M. Debbah, N. Man- dayam, Z. Han, Artificial general intelligence (agi)-native wireless sys- tems: A journey beyond 6g, arXiv preprint arXiv:2405.02336 (2024)

  3. [11]

    Bariah, H

    L. Bariah, H. Zou, Q. Zhao, B. Mouhouche, F. Bader, M. Debbah, Understanding telecom language through large language models, arXiv preprint arXiv:2306.07933 (2023)

  4. [12]

    Huang, W

    L. Huang, W. Yu, W. Ma, W. Zhong, Z. Feng, H. Wang, Q. Chen, W. Peng, X. Feng, B. Qin, et al., A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions, ACM Transactions on Information Systems 43 (2) (2025) 1–55

  5. [13]

    Patil, V

    R. Patil, V. Gudivada, A review of current trends, techniques, and chal- lenges in large language models (llms), Applied Sciences 14 (5) (2024) 2074

  6. [14]

    L. Yuan, Y. Chen, G. Cui, H. Gao, F. Zou, X. Cheng, H. Ji, Z. Liu, M. Sun, Revisiting out-of-distribution robustness in nlp: Benchmarks, analysis, and llms evaluations, Advances in Neural Information Process- ing Systems 36 (2023) 58478–58507

  7. [15]

    E. Tabassi, Artificial Intelligence Risk Management Framework (AI RMF 1.0), Nist ai 100-1, National Institute of Standards and Technol- ogy, Gaithersburg, MD, nIST Trustworthy and Responsible AI Program (Jan. 2023). doi:10.6028/NIST.AI.100-1. URL https://tsapps.nist.gov/public...

  8. [16]

    ISO/IEC 42001:2023 Information technology — Artificial intelligence — Management system (Dec. 2023). URL https://www.iso.org/standard/81230.html

  9. [17]

    J. Song, Z. Zhou, J. Liu, C. Fang, Z. Shu, L. Ma, Self-refined large language model as automated reward function designer for deep rein- forcement learning in robotics, arXiv preprint arXiv:2309.06687 (2023)

  10. [18]

    Shinn, F

    N. Shinn, F. Cassano, A. Gopinath, K. Narasimhan, S. Yao, Reflexion: Language agents with verbal reinforcement learning, Advances in Neural Information Processing Systems 36 (2023) 8634–8652. 39

  11. [19]

    Guo, Y.-H

    P.-F. Guo, Y.-H. Chen, Y.-D. Tsai, S.-D. Lin, Towards optimizing with large language models, arXiv preprint arXiv:2310.05204 (2023)

  12. [20]

    H. Chen, G. E. Constante-Flores, C. Li, Diagnosing infeasible optimiza- tion problems using large language models, INFOR: Information Sys- tems and Operational Research 62 (4) (2024) 573–587

  13. [21]

    F. Liu, X. Lin, S. Yao, Z. Wang, X. Tong, M. Yuan, Q. Zhang, Large language model for multiobjective evolutionary optimization, in: In- ternational Conference on Evolutionary Multi-Criterion Optimization, Springer, 2025, pp. 178–191

  14. [22]

    Shahid, A

    A. Shahid, A. Kliks, A. Al-Tahmeesschi, A. Elbakary, A. Nikou, A. Maa- touk, A. Mokh, A. Kazemi, A. De Domenico, A. Karapantelakis, et al., Large-scale ai in telecom: Charting the roadmap for innovation, scalabil- ity, and enhanced digital experiences, arXiv preprint arXiv:250...

  15. [23]

    H. Zhou, C. Hu, Y. Yuan, Y. Cui, Y. Jin, C. Chen, H. Wu, D. Yuan, L. Jiang, D. Wu, X. Liu, C. Zhang, X. Wang, J. Liu, Large language model (llm) for telecommunications: A comprehensive survey on princi- ples, key techniques, and opportunities, IEEE Communications Surveys & Tut...

  16. [24]

    H. Zou, Q. Zhao, Y. Tian, L. Bariah, F. Bader, T. Lestable, M. Deb- bah, Telecomgpt: A framework to build telecom-specfic large language models, arXiv preprint arXiv:2407.09424 (2024)

  17. [25]

    Z. Shi, N. Luktarhan, Y. Song, G. Tian, Bfcn: A novel classification method of encrypted traffic based on bert and cnn, Electronics 12 (3) (2023) 516

  18. [26]

    Tsourdinis, I

    T. Tsourdinis, I. Chatzistefanidis, N. Makris, T. Korakis, Ai-driven service-aware real-time slicing for beyond 5g networks, in: IEEE INFOCOM 2022 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), 2022, pp. 1–6. doi:10.1109/ INFOCOMWKSHPS54753.2022.9798391

  19. [27]

    Tsourdinis, I

    T. Tsourdinis, I. Chatzistefanidis, N. Makris, T. Korakis, N. Nikaein, S. Fdida, Service-aware real-time slicing for virtualized beyond 5g net- works, Computer Networks 247 (2024) 110445. 40

  20. [28]

    Sousa, M

    I. Sousa, M. P. Queluz, A. Rodrigues, A survey on qoe-oriented wireless resources scheduling, Journal of Network and Computer Applications 158 (2020) 102594

  21. [29]

    Chatzistefanidis, N

    I. Chatzistefanidis, N. Makris, V. Passas, T. Korakis, Ml-based traf- fic steering for heterogeneous ultra-dense beyond-5g networks, in: 2023 IEEE Wireless Communications and Networking Conference (WCNC), 2023, pp. 1–6. doi:10.1109/WCNC55385.2023.10118923

  22. [30]

    Chatzistefanidis, N

    I. Chatzistefanidis, N. Makris, V. Passas, T. Korakis, Which ml model to choose? experimental evaluation for a beyond-5g traffic steering case, in: ICC 2023 - IEEE International Conference on Communications, 2023, pp. 5185–5190. doi:10.1109/ICC45041.2023.10279485

  23. [31]

    H. Zou, Q. Zhao, L. Bariah, M. Bennis, M. Debbah, Wireless multi- agent generative ai: From connected intelligence to collective intelli- gence, arXiv preprint arXiv:2307.02757 (2023)

  24. [32]

    Ameur, B

    M. Ameur, B. Brik, A. Ksentini, Leveraging llms to explain drl decisions for transparent 6g network slicing, in: 2024 IEEE 10th International Conference on Network Softwarization (NetSoft), IEEE, 2024, pp. 204– 212

  25. [33]

    X. Wu, J. Farooq, Y. Wang, J. Chen, Llm-xapp: A large language model empowered radio resource management xapp for 5g o-ran, in: Sympo- sium on Networks and Distributed Systems Security (NDSS), Workshop on Security and Privacy of Next-Generation Networks (FutureG 2025), San Die...

  26. [34]

    Mekrache, A

    A. Mekrache, A. Ksentini, C. Verikoukis, Intent-based management of next-generation networks: An llm-centric approach, Ieee Network (2024)

  27. [35]

    Chatzistefanidis, A

    I. Chatzistefanidis, A. Leone, N. Nikaein, Maestro: Llm-driven collabo- rative automation of intent-based 6g networks, IEEE Networking Letters 6 (4) (2024) 227–231. doi:10.1109/LNET.2024.3503292

  28. [36]

    Chatzistefanidis, N

    I. Chatzistefanidis, N. Nikaein, A. Leone, A. Maatouk, L. Tassioulas, R. Morabito, I. Pitsiorlas, M. Kountouris, Agoran: An agentic open marketplace for 6g ran automation, arXiv preprint arXiv:2508.09159 (2025). 41

  29. [37]

    Chatzistefanidis, A

    I. Chatzistefanidis, A. Leone, A. Yaghoubian, M. Irazabal, S. Nassim, L. Bariah, M. Debbah, N. Nikaein, Mx-ai: Agentic observability and control platform for open and ai-ran, arXiv preprint arXiv:2508.09197 (2025)

  30. [38]

    Visioli, Practical PID control, Springer Science & Business Media, 2006

    A. Visioli, Practical PID control, Springer Science & Business Media, 2006

  31. [39]

    K. J. ˚Astr¨ om, T. H¨ agglund, Advanced PID Control, ISA – The Instru- mentation, Systems, and Automation Society, Research Triangle Park, NC, 2006

  32. [40]

    Scutari, F

    G. Scutari, F. Facchinei, P. Song, D. P. Palomar, J.-S. Pang, Decom- position by partial linearization: Parallel optimization of multi-agent systems, IEEE Transactions on Signal Processing 62 (3) (2013) 641– 656

  33. [41]

    X. Wang, G. Wang, S. Li, A distributed fixed-time optimization algo- rithm for multi-agent systems, Automatica 122 (2020) 109289

  34. [42]

    S. Boyd, L. Vandenberghe, Convex Optimization, Cambridge University Press, Cambridge, UK, 2004. URL https://web.stanford.edu/~boyd/cvxbook/

  35. [43]

    D. P. Bertsekas, Nonlinear Programming, 2nd Edition, Athena Scien- tific, Belmont, MA, USA, 1999

  36. [44]

    Milosevic, J

    N. Milosevic, J. M¨ uller, N. Scherf, Central path proximal policy opti- mization (2025). arXiv:2506.00700. URL https://arxiv.org/abs/2506.00700

  37. [45]

    Saxena, B

    V. Saxena, B. Guldogan, D. D. Nimara, On-policy and off-policy Reinforcement Learning: Key features and differences, Ericsson Blog, published December 13, 2023. Accessed 2025-08-22 (Dec. 2023). URL https://www.ericsson.com/en/blog/2023/12/ online-and-offline-reinforcement-lear...

  38. [46]

    Eldeeb, H

    E. Eldeeb, H. Alves, Offline and distributional reinforcement learning for wireless communications (2025). arXiv:2504.03804. URL https://arxiv.org/abs/2504.03804 42

  39. [47]

    Nikaein, M

    N. Nikaein, M. K. Marina, S. Manickam, A. Dawson, R. Knopp, C. Bon- net, Openairinterface: A flexible platform for 5g research, ACM SIG- COMM Computer Communication Review 44 (5) (2014) 33–38

  40. [48]

    Schmidt, M

    R. Schmidt, M. Irazabal, N. Nikaein, Flexric: an sdk for next-generation sd-rans, in: Proceedings of the 17th International Conference on emerg- ing Networking EXperiments and Technologies, 2021, pp. 411–425

  41. [49]

    D’Oro, M

    S. D’Oro, M. Polese, L. Bonati, H. Cheng, T. Melodia, dapps: Dis- tributed applications for real-time inference and control in o-ran, IEEE Communications Magazine 60 (11) (2022) 52–58

  42. [50]

    OpenAI, Gpt-family, https://openai.com/api/, accessed: July 2025 (2024)

  43. [51]

    AI, Mistral 7b, https://mistral.ai/news/ announcing-mistral-7b/, accessed: June 2024 (2024)

    M. AI, Mistral 7b, https://mistral.ai/news/ announcing-mistral-7b/, accessed: June 2024 (2024)

  44. [52]

    URL https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence

    OpenAI, Gpt-4o mini: Advancing cost-efficient intelligence (2024). URL https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence

  45. [53]

    Grattafiori, A

    A. Grattafiori, A. Dubey, et al., The llama 3 herd of models, arXiv preprint arXiv:2407.21783 (2024). URL https://arxiv.org/abs/2407.21783

  46. [54]

    A. Yang, B. Yang, et al., Qwen2 technical report, arXiv preprint arXiv:2407.10671 (2024). URL https://arxiv.org/abs/2407.10671

  47. [55]

    Team, Gemma: Open models based on gemini research and technol- ogy, arXiv preprint arXiv:2403.08295 (2024)

    G. Team, Gemma: Open models based on gemini research and technol- ogy, arXiv preprint arXiv:2403.08295 (2024). URL https://arxiv.org/abs/2403.08295

  48. [56]

    Team, Ollama: Run large language models locally (2024)

    O. Team, Ollama: Run large language models locally (2024). URL https://github.com/ollama/ollama

  49. [57]

    Tsourdinis, I

    T. Tsourdinis, I. Chatzistefanidis, N. Makris, T. Korakis, Ue network traffic time-series (applications, throughput, latency, cqi) in lte/5g net- works, IEEE Dataport (2022)

  50. [58]

    Chatzistefanidis, N

    I. Chatzistefanidis, N. Makris, V. Passas, T. Korakis, Ue statistics time- series (cqi) in lte networks (2022). 43

  51. [59]

    Evolved universal terrestrial radio access (e-utra); physical layer pro- cedures, Tech. Rep. TS 36.213, 3rd Generation Partnership Project (3GPP), available at: https://www.3gpp.org/DynaReport/36213.htm (June 2018)

  52. [60]

    Y. Liu, D. Iter, Y. Xu, S. Wang, R. Xu, C. Zhu, G-eval: Nlg evaluation using gpt-4 with better human alignment, arXiv preprint arXiv:2303.16634 (2023)

  53. [61]

    Fu, S.-K

    J. Fu, S.-K. Ng, Z. Jiang, P. Liu, Gptscore: Evaluate as you desire, arXiv preprint arXiv:2302.04166 (2023)

  54. [62]

    C.-M. Chan, W. Chen, Y. Su, J. Yu, W. Xue, S. Zhang, J. Fu, Z. Liu, Chateval: Towards better llm-based evaluators through multi-agent de- bate, arXiv preprint arXiv:2308.07201 (2023)

  55. [63]

    Z. Guo, R. Jin, C. Liu, Y. Huang, D. Shi, L. Yu, Y. Liu, J. Li, B. Xiong, D. Xiong, et al., Evaluating large language models: A comprehensive survey, arXiv preprint arXiv:2310.19736 (2023)

  56. [64]

    Spearman, The proof and measurement of association between two things, The American Journal of Psychology 15 (1) (1904) 72–101

    C. Spearman, The proof and measurement of association between two things, The American Journal of Psychology 15 (1) (1904) 72–101. doi: 10.2307/1412159

  57. [65]

    M. G. Kendall, A new measure of rank correlation, Biometrika 30 (1-2) (1938) 81–93. doi:10.1093/biomet/30.1-2.81. 44 Appendix A: Optimization Algorithm for Type II Agents The optimization algorithm 1 is used by the Type II agents to model the whole topology as a distributed op...

  58. [66]

    throughput\

    Second your SLA list of parameters. Important: This is your response SLA template: Reasoning: (Your reasoning in 1-2 sentences) {\"throughput\": 0}. An optimization algorithm calculates a confidence interval of SLA values where the optimal SLA exists. The algorithm suggests th...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.