English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

AIRA Deep Dive: When AI Agents Design Their Own Neural Networks—How Far Are We from AI Researching AI?

Forum topic · 小凯 · 2026-06-21

Summary

This zhichai.net forum post analyzes AIRA (Agentic Discovery of Neural Architectures), a Meta FAIR research framework (arXiv:2605.15871) where LLM agents design neural network architectures from raw computational primitives like Attention, MLP, and Mamba. AIRA consists of two complementary systems: AIRA-Compose, where 11 parallel agents propose and evaluate 16-layer candidate architectures before extrapolating winners to 350M–3B parameters, and AIRA-Design, where 20 agents implement novel attention mechanisms and training scripts from scratch. The resulting AIRAformer and AIRAhybrid model families reportedly outperform Llama 3.2 baselines by up to +3.8% downstream accuracy at 1B parameters, with AIRAformer-C showing +54% scaling speed and AIRAhybrid-C showing +23% scaling performance versus Nemotron-2. Agents spontaneously discovered heterogeneous layer patterns mixing Attention and Mamba rather than uniform stacking. However, on Long Range Arena tasks agents still trail human SOTA (Normalized Scores 0.51–0.88), primitives remain human-supplied, and giving agents more freedom (configurable setups) degraded results. The post concludes AIRA demonstrates AI-assisted—not autonomous—architecture research is viable, and the key open question is whether agents can invent entirely new computational primitives.

Overview

This post dissects AIRA (Agentic Discovery of Neural Architectures: AIRA-Compose and AIRA-Design) by Alberto Pepe, Chien-Yu Lin, Despoina Magka, Bilge Acun, Yannan Nellie Wu, Anton Protopopov, Carole-Jean Wu, and Yoram Bachrach (Meta FAIR).

  • Paper: https://arxiv.org/abs/2605.15871
  • The core question: since Transformers have dominated deep learning since 2017's *Attention Is All You Need*, would an AI agent designing architectures from scratch choose a Transformer? Traditional NAS searches within human-defined spaces using deterministic strategies (Bayesian optimization, evolution). AIRA asks something more radical: give agents raw computational primitives (Attention, MLP, Mamba, convolution) and let them decide how to combine, order, and scale them.

    Key points

    Two-layer framework

    AIRA-Compose (architecture search), built on an agentic adaptation of Meta's Composer framework: 1. Agentic Exploration — 11 LLM agents work in parallel, freely proposing structural hypotheses that combine primitives into 16-layer small-scale candidates 2. Fast-Proxy Evaluation — quick train/validation on small datasets, scored by validation loss and downstream accuracy 3. Aggregation — layer-wise clustering picks the most frequent primitives, smoothing proxy-training noise 4. Extrapolation — stretching contiguous blocks and stacking the architecture to reach 350M / 1B / 3B parameters

    The 16-layer proxy is key: small-scale performance correlates strongly with large-scale results, so agents iterate fast at small scale before winners are scaled up.

    AIRA-Design (low-level implementation): 20 agents write code from scratch—novel attention mechanisms for Long Range Arena tasks (ListOps, Text, Retrieval) and training-script optimization (minimizing validation loss under fixed time). Compose searches the macro architecture; Design handles micro implementation.

    Outputs: two model families, 14 models

  • AIRAformers: pure Transformer variants differing from standard Transformers in layer ordering, width ratios, and connectivity patterns
  • AIRAhybrids: Transformer + Mamba hybrids where agents found Mamba more efficient than Attention at certain layers
  • Experimental results (1B parameters, fixed token budget)

    | Model | vs Llama 3.2 | vs Nemotron-2 | Note | |:---|:---|:---|:---| | AIRAhybrid-D | +3.8% downstream accuracy | - | Best hybrid | | AIRAhybrid-C | - | +23% scaling performance | Better scaling | | AIRAformer-C | - | - | +54% scaling speed (highest efficiency) |

    Agent-designed architectures also show steeper scaling curves—more capability gain per parameter added.

    Long Range Arena (Normalized Score, 1.0 = human SOTA):

    | Task | Best Agent | Score | Gap | |:---|:---|:---|:---| | ListOps | Greedy Opus 4.6 | 0.51 | -49% | | Text | Greedy Gemini 3 Pro | 0.88 | -12% | | Retrieval | Greedy Opus 4.6 | 0.79 | -21% |

    Key findings:

  • Configurable setups performed worse: more freedom (tunable hyperparameters) led agents to get lost in oversized search spaces
  • Greedy beats one-shot: iterative refinement significantly outperforms one-shot generation
  • Opus 4.6 vs Gemini 3 Pro: different base LLMs excel at different tasks, bounding agent capability
  • What did agents discover that humans missed?

    1. Emergent hybrid architectures — agents spontaneously proposed mixing Attention and Mamba across layers; unlike NAS, the search space itself was agent-invented 2. Heterogeneous layer patterns — instead of uniform stacking, layers at different depths show distinct primitive distributions, suggesting shallow layers do local feature extraction while deep layers do global semantic integration 3. Improved scaling efficiency — possibly a more efficient parameter-to-capability conversion

    Limitations (per the paper)

  • Significant gaps remain on long-range tasks vs human-designed specialists (S4, H3, Mamba)
  • Primitives themselves are still human-provided; agents cannot yet invent new computational primitives
  • Configurable setups hurt performance; agents can't yet handle high-dimensional continuous spaces
  • The 16-layer proxy correlation is imperfect; winners may not stay optimal at 3B
  • Non-trivial compute cost (24-hour GPU budgets, multi-agent parallelism)
  • No theoretical explanation of why agent designs work; reproducibility variance underreported; humans still build the training infrastructure
  • How far are we from "AI researching AI"?

    Current AIRA is AI-assisted architecture search: agents propose novel combinations, but primitives, evaluation criteria, infrastructure, and final code review are all human. Next steps require:

    1. Primitive discovery — inventing new computational primitives, not just recombining known ones 2. End-to-end loops — design → training code → experiments → analysis → improvement 3. Theoretical explanation — agents deriving why their designs work from first principles 4. Self-improvement — agent-designed architectures training stronger agents (recursive self-improvement)

    Context vs related work: traditional NAS (AutoML, DARTS) uses fixed search strategies; Composer (Meta, 2025) is non-agentic; AlphaEvolve (DeepMind, 2025) targets math algorithms; AI Scientist v2 (Sakana, 2025) is more end-to-end but focused on paper generation.

    2026 signals across this week's papers (OProver, HarnessX, LFM2.5, AIRA): algorithms beat parameters (32B beating 560B; 350M beating 600M); system-level optimization can outweigh model swaps; and AI-assisted R&D—not AI-replacing R&D—is the realistic path.

    AIRA's real value isn't that agents beat humans (they don't, especially on long-range tasks), but that it proves the path is viable: given raw primitives, agents find effective combinations humans never tried. The pivotal question: when will agents invent computational structures beyond Transformers and Mamba? That's when "AI researching AI" truly begins.

    References

  • Paper: https://arxiv.org/abs/2605.15871
  • Composer (Meta NAS framework): https://arxiv.org/abs/2502.07897
  • Long Range Arena: https://arxiv.org/abs/2011.04006
  • Mamba (state space model): https://arxiv.org/abs/2312.00752

Tags

#ai-agents#neural-architecture-search#meta-fair#transformer#mamba#hybrid-architectures#llm#scaling-laws

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178203246