SYNTHESIS NOTE
Do large language models actually perform iterative optimization?
Explores whether LLMs execute genuine numerical procedures like Newton-Raphson or instead pattern-match to memorized solution templates when solving constrained optimization problems.
The constraint-optimization study identifies the mechanism behind the 55-60% plateau directly. LLMs cannot actually perform Newton-Raphson iterations in their latent space. They cannot execute primal-dual updates, nor any other iterative numerical procedure that genuine optimization requires. When asked to do so, they fall back to what the paper calls "result guessing" — recognizing the problem as similar to a standard power grid (or financial dataset, or security scenario) and emitting values that pattern-match what a valid solution should look like.
The fallback is silent. The output is fluent, well-formatted, often plausible. It can pass surface-level inspection because the model has seen many examples of what answers in this domain look like. What it has not done is solve the problem. The constraint values are wrong in ways that physical or financial systems would actually reject.
This explains why scale, architecture, and training regime do not move the plateau. They improve the template but not the procedure. A larger model has seen more example solutions and can produce more convincing guesses. Reinforcement learning on outcome rewards reinforces the template-matching pattern. None of this installs the iterative-computation capability the problem requires.
The mechanism — pattern-match against memorized solution-shapes when genuine computation is required — generalizes beyond optimization. It is plausibly the same mechanism behind a class of mathematical-reasoning failures where models produce confidently wrong numerical answers that resemble the right shape. The category is "looks like a solution; is not derived from one."
Inquiring lines that read this note 122
This note is a source for these research framings, grouped by the broader line of inquiry each explores. Scan the bold lines of inquiry; follow any specific question forward.
How can AI alignment serve diverse human preferences at scale? Which computational strategies best support reasoning in language models?- Can closed-form solutions compete with gradient descent optimization?
- How does latent space diffusion enable evolutionary search in high dimensions?
- How does fitness-proportional selection guide LLM recombination in unstructured solution spaces?
- How can gradients flow through discrete document selection?
- Do task-specific heuristics emerge because they compress well enough?
- Do latent sequence vectors outperform per-token latent iterative computation for reasoning?
- Can textual gradients generalize natural language feedback across computation graphs?
- Can optimization algorithms exploit the shift between procedural and planning bottlenecks?
- How many particles and iterations does optimal expert discovery require?
- Can a trained decoder replace both search and parameter updates?
- What distinguishes minimal-pair asymmetry from standard accuracy evaluation?
- How do unstated constraints become invisible to training data distributions?
- Can benchmarks designed for shortcut learning detect heuristic override failures?
- Why does iterative refinement fail when information stays constant?
- Can trustworthy scoring prevent persistent iteration from compounding errors?
- Do iterative refinement methods reproduce the same overthinking failure mode?
- Why do sigmoid conflict curves look the same across different language models?
- Why do language models fall back on frequency heuristics under structural complexity?
- Do LLMs rely on surface heuristics instead of learning recursive grammar rules?
- How do description-based identifiers bias language model output distribution?
- Why do smaller models favor code formats while larger models prefer natural language?
- Do pretrained language models carry reusable computational scaffolding for length handling?
- How do parameter scaling and latent vectors interact in language models?
- Can explicit constraint statements override the dominance of surface heuristics?
- Would hybrid systems combining LLMs with symbolic solvers overcome the retraction limitation?
- How do deterministic symbolic solvers improve the reliability of language model reasoning?
- How should benchmarks test whether models fit algorithms or patterns?
- Does the Heuristic Override Benchmark measure enumeration or world knowledge?
- Can universal function approximators be expensive to learn in practice?
- What decomposition level minimizes both error rate and computational cost in practice?
- Why do task-specific heuristics fail at generalizing to sparse data regions?
- Can instance seeds work for tasks beyond language understanding benchmarks?
- Why do simple length heuristics outperform sophisticated semantic methods?
- Why does extended chain-of-thought reasoning fail to improve numerical optimization performance?
- Why do LLM outputs match researcher priors without solving tasks correctly?
- Why do backward-looking benchmarks underestimate LLM scientific value?
- Why do token-level language models fail at utterance-level pragmatic optimization?
- Does sequence prediction accuracy prove an underlying world model exists?
- Why is latent-level prediction more sample-efficient than token-level prediction?
- What makes fixed-point convergence better than learned halt tokens?
- Does scaling model size solve compositional generalization problems?
- Can recursion alone drive generalization better than model scale?
- How does the discrete token bottleneck prevent gradient flow in language model control?
- Why do different LLMs converge on similar outputs in open-ended tasks?
- Why does most refinement in iterative models maintain answers rather than improve them?
- Why does iterative refinement amplify rather than correct reasoning errors?
- Can prompt optimization alone inject knowledge models don't already have?
- Can prompt optimization inject new knowledge into language models?
- Can prompt optimization inject genuinely new knowledge into a model?
- What knowledge can prompt optimization actually activate in trained models?
- Why does prompt optimization alone fail to inject genuinely new knowledge?
- Can latent recurrence and energy minimization both escape the same computational depth constraints?
- Can neural networks implement genuine algorithms or only statistical pattern matching?
- What formal language complexity level matches transformer computational limits best?
- Do language models build world models or just task-specific heuristics?
- How can we probe LLM representations in channels that training did not target?
- Why do large language models still have systematic blind spots with complex structures?
- Why do rare complex structures in training data harm LLM generalization?
- How do general language model benchmarks predict specialized domain performance?
- Do standard language benchmarks underestimate what LLMs can actually do?
- Why do standard NLP benchmarks hide the most critical language limitations?
- Why does genetic programming outperform direct LLM generation by 86 percent?
- What mechanism causes LLMs to plateau on numerical optimization tasks?
- Why do language models plateau at 55 to 60 percent constraint satisfaction?
- Why do LLMs fail at directly solving stochastic control problems?
- What latent mechanisms do LLMs use when they cannot execute iterative methods?
- Why do language models fail at iterative numerical optimization despite scale?
- Why do language models plateau at constraint satisfaction regardless of scale?
- Can language models execute iterative numerical methods in latent space?
- What prevents monolithic LLMs from coordinating decomposition with execution?
- Why do LLMs fail at iterative numerical computation in latent space?
- What constraint satisfaction rate do LLMs achieve at scale?
- Why do LLMs struggle more when only numerical values change?
- Are newer larger language models actually worse at faithful summarization?
- How do LLM activations sparsify differently under out-of-distribution inputs?
- Why does fixing decomposition step count matter more than vocabulary alignment?
- How does algorithmic control flow define computational graph structure in LLM programs?
- What distinguishes hierarchical dual-recurrence from flat parameter-sharing recurrence?
- Do instruction-tuned models learn tasks or just output format distributions?
- Why does teacher forcing fail to capture long-range dependencies?
- Can LLMs reliably generate novel working architectures without structured representations?
- Can LLMs successfully translate natural language into formal solver specifications?
- What concrete problems do LLMs solve at the computational level?
- Can LLMs recover true joint distributions from marginal census data?
- Do independent LLM outputs converge enough to create artificial hiveminds?
- Why do large language models outperform fine-tuned models once repeated items are removed?
- What non-parametric methods could replace latent factors for inductive learning?
- Do small models show different parameter efficiency patterns than large models?
- How should tiny language models be architected differently than large ones?
- Can width-scaling replace depth-scaling on inherently sequential problems?
- Can a two-layer network outgeneralize billion-parameter models through recursion alone?
- What happens when we treat LLM outputs as sampled rather than stored?
- Can surface-level correctness hide failures in structural learning by LLMs?
- Why do smaller LLMs fail at zero-shot argument scheme classification?
- Can a tiny recursive network beat billion-parameter models on hard problems?
Related concepts in this collection 5
This note in its neighbourhood — explore the map, then jump to a related concept in the list below.
Concept map
Do large language models actually perform iterat…
Do larger language models solve constrained optimi…
Do reasoning models actually beat standard models …
Do fine-tuned language models actually learn optim…
Does chain-of-thought reasoning reveal genuine inf…
What do models actually learn from chain-of-though…
Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph
your link
semantically near
linked from elsewhere
-
Do larger language models solve constrained optimization better?
Explores whether scaling LLMs—through more parameters, better training, or reasoning extensions—improves their ability to satisfy constraints in real optimization problems like power grids and portfolios.
same paper, the plateau this mechanism explains
-
Do reasoning models actually beat standard models on optimization?
Explores whether extended chain-of-thought in reasoning models delivers performance gains on constraint-satisfaction problems like power-grid optimization. Matters because reasoning models are treated as automatic upgrades, but the evidence may not support that claim.
same paper, why extended CoT does not fix it
-
Do fine-tuned language models actually learn optimization procedures?
Can RL fine-tuning teach LLMs to solve constraint-optimization problems through genuine reasoning, or does it merely sharpen pattern-matching? Testing on out-of-distribution variants reveals the mechanism.
same paper, the diagnostic that exposes the memorization
-
Does chain-of-thought reasoning reveal genuine inference or pattern matching?
Explores whether CoT instructions unlock real reasoning capabilities or simply constrain models to mimic familiar reasoning patterns from training data. This matters for understanding whether language models can actually reason abstractly.
adjacent: imitation vs computation
-
What do models actually learn from chain-of-thought training?
When models train on reasoning demonstrations, do they memorize content details or absorb reasoning structure? Testing with corrupted data reveals which aspects of CoT samples actually drive learning.
adjacent: form-over-content failure mode
Related papers in this collection 8
Papers most semantically related to this note, ranked by cosine similarity in the embedding space.
- Branch-Solve-Merge Improves Large Language Model Evaluation and Generation
- Can Large Language Models Reason and Optimize Under Constraints?
- Algorithm of Thoughts: Enhancing Exploration of Ideas in Large Language Models
- Metacognitive Reuse: Turning Recurring LLM Reasoning Into Concise Behaviors
- Chain of Thoughtlessness? An Analysis of CoT in Planning
- Beyond Accuracy: Evaluating the Reasoning Behavior of Large Language Models -- A Survey
- A Mechanistic Analysis of Looped Reasoning Language Models
- Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models
Original note title
LLMs cannot execute iterative numerical methods in latent space and fall back to result guessing against memorized templates