Pith. sign in

REVIEW 12 cited by

Roll the dice & look before you leap: Going beyond the creative limits of next-token prediction

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2504.15266 v4 pith:OBR2JOXC submitted 2025-04-21 cs.LG cs.AIcs.CL

Roll the dice & look before you leap: Going beyond the creative limits of next-token prediction

classification cs.LG cs.AIcs.CL
keywords taskscreativelikenext-tokenopen-endedbeyondgoinglayer
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We design a suite of minimal algorithmic tasks that are a loose abstraction of open-ended real-world tasks. This allows us to cleanly and controllably quantify the creative limits of the present-day language model. Much like real-world tasks that require a creative, far-sighted leap of thought, our tasks require an implicit, open-ended stochastic planning step that either (a) discovers new connections in an abstract knowledge graph (like in wordplay, drawing analogies, or research) or (b) constructs new patterns (like in designing math problems or new proteins). In these tasks, we empirically and conceptually argue how next-token learning is myopic; multi-token approaches, namely teacherless training and diffusion models, comparatively excel in producing diverse and original output. Secondly, to elicit randomness without hurting coherence, we find that injecting noise at the input layer (dubbed seed-conditioning) works surprisingly as well as (and in some conditions, better than) temperature sampling from the output layer. Thus, our work offers a principled, minimal test-bed for analyzing open-ended creative skills, and offers new arguments for going beyond next-token learning and temperature sampling. We make part of the code available under https://github.com/chenwu98/algorithmic-creativity

discussion (0)

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

Forward citations

Cited by 12 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Language Modeling with Hyperspherical Flows

    cs.LG 2026-05 unverdicted novelty 8.0

    S-FLM rotates vectors on a hypersphere using a learned velocity field to generate language sequences, improving continuous flow models on large-vocabulary reasoning and closing the gap to masked diffusion at standard ...

  2. Assessing the Creativity of Large Language Models: Testing, Limits, and New Frontiers

    cs.AI 2026-05 conditional novelty 7.0

    The Divergent Remote Association Test (DRAT) is the first creativity test that significantly predicts LLMs' scientific ideation ability, unlike prior tests such as DAT or RAT.

  3. Language Modeling with Hyperspherical Flows

    cs.LG 2026-05 unverdicted novelty 7.0

    S-FLM is a hyperspherical latent flow language model that improves continuous flow language models on large-vocabulary reasoning tasks and closes the gap to masked diffusion at standard sampling temperature.

  4. Distilled Pretraining: A modern lens of Data, In-Context Learning and Test-Time Scaling

    cs.LG 2025-09 conditional novelty 7.0

    Distilled pretraining improves test-time scaling via generation diversity but impairs induction-head-based in-context learning, with the trade-off explained by a bigram model analysis.

  5. Hierarchical Domain Generalization

    cs.LG 2026-07 conditional novelty 6.0

    Over infinite domains, hierarchy-uniform domain generalization is impossible for every nontrivial hypothesis class; a length-generalization bound is a property of the length hierarchy, not a hierarchy-free guarantee.

  6. BlockGen: Flexible Blockwise Sequence Modeling with Hybrid Samplers

    cs.LG 2026-06 unverdicted novelty 6.0

    BlockGen enables flexible blockwise diffusion modeling with mixed block sizes and ARPC sampling, finding uniform diffusion outperforms masked under ancestral sampling in few-step regimes while the gap reverses with AR...

  7. MLReplicate: Benchmarking Autonomous Research Systems for Machine Learning Reproducibility

    cs.LG 2026-05 conditional novelty 6.0

    MLReplicate benchmark evaluates six autonomous systems on 45 manuscripts from ICML 2025 papers, finding that automated reviews accept flawed outputs with fabricated claims while human review exposes methodological fai...

  8. Language Modeling with Hyperspherical Flows

    cs.LG 2026-05 unverdicted novelty 6.0

    S-FLM is a hyperspherical latent flow language model that learns velocity fields on the unit sphere to generate token sequences via deterministic ODE integration without materializing one-hot vectors.

  9. Next-Latent Prediction Transformers Learn Compact World Models

    cs.LG 2025-11 conditional novelty 6.0

    Adding a next-latent prediction loss to next-token training makes transformer hidden states more predictive of future tokens and improves planning/reasoning on small benchmarks.

  10. Next-Latent Prediction Transformers Learn Compact World Models

    cs.LG 2025-11 unverdicted novelty 6.0

    NextLat augments next-token prediction with latent next-state prediction, theoretically converging latents to belief states and showing empirical gains in world modeling, reasoning, planning, and faster inference via ...

  11. HypoSpace: A Diagnostic Benchmark for Set-Valued Hypothesis Generation under Underdetermination and Sublinear Coverage Bounds

    cs.CL 2025-10 conditional novelty 6.0

    A benchmark with exactly enumerated valid hypothesis sets shows LLMs maintain high validity but lose uniqueness and coverage as the admissible solution space grows.

  12. Polychromic Objectives for Reinforcement Learning

    cs.LG 2025-09 unverdicted novelty 5.0

    Introduces polychromic objectives adapted into PPO via vine sampling and modified advantages, showing higher success rates and better coverage under perturbations on BabyAI, Minigrid, and algorithmic tasks.