REVIEW 20 cited by
Embers of Autoregression: Understanding Large Language Models Through the Problem They are Trained to Solve
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
Embers of Autoregression: Understanding Large Language Models Through the Problem They are Trained to Solve
read the original abstract
The widespread adoption of large language models (LLMs) makes it important to recognize their strengths and limitations. We argue that in order to develop a holistic understanding of these systems we need to consider the problem that they were trained to solve: next-word prediction over Internet text. By recognizing the pressures that this task exerts we can make predictions about the strategies that LLMs will adopt, allowing us to reason about when they will succeed or fail. This approach - which we call the teleological approach - leads us to identify three factors that we hypothesize will influence LLM accuracy: the probability of the task to be performed, the probability of the target output, and the probability of the provided input. We predict that LLMs will achieve higher accuracy when these probabilities are high than when they are low - even in deterministic settings where probability should not matter. To test our predictions, we evaluate two LLMs (GPT-3.5 and GPT-4) on eleven tasks, and we find robust evidence that LLMs are influenced by probability in the ways that we have hypothesized. In many cases, the experiments reveal surprising failure modes. For instance, GPT-4's accuracy at decoding a simple cipher is 51% when the output is a high-probability word sequence but only 13% when it is low-probability. These results show that AI practitioners should be careful about using LLMs in low-probability situations. More broadly, we conclude that we should not evaluate LLMs as if they are humans but should instead treat them as a distinct type of system - one that has been shaped by its own particular set of pressures.
Forward citations
Cited by 20 Pith papers
-
Gradient-Based Program Synthesis with Neurally Interpreted Languages
NLI autonomously discovers a vocabulary of primitive operations and interprets variable-length programs via a neural executor, allowing end-to-end training and gradient-based test-time adaptation that outperforms prio...
-
Brain-LLM Alignment Tracks Training Data, Not Typology
Training-language dominance, not English inherent properties, determines brain-LLM alignment across English, Chinese, and French, with additional independent effects from typological distance concentrated in syntactic...
-
Deep Reasoning in General Purpose Agents via Structured Meta-Cognition
DOLORES, an agent using a formal language for meta-reasoning to construct adaptive scaffolds on the fly, outperforms prior scaffolding methods by 24.8% on average across four hard benchmarks and multiple model sizes.
-
Is She Even Relevant? When BERT Ignores Explicit Gender Cues
A Dutch BERT model encodes gender linearly by epoch 20 but does not dynamically update its representations when explicit female cues contradict learned stereotypical associations in short sentence templates.
-
GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models
LLMs display high variance and major accuracy drops on GSM-Symbolic variants of grade-school math problems, indicating they replicate training patterns rather than execute logical reasoning.
-
SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering
SWE-agent introduces a custom agent-computer interface that lets LM agents solve software engineering tasks, reaching 12.5% pass@1 on SWE-bench and 87.7% on HumanEvalFix, exceeding prior non-interactive approaches.
-
Reasoning as Pattern Matching: Shared Mechanisms in Human and LLM Everyday Reasoning
Humans and LLMs exhibit similar error patterns in common-sense reasoning, consistent with shared pattern-matching mechanisms rather than abstract world models.
-
Consistency Training while Mitigating Obfuscation via Rate Matching
RMCT matches the rate of target behaviors like bias-following across input perturbations to reduce sycophancy in LLMs while preserving verbalization of bias cues.
-
Revisiting Parameter-Based Knowledge Editing in Large Language Models: Theoretical Limits and Empirical Evidence
Parameter-based knowledge editing in LLMs induces reasoning collapse via dimensional collapse and is consistently outperformed by a retrieval baseline across varied edit counts, knowledge complexity, and evaluation metrics.
-
SuperVoxelGPT: Adaptive and Ordered 3D Tokenization for Autoregressive Shape Generation
Adaptive saliency-guided supervoxel tokenization cuts 3D AR token length to 12.8% of uniform voxels while claiming SOTA quality and ~10× speedup on Trellis-500K.
-
Using Probabilistic Programs to Train Inductive Reasoning in Large Language Models
PPT generates probabilistic programs via LLM, runs inference for soft labels, and fine-tunes LLMs, yielding better accuracy, human alignment, and calibration on inductive tasks than baselines.
-
Investigating Concept Alignment Using Implausible Category Members
AI models misalign with humans on concept boundaries when probed with implausible category members, such as classifying words as vehicles or vegetables as fruit.
-
Measuring Distribution Shift in User Prompts and Its Effects on LLM Performance
The LENS framework applied to 192 real-world settings shows moderate natural prompt distribution shifts cause 73% average performance loss in deployed LLMs, especially across user groups and regions.
-
When Verification Fails: How Compositionally Infeasible Claims Escape Rejection
AI claim verification models rely on salient-constraint shortcuts instead of full compositional reasoning under the closed-world assumption, as revealed by their over-acceptance of claims with supported salient constr...
-
How Do Language Models Compose Functions?
LLMs solve compositional factual recall either by computing intermediates or directly, with mechanism choice correlated to translation geometry in embedding spaces.
-
Empirical Study for Structured Output Control in LLMs for Software Engineering
Empirical benchmarks on four SE tasks show grammar-constrained decoding and TTMG eliminate most syntax errors in LLM outputs while structural and semantic errors persist and cascade in downstream tools.
-
SuperVoxelGPT: Adaptive and Ordered 3D Tokenization for Autoregressive Shape Generation
SuperVoxelGPT creates shape-adaptive, deterministically ordered supervoxel tokens via saliency-guided CVT, cutting sequence length to 12.8% of uniform voxels while claiming SOTA quality and 10x speedup on Trellis-500K.
-
How Well Do LLMs Perform on the Simplest Long-Chain Reasoning Tasks: An Empirical Study on the Equivalence Class Problem
Non-reasoning LLMs fail the equivalence class problem while reasoning LLMs perform better but remain incomplete, with difficulty peaking at phase transition for the former and maximum diameter for the latter.
-
Position: Stop Evaluating AI with Human Tests, Develop Principled, AI-specific Tests instead
Human tests should not be applied to AI to measure traits like intelligence due to calibration, validity, contamination, and prompt sensitivity issues; develop AI-specific evaluation frameworks instead.
-
Losing our Tail, Again: (Un)Natural Selection & Multilingual LLMs
Position paper warns that model collapse in self-consuming multilingual LLM training loops risks flattening linguistic diversity and cultural nuance.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.