NEWWriting, teamwork, and fresh ideas · See what is new→

Research moves forward.
Your judgment leads.

Ask questions grounded in your papers and notes. Agents track new research and next steps.

Get started free → Read the guide

Open to everyone · Usage-based Credits

SciArk Background tasks Ddemo

← Library LLM

LLM Reasoning · Wiki Interest Point · LLM Reasoning

20 papers · Generated Oct 12, 2026 at 20:16 Wiki ready

Refresh Ask this Library

OVERVIEW

This Library brings together 20 papers on LLM reasoning. It traces how chain-of-thought prompting develops into self-consistency voting, Tree / Graph of Thoughts search, and step-by-step process reward models. Work on tool use (ReAct, PAL) and self-reflection (Reflexion, Self-Refine) adds 31 cross-references.

Papers 20

TRTree of Thoughts: Deliberate Problem Solving with Large Language ModelsIncluded
SESelf-Consistency Improves Chain of Thought Reasoning in Language ModelsIncluded
LELet's Verify Step by StepIncluded
CHChain-of-Thought Prompting Elicits Reasoning in Large Language ModelsIncluded
LALarge Language Models are Zero-Shot ReasonersIncluded
LELeast-to-Most Prompting Enables Complex Reasoning in Large Language ModelsIncluded
REReAct: Synergizing Reasoning and Acting in Language ModelsIncluded
PAPAL: Program-aided Language ModelsIncluded
STSTaR: Bootstrapping Reasoning With ReasoningIncluded
REReflexion: Language Agents with Verbal Reinforcement LearningIncluded
SESelf-Refine: Iterative Refinement with Self-FeedbackIncluded
GRGraph of Thoughts: Solving Elaborate Problems with Large Language ModelsIncluded
TRTraining Verifiers to Solve Math Word ProblemsIncluded
SHShow Your Work: Scratchpads for Intermediate Computation with Language ModelsIncluded
SOSolving Quantitative Reasoning Problems with Language ModelsIncluded
FAFaithful Chain-of-Thought ReasoningIncluded
SCScaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model ParametersIncluded

Graph statistics

Papers: 20
Entities: 51
Relations: 50
New entries: +4

Ask this Library

Search across every entity in the Library Wiki. Answers link to entities and source papers.

Open chat

Self-consistency or Tree of Thoughts: which suits math reasoning?

Retrieval · 1 tool call

It depends on the task. Self-consistency samples reasoning paths and votes on the answer. It is simple and works well on arithmetic benchmarks such as GSM8K. Tree of Thoughts adds search and backtracking, making it a better fit for trial-and-error tasks such as the Game of 24, at the cost of many more model calls.

self-consistencytree-of-thoughts

Self-consistency or Tree of Thoughts: which suits math reasoning?

Retrieval · 1 tool call

It depends on the task. Self-consistency samples reasoning paths and votes on the answer. It is simple and works well on arithmetic benchmarks such as GSM8K. Tree of Thoughts adds search and backtracking, making it a better fit for trial-and-error tasks such as the Game of 24, at the cost of many more model calls.

self-consistencytree-of-thoughts

Ask a follow-up…

Follow your research interests across open academic sources

arXiv
PubMed
Europe PMC
OpenAlex
Semantic Scholar
Crossref

Building custom research skills with labs at Zhejiang University, Xidian University, Henan University, and beyond

From a research interest to an answer.

Search, select, compile, and question. Every step in one place.

01 / 03

Let agents search. You choose what matters.

Agents search open academic sources for your Interest Points. You choose which Candidate Pool results enter your Library.

Grouped by search. Sources in view.
One-click import, with deduplication

SciArk Background tasks Ddemo

← Interest Points

LLM Reasoning

20 results · 3 ready to add

LLMAdd to “LLM Reasoning” Search again

To add 3In Library 17All batches 20

2026/10/12 20:14Scheduled searcharXiv20 results · 3 to add

Tree of Thoughts: Deliberate Problem Solving with Large Language Models
Shunyu Yao, Dian Yu, Jeffrey Zhao et al. · 2023 · arXiv:2305.10601 · To add
4.80
Self-Consistency Improves Chain of Thought Reasoning in Language Models
Xuezhi Wang, Jason Wei, Dale Schuurmans et al. · 2022 · arXiv:2203.11171 · To add
4.80
Let's Verify Step by Step
Hunter Lightman, Vineet Kosaraju, Yura Burda et al. · 2023 · arXiv:2305.20050 · To add
4.60
Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Jason Wei, Xuezhi Wang, Dale Schuurmans et al. · 2022 · arXiv:2201.11903 · In Library
4.90
Large Language Models are Zero-Shot Reasoners
Takeshi Kojima, Shixiang Shane Gu, Machel Reid et al. · 2022 · arXiv:2205.11916 · In Library
4.70
Least-to-Most Prompting Enables Complex Reasoning in Large Language Models
Denny Zhou, Nathanael Schärli, Le Hou et al. · 2022 · arXiv:2205.10625 · In Library
4.50
ReAct: Synergizing Reasoning and Acting in Language Models
Shunyu Yao, Jeffrey Zhao, Dian Yu et al. · 2022 · arXiv:2210.03629 · In Library
4.50
PAL: Program-aided Language Models
Luyu Gao, Aman Madaan, Shuyan Zhou et al. · 2022 · arXiv:2211.10435 · In Library
4.30
STaR: Bootstrapping Reasoning With Reasoning
Eric Zelikman, Yuhuai Wu, Jesse Mu et al. · 2022 · arXiv:2203.14465 · In Library
4.20
Reflexion: Language Agents with Verbal Reinforcement Learning
Noah Shinn, Federico Cassano, Edward Berman et al. · 2023 · arXiv:2303.11366 · In Library
4.20
Self-Refine: Iterative Refinement with Self-Feedback
Aman Madaan, Niket Tandon, Prakhar Gupta et al. · 2023 · arXiv:2303.17651 · In Library
4.10
Graph of Thoughts: Solving Elaborate Problems with Large Language Models
Maciej Besta, Nils Blach, Ales Kubicek et al. · 2023 · arXiv:2308.09687 · In Library
4.00

SciArk Background tasks Ddemo

← Library LLM

LLM Reasoning · Wiki Interest Point · LLM Reasoning

20 papers · Generated Oct 12, 2026 at 20:16 Wiki ready

Refresh Ask this Library

OVERVIEW

Papers 20

TRTree of Thoughts: Deliberate Problem Solving with Large Language ModelsIncluded
SESelf-Consistency Improves Chain of Thought Reasoning in Language ModelsIncluded
LELet's Verify Step by StepIncluded
CHChain-of-Thought Prompting Elicits Reasoning in Large Language ModelsIncluded
LALarge Language Models are Zero-Shot ReasonersIncluded
LELeast-to-Most Prompting Enables Complex Reasoning in Large Language ModelsIncluded
REReAct: Synergizing Reasoning and Acting in Language ModelsIncluded
PAPAL: Program-aided Language ModelsIncluded
STSTaR: Bootstrapping Reasoning With ReasoningIncluded
REReflexion: Language Agents with Verbal Reinforcement LearningIncluded
SESelf-Refine: Iterative Refinement with Self-FeedbackIncluded
GRGraph of Thoughts: Solving Elaborate Problems with Large Language ModelsIncluded
TRTraining Verifiers to Solve Math Word ProblemsIncluded
SHShow Your Work: Scratchpads for Intermediate Computation with Language ModelsIncluded
SOSolving Quantitative Reasoning Problems with Language ModelsIncluded
FAFaithful Chain-of-Thought ReasoningIncluded
SCScaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model ParametersIncluded

Graph statistics

Papers: 20
Entities: 51
Relations: 50
New entries: +4

Ask this Library

Search across every entity in the Library Wiki. Answers link to entities and source papers.

Open chat

SciArk Background tasks Ddemo

LLM Reasoning ⌄20 papers · Wiki ready

Self-consistency or Tree of Thoughts: which is better for math reasoning?

Search · 1 tool call · 8s

It depends on the problem. Self-consistency samples several reasoning chains and votes on the answer. It is simple to implement and improves results on arithmetic benchmarks such as GSM8K.

Tree of Thoughts uses search and backtracking. It suits trial-and-error tasks such as Game of 24, but needs many more model calls.

self-consistencytree-of-thoughtsgsm8k

Model: kimi-k3

Can process reward models work with either approach?

Search · 1 tool call · 6s

Yes, and they often do. A process reward model can score reasoning chains for weighted voting in self-consistency, or evaluate states to choose which Tree of Thoughts branches to explore.

process-reward-modelself-consistency

Model: kimi-k3

Which papers in this Library report results on GSM8K?

Search · 1 tool call · 5s

Chain-of-Thought, Zero-Shot Reasoners, Self-Consistency, PAL, and Least-to-Most. The dataset comes from Training Verifiers. Let's Verify Step by Step and Test-Time Compute use the harder MATH benchmark.

gsm8kmath-benchmark

Ask this Library Wiki…Send

LLM WIKI

Beyond RAG.
Start with the whole paper.

SciArk connects full papers with your notes, mapping concepts, methods, and citations. New papers are compared with your Library.

Incremental update · 3 papers20 processed papers skipped

Tree of Thoughts: Deliberate Problem Solving with Large Language ModelsParsing✓ Done
Self-Consistency Improves Chain of Thought Reasoning in Language ModelsParsing✓ Done
Let's Verify Step by StepParsing✓ Done

12:04:11 Concepts · tree search, self-evaluation

12:04:13 Citations · Tree of Thoughts → Chain-of-Thought

12:04:16 Methods · sample multiple paths, then majority vote

12:04:19 Compare · checking against 20 Library papers

12:04:22 Wiki · 4 entries added, 7 updated

12:04:11 Concepts · tree search, self-evaluation

12:04:13 Citations · Tree of Thoughts → Chain-of-Thought

12:04:16 Methods · sample multiple paths, then majority vote

12:04:19 Compare · checking against 20 Library papers

12:04:22 Wiki · 4 entries added, 7 updated

Cancel anytimeView live logs →

1 : 1

One Library. One Wiki.

Ask questions within your Library’s knowledge.

Full text

Read deeply. Connect broadly.

Extract concepts and methods, map citations across papers, and uncover new connections.

+ New

Only process what is new.

Track progress, inspect live logs, or cancel anytime in the Task Center.

Photos

Camera

Weather

Mail

Maps

Notes

Clock

SciArk

Files

Podcasts

Settings

Calendar

Tuesday, October 14

Today

4 Interest Points · Scheduled search at 08:03

32new papers

5highlights

2to review

HighlightsSee all

Let's Verify Step by Step

Hunter Lightman et al. · 2023 · arXiv

Tree of Thoughts: Deliberate Problem Solving with Large Language Models

Shunyu Yao et al. · 2023 · arXiv

Self-Consistency Improves Chain of Thought Reasoning in Language Models

Xuezhi Wang et al. · 2022 · arXiv

Your review

Add 3 papers to “LLM Reasoning”

Add

Update Wiki with 3 new papers

Start

TodayInterest PointsLibraryProfile

9:41

SciArk · Scheduled search

LLM Reasoning

08:03

4 of 5 sources searched32 new papers

5 highlights in your Candidate Pool

View

RESEARCH BRIEF

Research keeps moving.
Even before you ask.

Agents track your Interest Points and highlight key research updates. They pause when decisions need your judgment.

SciArk · Overlap alertJust now

1 new paper overlaps with your work. Worth a look today.

SciArk · Scheduled search08:03

5 papers worth a look are now in your Candidate Pool.

SciArk · Conference reminderYesterday

The deadline is 12 days away. Your submission materials are ready for review.

KEEP RESEARCH MOVING

From a new finding to work you finish together.

Build on your papers, ideas, and shared knowledge.

Support through the writing process

Develop experiments, outlines, and drafts, compile PDFs, and revise with the literature in view.

A shared space for research

Share papers, ideas, and progress in one project.

Ideas that find their way to you

Connect new findings with your interests and shape ideas into testable directions.

Start researching → Explore recent updates →

Your research deserves
more than a conversation.

Capability	General-purpose assistants	SciArk
Paper discovery	Repeat your question, then sift through results.	Agents search open academic sources. Choose papers from the Candidate Pool or import your own into one Library.
Lasting knowledge	Context can fade between conversations.	Full papers and reading notes become lasting knowledge for your project.
How work moves	Work advances only when you ask.	Briefs, overlap alerts, and paper comparisons keep you informed. Agents pause for your judgment.
Team continuity	New members start without shared research context.	PIs follow group progress. Papers and ideas stay when students graduate.

Let agents find papers, read the details, and follow new work.
You decide what is worth reading, and where to go next.

“General-purpose assistants” refers to AI assistants built primarily for conversation. This compares capability types, not specific products.

Three principles.
In plain sight.

Read our privacy policy →

No tracking

This website uses no third-party analytics or tracking scripts, and no cookies.

Your choice

Papers arrive in the Candidate Pool first. You choose what to keep and which research directions to pursue.

Respect access rights

SciArk does not bypass paywalls or access controls. It uses only the access you already have permission to use.

Begin your next chapter.

Create your first Library. Discover papers, exchange ideas, and move your research forward together.

Get started free → Read the guide

SciArk on Xiaohongshu

Account ID: 1481499625

Follow the account to join the community

Feishu community

Help, feedback, and event codes

Research moves forward.
Your judgment leads.

From a research interest to an answer.

Let agents search. You choose what matters.

Turn papers into knowledge you can question.

Keep asking. Check every citation.

Beyond RAG.
Start with the whole paper.

One Library. One Wiki.

Read deeply. Connect broadly.

Only process what is new.

Research keeps moving.
Even before you ask.

From a new finding to work you finish together.

Support through the writing process

A shared space for research

Ideas that find their way to you

Your research deserves
more than a conversation.

Three principles.
In plain sight.

No tracking

Your choice

Respect access rights

Begin your next chapter.

Research moves forward.Your judgment leads.

From a research interest to an answer.

Let agents search. You choose what matters.

Turn papers into knowledge you can question.

Keep asking. Check every citation.

Beyond RAG.Start with the whole paper.

One Library. One Wiki.

Read deeply. Connect broadly.

Only process what is new.

Research keeps moving.Even before you ask.

From a new finding to work you finish together.

Support through the writing process

A shared space for research

Ideas that find their way to you

Your research deservesmore than a conversation.

Three principles.In plain sight.

No tracking

Your choice

Respect access rights

Begin your next chapter.

Research moves forward.
Your judgment leads.

Beyond RAG.
Start with the whole paper.

Research keeps moving.
Even before you ask.

Your research deserves
more than a conversation.

Three principles.
In plain sight.