Glide Through the Sea of Research

Never miss a breakthrough in Machine Learning and NLP

PaperManta uses Agentic Search to automatically discover, prioritize, and summarize the exact research papers that matter to your unique interests.

Start Your Search

How It Works

Your Personal AI Research Assistant

Stop drowning in endless publications. Let PaperManta find the work that is important for you.

1

Describe what you're building or researching

Enter a plain-text description about your specific niche (e.g., "Explainable architectures for LLMs").

2

Get a curated, summarized list

PaperManta scans thousands of papers, finding the ones most relevant to your query and providing personalised, easy-to-read summaries.

3

Wake up to new breakthroughs

Our system runs your queries daily, finding the latest uploads from arXiv and top conferences, so you are always the first to know.

Why PaperManta?

See how PaperManta compares to other ways of keeping up with the latest research.

Google Scholar alerts arXiv alerts Chatbots AI Agents PaperManta
Specify exactly what to search ✗ ✗ ✓ ✓ ✓
Searches automatically & periodically ✓ ✓ ✗ ✗ ✓
Personalised summaries ✗ ✗ ✓ ✓ ✓
Relevance ranking & prioritisation ✗ ✗ ✓ ✓ ✓
Predictable, fixed cost ✓ ✓ ✓ ✗ ✓
Only real papers
(no hallucinations)		 ✓ ✓ ✗ ✗ ✓

Live Research Queries

Explore what our AI is uncovering right now for some of the most popular NLP and ML topics.

Parameter-Efficient Fine-Tuning (PEFT) in Large Language Models

QLoRA: Efficient Finetuning of Quantized LLMs

This paper introduces QLoRA, a highly memory-efficient finetuning approach that allows a 65B parameter model to be finetuned on a single 48GB GPU while preserving full 16-bit finetuning task performance. The authors combine 4-bit NormalFloat quantization, double quantization, and paged optimizers to dramatically reduce memory footprint without sacrificing model quality.

View Full Results List

Retrieval-Augmented Generation for Specialized Domains

Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection

Proposes a new framework where the language model learns to dynamically retrieve information on demand and self-reflect on retrieved passages and its own generations, improving factual accuracy across a range of knowledge-intensive tasks and reducing hallucinations compared to standard RAG pipelines.

View Full Results List

Mixture-of-Experts Architectures for Efficient Scaling

Mixtral of Experts: Sparse Mixture-of-Experts Language Models

Introduces a sparse mixture-of-experts model that activates only a fraction of its total parameters per token, achieving the quality of a much larger dense model while dramatically reducing inference cost. The paper details routing strategies, load balancing techniques, and shows strong results on reasoning, code, and multilingual benchmarks.

View Full Results List

Long-Context Transformers and Efficient Attention

Ring Attention with Blockwise Transformers for Near-Infinite Context

Presents a distributed attention mechanism that partitions the sequence across devices and overlaps communication with computation, enabling context lengths far beyond what a single accelerator can hold. Demonstrates practical training and inference at multi-million token contexts with linear memory scaling in the number of devices.

View Full Results List

Diffusion Models for Text Generation

SEDD: Score Entropy Discrete Diffusion for Language Modeling

Extends score-based diffusion to discrete text spaces by introducing a novel score-entropy loss that stabilizes training on discrete tokens. The resulting model matches or surpasses autoregressive language models of similar size on perplexity while enabling parallel non-autoregressive sampling and controllable generation.

View Full Results List

Choose Your Plan

Whether you are casually browsing or doing intensive R&D, we have a plan for you.

PaperManta Basic Full Capabilities PaperManta Pro
Results Limit Up to 8 highly relevant papers Unlimited papers per topic
Paper Sources Conference proceedings Conference proceedings + daily arXiv releases
Concurrent Queries 1 Static Search 3 Active Running Searches
Automated Updates No (One-time search only) Daily automated updates
Personalised Summaries Included Included
Price $0 $12 / month
Start Free Join Pro

Frequently Asked Questions

Everything you need to know about how PaperManta works.

What fields of research do you support?
We are currently highly specialized in Machine Learning (ML), Artificial Intelligence (AI), and Natural Language Processing (NLP), with more disciplines planned in the future.
Which conference proceedings do you search?
PaperManta currently covers the following conferences: AAAI, AACL, ACL, CL, COLING, CoNLL, EACL, EMNLP, Findings (ACL), ICLR, ICML, NAACL, NeurIPS, TACL. We are analysing papers published at 2022 or later at these conferences.
Do you cover the whole of arXiv?
Given the high volume of papers submitted to arXiv every day, we are currently focusing on the categories of cs.CL, cs.AI, cs.LG. PaperManta analyses the new papers posted to these categories every day and selects the ones that are most relevant to your topic of interest. Some older papers may still show up in the results, for example when authors post an updated version or when the system has already seen an older paper previously and finds it particularly relevant.
How does the daily update work on the Pro plan?
Once you set up a query, PaperManta acts as a radar. Every 24 hours, it scans new publications from conferences and arXiv. It then adds relevant, newly published papers to your query dashboard with fresh summaries.
How does the search work on the Free plan?
You describe a topic to set up a query. PaperManta will then search for relevant papers until it has analysed through a sufficient number of publications or until it has found 7 highly relevant papers. On the free plan, the search finishes and doesn't automatically keep searching for the newest papers. Also, the search is limited to only conference proceedings and arxiv isn't covered.
Can I change my active queries?
Yes! Pro users can have 3 active queries running simultaneously and can pause, delete, or swap them out at any time.
Is there a risk of hallucinated papers?
The paper titles, conferences and author lists are not generated by AI. They come from a controlled database so the system is not able to invent any non-existent papers.