Finding relevant papers for literature review takes a lot of time.
So, MIT researchers built Undermind, an AI-powered search engine.
It can generate well-researched overviews on any topic and is 10-50 times better than Google Scholar.
Here's how to use it:
This tool is hands down the MOST powerful AI literature search engine (developed by MIT grads)
It retrieved 70% of the studies I included in my systematic review (+1 new study I had missed)
All by just entering my review question!
------------------------------------------------
๐ฌ Want me to create a video showing each step in detail?
Comment 'Video' & I'll be sure to let you know when I do!
------------------------------------------------
You can ask questions of up to 1000 characters in a single query.
And it helps you get super specific with your input
โคด so it can give you super relevant output
--------------------------------
Found this helpful? Pass it on!
๐ Retweet
--------------------------------
@UndermindAI is Google for scientific research.
It finds EXACTLY what you need..
..no matter how COMPLEX your research question is!
------------------------
โ Follow @RaziaAliani to get more useful AI in #research content (no clickbait stuff!) in your X feed.
---------------------------
#aiinresearch #scientificliterature #aisearch
Thrilled to invest in @UndermindAI YC S24, a search engine that can handle extremely complex questions for experts like research scientists & doctors.
Many YC bio-founders are paying subscribersโand investors!
Backed by @Ycombinator & @Pioneer_fund.
If you're a researcher and want to try it out, let me know!
๐@UndermindAI (YC S24) is an AI-powered search engine for scientific research. It finds extremely complex information using an AI agent.
https://t.co/0mVFzZtXpo
Congrats on the launch, @RametteJoshua and @tom_hartke!
๐ @UndermindAI launched! Elevate your research with a radically thorough AI research assistant ๐
"๐ด ๐ ๐๐๐๐โ ๐๐๐๐๐๐ ๐๐๐ ๐๐๐๐๐๐๐ฅ ๐๐ข๐๐ ๐ก๐๐๐๐ " ๐ก
๐ https://t.co/h45esOVy0K
๐ Started by two quantum physics PhDs from MIT, with decades of experience in deep research, it's like "Google for scientific research"
๐๐ก๐ฒ ๐๐ง๐๐๐ซ๐ฆ๐ข๐ง๐? โคต
โ ๐๐ซ๐จ๐๐ฅ๐๐ฆ: 50 million researchers searching for scientific literature on PubMed and Google Scholar every month, and these tools are insufficient for complex research queries, often requiring hours to sift through mostly irrelevant results.
โ ๐๐จ๐ฅ๐ฎ๐ญ๐ข๐จ๐ง: Undermind has rebuilt search from the ground up, mimicking how a human researcher carefully discovers information. Their AI agent adapts based on what it finds, providing results that are 10-50x better than traditional keyword search engines.
๐ง ๐๐ฆ๐๐ ๐ข๐ง๐ ๐ฌ๐๐ง๐๐ข๐ง๐ ๐ ๐ฌ๐ฆ๐๐ซ๐ญ ๐๐จ๐ฅ๐ฅ๐๐๐ ๐ฎ๐ ๐ญ๐จ ๐๐ข๐ง๐ ๐๐ฑ๐๐๐ญ๐ฅ๐ฒ ๐ฐ๐ก๐๐ญ ๐ฒ๐จ๐ฎ ๐ง๐๐๐: Undermind's AI research assistant finds exactly what you need, no matter how complex. It carefully reads through hundreds of papers for you, identifying the very best resources, driving crucial insights.
โฑ ๐ ๐ฎ๐ฅ๐ฅ๐ฒ ๐๐ฎ๐ญ๐จ๐ฆ๐๐ญ๐ ๐ก๐จ๐ฎ๐ซ๐ฌ ๐จ๐ ๐๐๐๐ค๐ ๐ซ๐จ๐ฎ๐ง๐ ๐ฅ๐ข๐ญ๐๐ซ๐๐ญ๐ฎ๐ซ๐ ๐ฌ๐๐๐ซ๐๐ก ๐ฐ๐ข๐ญ๐ก ๐ฎ๐ง๐ฉ๐๐ซ๐๐ฅ๐ฅ๐๐ฅ๐๐ ๐๐๐๐ฎ๐ซ๐๐๐ฒ ๐๐ง๐ ๐ช๐ฎ๐๐ฅ๐ข๐ญ๐ฒ: Radically better research, radically faster. โก
โ๏ธ ๐๐จ๐ฐ ๐ญ๐จ ๐ฎ๐ฌ๐ ๐ข๐ญ ๐ Just write down your complex research topic, describing exactly what you want, as if to a colleague. After 2-3 minutes, Undermind will give you the precisely relevant articles you need.
๐ฏ ๐ ๐ข๐ซ๐ฌ๐ญ ๐ญ๐๐ซ๐ ๐๐ญ ๐ฎ๐ฌ๐๐ซ๐ฌ: Geared toward experts like research scientists and doctors, who need to find very specific resources to solve high-stakes problems. The Undermind team aims to expand their technology into law, consulting, and other areas requiring focused research solutions.
โญ๏ธ ๐๐ซ๐ฒ ๐ข๐ญ ๐๐จ๐ซ ๐๐ซ๐๐: https://t.co/imysT5lsEr
Congrats on the launch @RametteJoshua@tom_hartke!!
@aarontay on "AI-Powered Search Tools" such as: https://t.co/UhxHG9STs6
๐๏ธ@elicitorg
๐ฌ@ConnectedPapers
โ๏ธ@UndermindAI
To explore AI-powered search & discovery tools; sharing best practices, tips, and tricks. Will you share your powerpoints with me? Thank you!๐Bonne chance!
@deedydas Reminds me of the "end of history" illusion, where every generation thinks they're done changing beliefs.
Curious what other studies fall into this similar abstract boat.
https://t.co/y53lTXseIP
https://t.co/L7dY7wa4mH
Finding the right papers for complex research topics can be frustrating. @UndermindAI systematically digs through the literature to identify the precisely relevant research, as if you had sent a PhD level scientist to do hours of searching: https://t.co/HwoKh4VfFJ
[blogged] @UndermindAI- a different type of AI agent style search optimized for high recall? https://t.co/IlfOTLYQy9 - a different spin on how LLM capabilities are used in academic search compared to those we have seen from 2021/2022 onwards
@eshear Looks like intense exercise has significant effects on both physical and cognitive reaction times. Also
combining with cognitive training in some cases.
See wide range of studies here: https://t.co/FqQPwnjHoj
Undermind has just upgraded:
1) Now covers all scientific topics
2) is 10x faster than before
3) allows free unlimited usage for individuals
Register and check out the new version here:
https://t.co/KXEw91aTBd
@ericjang11 We built a "deep search" system to handle exactly this problem - complex tasks, where you have to dig through many papers on ArXiv. It highlights precisely relevant papers.
For your search, it found ~11 papers (hope I captured what you wanted): https://t.co/tBnW5k1jJA
@ChrSzegedy An ideal retrieval system must sequentially reason/adapt as it goes. It should deploy more compute as necessary, and continue until it finds everything.
This is already possible if you use LLMs as subcomponents. See systematic retrieval on all of Arxiv: https://t.co/JIOMUBDwvs
@jerryjliu0 Great review! You might also want to check out this new literature search tool we built which can autonomously gather precise papers on complex topics on demand.
Just tried using it to list out the relevant works on RAG research: https://t.co/TKGEhyJBjV
Physicists donโt need a chatbot.
To make new scientific discoveries, we need careful, thorough, analysis of everything ever published in our field.
Hereโs how weโll build it with LLMs:
https://t.co/dWsUTCjr7i
- talk at @MIT_Physics with @RametteJoshua
We may be seeing sparks of functional AGI - by amplifying GPT-4 with memory and feedback.
However, the *quality* of these memories is crucial.
One solution is bootstrapping: feed in a knowledge graph of key facts (ie., flashcards). Here's my attempt: https://t.co/NwZNzWz58Y
3/3. Enhancing interpretability: The way the language model structures the graph indirectly reveals the language model's internal concept hierarchy. In addition, at runtime we can see what information the model accessed, and why, when it makes a decision.