Elicit

We’ve seen some of Elicit’s competitors rush ahead with the deployment of the DeepSeek’s R1 model and we’d like to explain why we’ve been a bit more cautious so far. DeepSeek’s models align closely with narratives spread by the Chinese Communist Party (CCP). While we respect people’s First Amendment right to engage or share such material, we believe that any government propaganda is fundamentally incompatible with Elicit’s mission to be maximally truth-seeking. Attached are examples of DeepSeek's V3 model defending the Tiananmen Square massacre. And another example where V3 says that China is a democracy. ChatGPT makes no such mistake. Even when deployed within Perplexity, R1 has clear pro-CCP bias compared to OpenAI's o1. While R1 is a stunning technical achievement and we have deep respect for the people behind it, the examples above show that without further post-training and evaluation, DeepSeek’s models are not suitable for deployment in high-stakes settings such as academic research. Given the open-source nature of these models we’re confident that the community will be able to address these issues and bring high-quality affordable reasoning to all Elicit users.

“Elicit is a step above other tools I've tried. I prefer Elicit when it comes to actually interpreting evidence. It doesn't make things up like ChatGPT.” - James Compagno, Director of Marketing MicroGenDX is a clinical laboratory that offers microbial diagnostic tests via next-generation DNA sequencing (NGS). The company has over 70 peer-reviewed publications and has processed over 2.5 million molecular samples. The Biostatistics, Medical Affairs, and Marketing teams all use Elicit to do much deeper lit reviews, better validate clinical applications of next-generation microbial DNA tests, and get these diagnostics to more patients. Read the case study: https://lnkd.in/g36KzjZB

Stay tuned for some big things from this design team.

Research Residency at Elicit (Q1 2025) We're building an epistemically sound research agent that can leverage unlimited test-time compute while maintaining transparent and verifiable reasoning The challenge: Transformers do fixed computation per token & rapidly diverge/converge when applied iteratively. We want operators that reliably improve a knowledge state over 1000s of iterations Think git for reasoning, where each commit needs to improve the repo. The repo / knowledge state = structured info (claims, evidence, reasoning), scratchpads, search trees. Scientific papers broken down into components that can be iteratively refined Goal: Like scientists, we want LLMs to make genuine intellectual progress - separating evidence from inference, finding connections, building clearer explanations. Each step must maintain local validity - no "trust me, it works out in the end" RL Looking for: Experience w/ LLMs, intuitions about systematic reasoning, care about AI transparency. Strong software engineering background & prior experience creating novel abstractions for automated reasoning are big pluses 3-month contract, $12-15k/month, Oakland/remote US. Details & apply: https://lnkd.in/e2sDBUve

In 2023, I discovered Elicit, a tool recommended by Dr. Andrew Stapleton that has revolutionised how I explore research questions.   Elicit lets you ask a question 🤔 and provides a concise summary of top results from peer-reviewed scientific literature 📚, complete with proper citations. 🌟 When I first stumbled upon it, Elicit was still in its early stages. Over time, it has evolved into a highly reliable resource. ✅ 🔍 It's a far more trustworthy alternative to popular web search engines, which often return unreliable or misleading results. If you're passionate about discovering credible, science-backed information, I highly recommend giving Elicit a try...!

🧱🔨 How are you using today’s tools to help chip away at long-standing walls in education? One of the most exciting uses of #AI that I’ve found with my students is the combination of Elicit with #NotebookLM, because it genuinely seems to help remove a barrier to learning that I’ve have seen since I started teaching 18years ago! My vocational HND learners have always struggled with researching from academic journals because of the style of language used in them. Now, supported by these two AI tools, they can find relevant research quickly and build a portfolio of relevant sources in NotebookLM that they can then interrogate and have explained to them. There are more ways that I’m using AI with my students in my most recent blog (link in the comments), but please share below what you’re doing too! #AmplifyFE #HEinFE #AIinEducation #EdTechUK #LoveOurColleges #GoogleEI

Research Residency at Elicit (Q1 2025) We're building an epistemically sound research agent that can leverage unlimited test-time compute while maintaining transparent and verifiable reasoning The challenge: Transformers do fixed computation per token & rapidly diverge/converge when applied iteratively. We want operators that reliably improve a knowledge state over 1000s of iterations Think git for reasoning, where each commit needs to improve the repo. The repo / knowledge state = structured info (claims, evidence, reasoning), scratchpads, search trees. Scientific papers broken down into components that can be iteratively refined Goal: Like scientists, we want LLMs to make genuine intellectual progress - separating evidence from inference, finding connections, building clearer explanations. Each step must maintain local validity - no "trust me, it works out in the end" RL Looking for: Experience w/ LLMs, intuitions about systematic reasoning, care about AI transparency. Strong software engineering background & prior experience creating novel abstractions for automated reasoning are big pluses 3-month contract, $12-15k/month, Oakland/remote US. Details & apply: https://lnkd.in/e2sDBUve

Research Residency at Elicit (Q1 2025) We're building an epistemically sound research agent that can leverage unlimited test-time compute while maintaining transparent and verifiable reasoning The challenge: Transformers do fixed computation per token & rapidly diverge/converge when applied iteratively. We want operators that reliably improve a knowledge state over 1000s of iterations Think git for reasoning, where each commit needs to improve the repo. The repo / knowledge state = structured info (claims, evidence, reasoning), scratchpads, search trees. Scientific papers broken down into components that can be iteratively refined Goal: Like scientists, we want LLMs to make genuine intellectual progress - separating evidence from inference, finding connections, building clearer explanations. Each step must maintain local validity - no "trust me, it works out in the end" RL Looking for: Experience w/ LLMs, intuitions about systematic reasoning, care about AI transparency. Strong software engineering background & prior experience creating novel abstractions for automated reasoning are big pluses 3-month contract, $12-15k/month, Oakland/remote US. Details & apply: https://lnkd.in/e2sDBUve

Y aquí mi mentorizado Alejandro Boix Català de ODEC, dándonos algunas pistas prácticas sobre el uso de IA en investigación de mercados: Elicit, para analizar artículos de investigación a una velocidad sobrehumana Al Día con Insights + Analytics España

Every week, Andreas Stuhlmüller kicks off our team sync with a view on "Elicit at 30,000 feet." Last Friday, we talked about Exa's websets and what it means to scale up inference-time compute. With performance gains from inference-time compute, we'll switch from humans calling models to models calling humans. "If you can run for longer without human intervention and keep doing useful work, you don’t need to pause all the time." We'll need to develop new interaction paradigms for models calling humans, and train models to know when to call humans to avoid getting off-track. Full notes: