Things & Thinks-Issue LXII

Things & Thinks-Issue LXII

??Research Digest

Can AI Conduct Research? Evaluating an Autonomous LLM Research Framework

What the Publication is About

This paper introduces Agent Laboratory, an AI-powered framework that automates the research process using large language models (LLMs) while allowing human oversight. It streamlines three key stages—literature review, experimentation, and report writing—to help researchers execute their ideas efficiently. Evaluations show that o1-preview produces the most useful research reports, while o1-mini excels in experimental quality. The framework also reduces research costs by 84% and achieves state-of-the-art performance on machine learning benchmarks. However, human feedback significantly improves quality, as automated evaluations tend to overestimate research effectiveness.

What It Means for the Future of Research

From my perspective, this work highlights both the promise and limitations of AI-driven research. While automation can significantly accelerate the research process, human involvement remains essential to ensure quality, originality, and alignment with scientific standards. AI-generated research still faces challenges, such as hallucinations, biases, and gaps in originality, emphasizing the need for human oversight. The study suggests that future improvements in human-AI collaboration, research validation, and automated peer review could further refine AI-assisted scientific workflows.

GPT to DeepSeek: What It Means for the Future of AI in Healthcare

What the Publication is About:


It has been about a month since the general release of DeepSeek R1. This perspective examines the emergence of DeepSeek-R1, an open-source AI model that has outperformed proprietary models like GPT-4o in reasoning, math, and coding tasks. The paper discusses how DeepSeek’s chain-of-thought (CoT) reasoning, cost-efficient training methods, and multimodal capabilities could shape AI’s role in healthcare. However, it also highlights significant gaps in applying such models to clinical settings, particularly in medical text analysis, diagnostic reasoning, and ethical considerations. The authors emphasize challenges such as data security, transparency, over-reliance on AI, and regulatory uncertainties, which must be addressed before models like DeepSeek can be widely adopted in healthcare.

What It Means for the Future of Healthcare:

This study reinforces the potential of open-source AI in medical applications, but also serves as a reality check. While DeepSeek’s advancements in explainable AI and efficient training methods are impressive, its clinical reliability remains unproven. The lack of rigorous testing in real-world medical scenarios means healthcare professionals must remain cautious about integrating such models into practice. Additionally, data privacy and regulatory compliance remain unresolved issues, particularly for AI systems operating outside institutional control. Moving forward, I see the need for collaborative frameworks between AI developers, clinicians, and regulators to ensure that AI tools in healthcare are trustworthy, ethical, and clinically validated before widespread adoption.


??Digital Healthcare News

#GenAI and #BigTech in #Healthcare

Apple launched the Apple Health Study, aiming to further understand how technology — including iPhone, Apple Watch, and AirPods — can play a role in advancing and improving physical health, mental health, and overall wellbeing. The study will be conducted in collaboration with Brigham and Women’s Hospital, a leading research hospital and a major teaching affiliate of Harvard Medical School.

Community Health Systems migrated its data systems onto a single Google Cloud data platform that is paving the way for generative AI to help manage complex data and automate administrative tasks.

Google announced its partnershop with the Institute of Women's Cancers, founded by France's cancer research and treatment center, Institut Curie, to study how AI tools can help address cancer, share science-based health information and support postdoctoral researchers with funding

Memorial Sloan Kettering Cancer Center partnered with AWS to accelerate its research with AI. The collaboration seeks to speed up drug discovery and increase activity within the health system’s startup accelerator.

NVIDIA will partner with the Arc Institute and Stanford University to launch a new foundation model, Evo 2, being touted as the largest publicly available AI model for genomic data.

Innovaccer introduced suite of AI agents for providers that are voice-activated and can communicate with patients for appointment scheduling, protocol intake, managing referrals and answering routine patient inquiries. The suite of eight AI agents also tackles tasks like prior authorization, care gap closure, hierarchical condition category coding and transitional care management

Tech in Digital Health

Corti and Wolters Kluwers announced collaboration to integrate W-K's clinical support platform UpToDate into Corti's real-time AI platform with the aim to enhance clinicians' access to credible medical data.

Regulatory Brief

Aidoc received FDA clearance for a computer-aided rib fracture triage offering (CADt AI) powered by a foundation model.

Pharma/Device Brief

Moderna is revamping its digital team, eliminating 10% of roles under two digital departments-“digital for business” and “digital core” functions, affecting about 50 employees.

Cognizant and Gilead Sciences extended their partnership for genAI

Funding, Deals, Mergers & acquisitions

Abridge, a generative AI platform for clinical conversations, raised a $250M

Harrison.ai has recently raised $112M

OpenEvidence, an AI copilot for physicians that helps them make better decisions at the point of care, raised $75M.

Rad AI, a generative AI radiology workflow company, has secured $60

Behavioral health tech company Eleos Health raised $60M

Candid Health, a revenue cycle management platform for healthcare providers, raised $52.5M

SafelyYou, a provider of AI-powered solutions for senior living communities, raised $43M

Lynx, which aims to simplify healthcare payments and administration, raised $27M

LinkedIn founder Reid Hoffman and Siddhartha Mukherjee, cancer researcher and author of the book “The Emperor of All Maladies,” co-founded an AI-powered drug discovery startup, Manas AI. The company, which will initially focus on breast cancer, prostate cancer, and lymphoma, raised $24.6M

Consumer Digital Health & Other News

Teladoc posted $1B loss in 2024 as BetterHelp lagged., the d2c mental health segment’s adjusted earnings before interest, taxes, depreciation and amortization was $77.8 million last year, down 43% from 2023. On he other hand, it announced acquiring virtual preventative care company Catapult Health for $65M.

Athenahealth rolled out Abridge AI scribe to community practices, hospitals


??Longread of the Month

This paper by Jean Dreze and Anfaz Abdul Wahab is a great deep-dive in India's national and state-level health insurance schemes

??Tweet of the Month

This is an interesting use-case for GPT o1.....

??Chart of the Month

This is a good graph depicting AI's energy footprint via Nature

h/t Eric Topol

Liked what you read? Subscribe & Share! I would love to hear your feedback and thoughts. You can also connect with me via Twitter and LinkedIn!

要查看或添加评论,请登录

Santosh Shevade的更多文章

  • Digital Health Goes to Market: The Journey Five Years Later (Part 1)

    Digital Health Goes to Market: The Journey Five Years Later (Part 1)

    Five years ago, I wrote two blogs examining the wave of digital health IPOs that took place between 2019 and 2021. At…

    5 条评论
  • Innovators & Economists: A Tale of Two AI Reports

    Innovators & Economists: A Tale of Two AI Reports

    Two reports caught my eye recently. On one side, there's a Google Cloud report-'321 real-world gen AI use cases from…

    2 条评论
  • Things & Thinks-Issue LXI

    Things & Thinks-Issue LXI

    ??Research Digest AI in Hospitals: Are Predictive Models Being Tested for Accuracy and Bias? What the publication is…

  • Reading about GenAI in Healthcare: From Academic Tropes via Preprint Dreams to Industry Realities

    Reading about GenAI in Healthcare: From Academic Tropes via Preprint Dreams to Industry Realities

    As someone who spends a significant chunk of time reading academic articles, skimming preprints, and diving into…

    2 条评论
  • Things & Thinks-Issue LX

    Things & Thinks-Issue LX

    ??Research Digest Testing and Evaluation of Health Care Applications of Large Language Models This Systematic Review…

    3 条评论
  • Things & Thinks-Issue LIX

    Things & Thinks-Issue LIX

    ??Research Digest PAM Chat and The Role of AI-Generated Drafts What is this about This study, titled Utility of…

  • I used NotebookLM to create my first AI Generated Podcast!

    I used NotebookLM to create my first AI Generated Podcast!

    I recently started using NotebookLM, a new AI tool from Google, and it is indeed quite an interesting. What follows…

    4 条评论
  • Things & Thinks-Issue LVIII

    Things & Thinks-Issue LVIII

    ??Research Digest- Enhanced Transformer for Health Outcome Simulation (ETHOS) What the publication is about: The paper…

    1 条评论
  • Things & Thinks-Issue LVII

    Things & Thinks-Issue LVII

    ??Digital Healthcare News- #GenAI and #BigTech in #Healthcare Epic's MyChart in-basket augmented response technology…

    6 条评论
  • Things & Thinks-Issue LVI

    Things & Thinks-Issue LVI

    ??Research Digest- High Accuracy, Hidden Errors This article, "Hidden flaws behind expert-level accuracy of multimodal…

    2 条评论

社区洞察