Here’s your biweekly AI in Healthcare and Digital Health video brief on data-use policy, point-of-care AI, scientific benchmarks, R&D tie-ups, organizational build-outs, and funding trends
🎯 Watch Our Video Summary Capturing AI in Healthcare and Digital Health News from the Last Two Weeks
Dive deeper
🗓️ Explore weekly details and sources
Week 3–9 November 2025
Week 10–16 November 2025
📚 Find your one-stop page for the full AI and Digital Health archive.
Top Stories Covered In This Video
Chapters
0:00 Introduction
0:09 Language models struggle to distinguish belief from knowledge on KaBLE, with large drops on first-person false-belief tasks
0:56 Eli Lilly expands XtalPi collaboration via Ailux in bispecific-antibody engineering deal worth up to 345 million dollars
1:21 Insilico unveils AI-designed cardiometabolic portfolio, including oral GLP-1RAs and NLRP3 inhibitor ISM8969
2:09 American College of Cardiology partners with OpenEvidence to deliver AI-powered, guideline-based point-of-care support
2:30 Michael Burry alleges hyperscalers understate depreciation to boost AI-era earnings; CNBC unable to confirm
3:14 European Commission draft would shift cookies into GDPR and permit AI training on personal data under legitimate interest
3:45 Pfizer job posting signals PDURS Center of Excellence to scale drug-digital solutions and FDA engagement
4:07 Reports indicate OpenAI is developing a consumer health assistant and health data aggregator
4:30 American Medical Association launches Center for Digital Health and AI to shape policy, workflow fit, and training
4:50 How to reach us
Transcript
Welcome to the latest edition of AI in Life Sciences Updates, covering breakthroughs in the past two weeks. Brought to you by LucidQuest.
Nature Machine Intelligence reports a 24-model evaluation on the 13,000-question KaBLE benchmark. Models struggled to separate belief from knowledge, especially for first-person false beliefs. GPT-4o fell from 98.2% to 64.4% on those items, while DeepSeek R1 dropped to 14.4%. Newer models hit about 95% on third-person false beliefs versus 62.6% on first-person, revealing attribution bias and inconsistent reasoning strategies. The authors say most models lack a robust grasp that knowledge requires truth.
Eli Lilly expanded its XtalPi collaboration via subsidiary Ailux in a bispecific-antibody pact worth up to 345 million dollars. Lilly will make double-digit-million upfront and near-term payments, can nominate undisclosed target pairs, and may license Ailux’s engineering platform for internal use.
Insilico Medicine unveiled eight oral cardiometabolic candidates from its Pharma.AI platform, including oral GLP-1 receptor agonists with improved safety and pharmacokinetics, a brain-penetrant NLRP3 inhibitor known as ISM8969, and programs against NR3C1, GIPR, Amylin, APJ, and Lp(a). Two oral GLP-1RAs, daily and once-weekly, are at or near preclinical candidate stage. Insilico cites 22 preclinical programs and 12–18 month discovery cycles, plus clinical assets Rentosertib and ISM5411.
The American College of Cardiology partnered with OpenEvidence to embed ACC-sourced guidelines into generative-AI decision support at the point of care. Plans include an expert work group, education materials, an AI Resource Center, and hands-on demos at ACC’s Future Hub.
CNBC reports investor Michael Burry alleges hyperscalers are inflating AI-era earnings by extending server useful lives, understating 2026–2028 depreciation by about 176 billion dollars. He flagged potential 2028 impacts at Oracle and Meta, 27% and 21% respectively. CNBC could not independently confirm. Burry’s filings showed puts against Nvidia and Palantir as of September 30, with notional values of 187 million and 912 million dollars.
According to a leaked draft summarized by Computerworld, the European Commission plans a “Digital Omnibus” to move cookies into GDPR via Articles 88a–88b, shift many uses to opt-out, enable browser-level consent signals, and allow AI training on personal data under legitimate interest with safeguards. It would narrow when data counts as sensitive and carve out consent rules for media.
A new Pfizer job posting signals a PDURS Center of Excellence. The Senior Director will translate PDURS regulatory requirements into drug-digital product, trial, and commercialization strategy, guide FDA engagement, and use cardiometabolic disease as a foundation for scale.
Reports indicate OpenAI is exploring a consumer health assistant and data aggregator, following hires of Nate Gross as head of healthcare strategy and Ashley Alexander as VP of health products. At HLTH, the company highlighted very high weekly usage for health queries. No official comment on product plans.
The American Medical Association launched the Center for Digital Health and AI to shape policy, drive workflow-ready tools, and train clinicians. AMA surveys find roughly two-thirds of physicians using AI-enabled tools in some form, with about a quarter still more concerned than excited.
Stay ahead in AI in Life Sciences research! Like, share, and subscribe for our updates. Visit www.lqventures.com
or email us at info@lqventures.com for expert healthcare consulting. See you next time!
Why It matters
-
Clinical safety: New evidence suggests LMs misread first-person beliefs, a risk for autonomous or semi-autonomous clinical advice [1].
-
Pipeline bets: Pharma is scaling AI-native discovery in antibodies and metabolism, broadening beyond small molecules [2][3].
-
Point-of-care shift: Major societies are hard-wiring guidelines into AI tools, aiming for workflow-first execution [4][9].
-
Policy turn: EU proposals could ease AI data use and cookie tracking, reshaping compliance and data supply chains [6].
-
Market signals: Investor skepticism about AI accounting intersects with capex-heavy health AI infrastructure [5].
-
Pharma digital: PDURS focus hints at drug-digital combinations moving toward regulated, reimbursable products [7].
-
Consumer front door: Big AI is circling health assistants, potentially redefining triage and navigation [8].
🗓️ Explore weekly details and sources
Week 3–9 November 2025
Week 10–16 November 2025
📚 Find your one-stop page for the full AI and Digital Health archive.
FAQ
What is KaBLE, and which failure mode mattered most?
KaBLE is a 13k-question benchmark spanning 13 epistemic tasks; first-person false-belief recognition was the consistent failure across 24 models, with attribution bias favoring third-person cases [1].
What exactly did Lilly get from Ailux?
Access to a bispecific-antibody engineering platform, the right to nominate undisclosed target pairs, and an option to license the platform internally; total deal value up to $345M with milestones [2].
Which cardiometabolic targets is Insilico pursuing?
Programs include oral GLP-1RAs, a selective NLRP3 inhibitor, and candidates against NR3C1, GIPR, Amylin, APJ, and Lp(a); some are nearing preclinical candidate stage [3].
How will ACC and OpenEvidence deliver guidance at the bedside?
By integrating ACC literature and guidelines into a generative-AI interface, plus an expert group, resource center, and ACC conference demos to refine point-of-care use cases [4].
Did CNBC confirm Burry’s depreciation claim?
No. CNBC reported the allegations and noted they were not independently confirmed; companies had not commented at the time, and details were promised later in November [5].
What GDPR changes would affect AI training?
A draft would allow AI training on personal data under “legitimate interest,” narrow sensitive-data scope, and shift cookie practices toward opt-out for defined purposes, pending formal proposal [6].
What is Pfizer’s PDURS Center of Excellence aiming to do?
Translate PDURS regulatory expertise into drug-digital product design, trials, and commercialization, starting with cardiometabolic use cases, and lead FDA engagement strategies [7].
Is OpenAI officially launching a health assistant?
Reports indicate exploration of a consumer assistant and aggregator, supported by recent health leadership hires; no formal company comment yet [8].
How will AMA’s new Center support clinicians?
By shaping policy, ensuring workflow-fit of AI tools, and providing training, aiming to boost adoption while addressing privacy and safety concerns [9].
Entities / Keywords
Eli Lilly; XtalPi/Ailux; Insilico Medicine; American College of Cardiology; OpenEvidence; Pfizer; OpenAI; American Medical Association; European Commission; GDPR; cookies; legitimate interest; GLP-1RA; NLRP3; bispecific antibodies; PDURS; clinical decision support; Michael Burry; Oracle; Meta.