Yuri Quintana
@yuriquintana.com
Chief Clinical Informatics @ BIDMC & Faculty Harvard Medical School, Chief of @DCINetwork.org #MedicalInformatics #AI #DigitalHealth #MobileHealth #LearningHealthSystems
https://www.LinkedIn.com/in/yuriquintana
http://www.yuriquintana.com
https://www.LinkedIn.com/in/yuriquintana
http://www.yuriquintana.com
Pinned
Call for Presentations: DCI Network 2026 Conference on Patient-Powered AI in Healthcare, June 22-24 at Harvard. Submit research on AI engagement, patient co-design, ethics, and real-world impact. Published in PubMed-indexed proceedings. www.dcinetwork.org/patients2026...
Oracle says 57% of its $523B backlog comes from OpenAI, projecting $74B losses by 2028. Nvidia invests $100B, OpenAI commits $300B to Oracle, Oracle buys Nvidia chips, a $610B circular flow. In the dotcom bubble Lucent collapsed at 24% vendor financing; Nvidia’s exposure is 67%.
December 14, 2025 at 8:43 AM
Oracle says 57% of its $523B backlog comes from OpenAI, projecting $74B losses by 2028. Nvidia invests $100B, OpenAI commits $300B to Oracle, Oracle buys Nvidia chips, a $610B circular flow. In the dotcom bubble Lucent collapsed at 24% vendor financing; Nvidia’s exposure is 67%.
NIH received strong public feedback on its proposal to limit publication costs. Universities and societies favored flexible award level caps over strict per article limits and urged protecting open access and research quality. grants.nih.gov/grants/guide...
NOT-OD-25-138: Request for Information on Maximizing Research Funds by Limiting Allowable Publishing Costs
NIH Funding Opportunities and Notices in the NIH Guide for Grants and Contracts: Request for Information on Maximizing Research Funds by Limiting Allowable Publishing Costs NOT-OD-25-138. NIH
grants.nih.gov
December 13, 2025 at 3:05 PM
NIH received strong public feedback on its proposal to limit publication costs. Universities and societies favored flexible award level caps over strict per article limits and urged protecting open access and research quality. grants.nih.gov/grants/guide...
FDA researchers developed AnimalGAN, an AI model trained on 8000 rats and 138 drugs that simulates clinical pathology without animal tests. It achieved up to 96 percent agreement with real studies and enables large scale digital drug safety screening doi.org/10.1038/s414...
A generative adversarial network model alternative to animal studies for clinical pathology assessment - Nature Communications
Generative AI has the potential to transform the way chemical and drug safety research is conducted. Here the authors show AnimalGAN, a model developed using Generative Adversarial Networks, which sim...
doi.org
December 13, 2025 at 2:40 PM
FDA researchers developed AnimalGAN, an AI model trained on 8000 rats and 138 drugs that simulates clinical pathology without animal tests. It achieved up to 96 percent agreement with real studies and enables large scale digital drug safety screening doi.org/10.1038/s414...
National Academy of Medicine launches AI patient safety initiative to develop harm prevention frameworks across healthcare systems www.beckershospitalreview.com/healthcare-i...
www.beckershospitalreview.com
December 12, 2025 at 12:17 PM
National Academy of Medicine launches AI patient safety initiative to develop harm prevention frameworks across healthcare systems www.beckershospitalreview.com/healthcare-i...
Adoption of AI assistants and public health AI strategies abroad reinforce that patients will increasingly interact with AI in both clinical and population health contexts, raising questions of trust, transparency, and access. www.businessinsider.com/included-hea...
This startup is launching an AI personal health assistant that'll face off with Big Tech from Verily to OpenAI
The startup's new tech could compete with Big Tech's health AI projects from heavyweights like Alphabet and OpenAI.
www.businessinsider.com
December 12, 2025 at 12:11 PM
Adoption of AI assistants and public health AI strategies abroad reinforce that patients will increasingly interact with AI in both clinical and population health contexts, raising questions of trust, transparency, and access. www.businessinsider.com/included-hea...
Measuring Agents in Production: arxiv.org/abs/2512.04123
Berkeley’s study shows successful AI agents are not autonomous. Sixty-eight percent run fewer than ten steps and seventy-four percent require human verification. Most teams build simple, tightly controlled agents that reliably deliver value.
Berkeley’s study shows successful AI agents are not autonomous. Sixty-eight percent run fewer than ten steps and seventy-four percent require human verification. Most teams build simple, tightly controlled agents that reliably deliver value.
Measuring Agents in Production
AI agents are actively running in production across diverse industries, yet little is publicly known about which technical approaches enable successful real-world deployments. We present the first lar...
arxiv.org
December 11, 2025 at 4:25 PM
Measuring Agents in Production: arxiv.org/abs/2512.04123
Berkeley’s study shows successful AI agents are not autonomous. Sixty-eight percent run fewer than ten steps and seventy-four percent require human verification. Most teams build simple, tightly controlled agents that reliably deliver value.
Berkeley’s study shows successful AI agents are not autonomous. Sixty-eight percent run fewer than ten steps and seventy-four percent require human verification. Most teams build simple, tightly controlled agents that reliably deliver value.
NAACP release a 75-page report Building a Healthier Future urging an equity-first approach to AI in healthcare, warning that unchecked AI tools risk amplifying biases and disparities. Proposes bias audits, transparency, governance councils, and community partnerships.
www.reuters.com/business/hea...
www.reuters.com/business/hea...
Exclusive: NAACP pressing for ‘equity-first’ AI standards in medicine
The NAACP unveiled a blueprint on Wednesday aimed at preventing artificial intelligence from deepening racial inequities in U.S. healthcare, calling for bias audits and “equity-first” standards as hos...
www.reuters.com
December 11, 2025 at 12:59 PM
NAACP release a 75-page report Building a Healthier Future urging an equity-first approach to AI in healthcare, warning that unchecked AI tools risk amplifying biases and disparities. Proposes bias audits, transparency, governance councils, and community partnerships.
www.reuters.com/business/hea...
www.reuters.com/business/hea...
Google’s Gemini 3 proves TPU power, raising big questions for NVIDIA and xAI. Gavin Baker says xAI will likely train the first Blackwell model in 2026 as Elon builds data centers fastest, helping NVIDIA debug at scale. GB300’s rollout could reshape Google’s strategy. x.com/patrick_osha...
x.com
December 10, 2025 at 1:04 AM
Google’s Gemini 3 proves TPU power, raising big questions for NVIDIA and xAI. Gavin Baker says xAI will likely train the first Blackwell model in 2026 as Elon builds data centers fastest, helping NVIDIA debug at scale. GB300’s rollout could reshape Google’s strategy. x.com/patrick_osha...
Cancer blood tests are rising fast—but evidence isn’t. Studies show limited proof they save lives, especially in low-risk adults 50+. False positives, unclear benefits, and costly workups remain major concerns. More: NYT + Topol review nyti.ms/4h4jJfF
December 10, 2025 at 12:56 AM
Cancer blood tests are rising fast—but evidence isn’t. Studies show limited proof they save lives, especially in low-risk adults 50+. False positives, unclear benefits, and costly workups remain major concerns. More: NYT + Topol review nyti.ms/4h4jJfF
Meta’s new paper shows a VLM can train its own judge—no humans, no labels. By generating good vs. flawed answers, the model learns to score responses and even beats much larger judges. A big shift for multimodal AI. arxiv.org/abs/2512.05145
Self-Improving VLM Judges Without Human Annotations
Effective judges of Vision-Language Models (VLMs) are crucial for model development. Current methods for training VLM judges mainly rely on large-scale human preference annotations. However, such an a...
arxiv.org
December 10, 2025 at 12:45 AM
Meta’s new paper shows a VLM can train its own judge—no humans, no labels. By generating good vs. flawed answers, the model learns to score responses and even beats much larger judges. A big shift for multimodal AI. arxiv.org/abs/2512.05145
LLMs can now construct full causal models by extracting, testing, and iteratively refining causal graphs, counterfactuals, and interventions directly from their latent world knowledge. A breakthrough showing AI can move beyond patterns to true causal reasoning.
arxiv.org/abs/2512.07796
arxiv.org/abs/2512.07796
Large Causal Models from Large Language Models
We introduce a new paradigm for building large causal models (LCMs) that exploits the enormous potential latent in today's large language models (LLMs). We describe our ongoing experiments with an imp...
arxiv.org
December 10, 2025 at 12:33 AM
LLMs can now construct full causal models by extracting, testing, and iteratively refining causal graphs, counterfactuals, and interventions directly from their latent world knowledge. A breakthrough showing AI can move beyond patterns to true causal reasoning.
arxiv.org/abs/2512.07796
arxiv.org/abs/2512.07796
HHS selects C3 AI to build a platform unifying NIH, Medicare/Medicaid, and registry data — a foundation for large-scale public health AI initiatives www.healthcareitnews.com/news/hhs-tap...
HHS taps C3 for interagency AI data platform
The department plans to use the company's agentic artificial intelligence to integrate disease-specific NIH data with Medicare, Medicaid, claims and state registry datasets to improve data quality and...
www.healthcareitnews.com
December 9, 2025 at 12:19 PM
HHS selects C3 AI to build a platform unifying NIH, Medicare/Medicaid, and registry data — a foundation for large-scale public health AI initiatives www.healthcareitnews.com/news/hhs-tap...
The EU and Canada just strengthened their digital cooperation, with a focus on AI, digital identity wallets, and independent media, announced during the first “Digital Partnership Council” session in Montreal.
europeansting.com/2025/12/09/e...
europeansting.com/2025/12/09/e...
European Union and Canada strengthen their digital partnership with a focus on artificial intelligence, digital identity wallets and independent media - The European Sting - Critical News & Insights o...
European Union and Canada strengthen their digital partnership with a focus on artificial intelligence, digital identity wallets and independent media The European Sting - Critical News & Insights on ...
europeansting.com
December 9, 2025 at 12:16 PM
The EU and Canada just strengthened their digital cooperation, with a focus on AI, digital identity wallets, and independent media, announced during the first “Digital Partnership Council” session in Montreal.
europeansting.com/2025/12/09/e...
europeansting.com/2025/12/09/e...
EU Digital Omnibus package proposes adjustments to how EU Artificial Intelligence Act (AI Act) . part of the EU’s effort to streamline its digital regulatory framework proposes delaying some obligations under the AI Act to give businesses more time to comply.
www.sidley.com/en/insights/...
www.sidley.com/en/insights/...
EU Digital Omnibus: The European Commission Proposes Important Changes to the EU's Digital Rulebook
www.sidley.com
December 9, 2025 at 12:12 PM
EU Digital Omnibus package proposes adjustments to how EU Artificial Intelligence Act (AI Act) . part of the EU’s effort to streamline its digital regulatory framework proposes delaying some obligations under the AI Act to give businesses more time to comply.
www.sidley.com/en/insights/...
www.sidley.com/en/insights/...
EU has launched a formal probe into Google’s use of online publishers’ content and YouTube videos to train its AI models. Regulators worry Google may be using this content without fair compensation or consent. potentially abusing its dominant position.
www.livemint.com/companies/ne...
www.livemint.com/companies/ne...
EU-Google antitrust case: European Commission opens probe against tech giant over market dominance using AI tools | Company Business News
EU-Google antitrust case: European Commission, has opened an investigation into Google, alleging that the company has abused its market dominance using artificial intelligence (AI) tools to squeeze co...
www.livemint.com
December 9, 2025 at 12:02 PM
EU has launched a formal probe into Google’s use of online publishers’ content and YouTube videos to train its AI models. Regulators worry Google may be using this content without fair compensation or consent. potentially abusing its dominant position.
www.livemint.com/companies/ne...
www.livemint.com/companies/ne...
Brookfield Asset Management (Brookfield), Qatar Investment Authority (via its AI subsidiary Qai) Team Up on a $20 Billion AI Investment Venture. The joint venture will invest in AI infrastructure in Qatar and select international markets.
www.reuters.com/technology/b...
www.reuters.com/technology/b...
Brookfield and Qatar launch $20 billion AI infrastructure joint venture
Brookfield and Qai, an artificial intelligence company owned by Qatar's sovereign wealth fund, have formed a $20 billion joint venture to develop artificial intelligence infrastructure in Qatar and se...
www.reuters.com
December 9, 2025 at 11:55 AM
Brookfield Asset Management (Brookfield), Qatar Investment Authority (via its AI subsidiary Qai) Team Up on a $20 Billion AI Investment Venture. The joint venture will invest in AI infrastructure in Qatar and select international markets.
www.reuters.com/technology/b...
www.reuters.com/technology/b...
AlphaGenome analyzes up to 1M DNA base-pairs at single-letter resolution, predicting effects on gene expression, splicing, chromatin and more. It can flag how one base change alters biology, disease risk or therapy.
deepmind.google/blog/alphage...
deepmind.google/blog/alphage...
AlphaGenome: AI for better understanding the genome
Introducing a new, unifying DNA sequence model that advances regulatory variant-effect prediction and promises to shed new light on genome function — now available via API.
deepmind.google
December 8, 2025 at 12:11 PM
AlphaGenome analyzes up to 1M DNA base-pairs at single-letter resolution, predicting effects on gene expression, splicing, chromatin and more. It can flag how one base change alters biology, disease risk or therapy.
deepmind.google/blog/alphage...
deepmind.google/blog/alphage...
Call for Presentations: DCI Network 2026 Conference on Patient-Powered AI in Healthcare, June 22-24 at Harvard. Submit research on AI engagement, patient co-design, ethics, and real-world impact. Published in PubMed-indexed proceedings. www.dcinetwork.org/patients2026...
December 8, 2025 at 6:30 AM
Call for Presentations: DCI Network 2026 Conference on Patient-Powered AI in Healthcare, June 22-24 at Harvard. Submit research on AI engagement, patient co-design, ethics, and real-world impact. Published in PubMed-indexed proceedings. www.dcinetwork.org/patients2026...
Peking University unveils AI-Newton, an autonomous system that rediscovers physical laws like Newton’s second law and gravity from raw data—no prior physics knowledge required.
arxiv.org/abs/2504.01538
arxiv.org/abs/2504.01538
AI-Newton: A Concept-Driven Physical Law Discovery System without Prior Physical Knowledge
Current limitations in human scientific discovery necessitate a new research paradigm. While advances in artificial intelligence (AI) offer a highly promising solution, enabling AI to emulate human-li...
arxiv.org
December 7, 2025 at 4:37 AM
Peking University unveils AI-Newton, an autonomous system that rediscovers physical laws like Newton’s second law and gravity from raw data—no prior physics knowledge required.
arxiv.org/abs/2504.01538
arxiv.org/abs/2504.01538
Foundation models can predict well but still fail to learn true underlying physics. New MIT/Harvard study shows they rely on heuristics, not world models. Important read for AGI debates.
arxiv.org/abs/2507.06952
arxiv.org/abs/2507.06952
What Has a Foundation Model Found? Using Inductive Bias to Probe for World Models
Foundation models are premised on the idea that sequence prediction can uncover deeper domain understanding, much like how Kepler's predictions of planetary motion later led to the discovery of Newton...
arxiv.org
December 7, 2025 at 4:37 AM
Foundation models can predict well but still fail to learn true underlying physics. New MIT/Harvard study shows they rely on heuristics, not world models. Important read for AGI debates.
arxiv.org/abs/2507.06952
arxiv.org/abs/2507.06952
A new multi-institution study defines what AGI actually requires and finds GPT-4 at 27 percent and GPT-5 at 58 percent toward full human-level cognition, with major gaps in memory and reasoning still unsolved.
arxiv.org/html/2510.18...
arxiv.org/html/2510.18...
A Definition of AGI
arxiv.org
December 7, 2025 at 3:42 AM
A new multi-institution study defines what AGI actually requires and finds GPT-4 at 27 percent and GPT-5 at 58 percent toward full human-level cognition, with major gaps in memory and reasoning still unsolved.
arxiv.org/html/2510.18...
arxiv.org/html/2510.18...
A review of 8,590 expert forecasts finds AI researchers expect AGI around 2040, far later than the timelines promoted by tech CEOs. The difference comes down to data, incentives and scientific rigor. research.aimultiple.com/agi-singular...
research.aimultiple.com
December 7, 2025 at 3:32 AM
A review of 8,590 expert forecasts finds AI researchers expect AGI around 2040, far later than the timelines promoted by tech CEOs. The difference comes down to data, incentives and scientific rigor. research.aimultiple.com/agi-singular...
1/7 Microsoft committed $13B to OpenAI, but now pays AWS—their cloud rival—to integrate Anthropic's Claude into Office 365. Why? Claude excels at PowerPoints and Excel tasks per internal tests. aimagazine.com/news/why-mic... #AI #Microsoft #OpenAI
Why Microsoft Adds Anthropic to Office 365 Alongside OpenAI
Microsoft amplifies its AI capabilities in Office 365 with Anthropic’s advanced Claude models alongside OpenAI’s technology and AWS’ infrastructure
aimagazine.com
December 4, 2025 at 4:44 AM
1/7 Microsoft committed $13B to OpenAI, but now pays AWS—their cloud rival—to integrate Anthropic's Claude into Office 365. Why? Claude excels at PowerPoints and Excel tasks per internal tests. aimagazine.com/news/why-mic... #AI #Microsoft #OpenAI
Is the AI Bubble about to burst? Microsoft stock sinks on report AI product sales are missing growth goals www.cnbc.com/2025/12/03/m...
Microsoft stock sinks on report AI product sales are missing growth goals
Microsoft stock sank after a report that multiple sales teams for the company's Azure Foundry product missed growth goals in the last fiscal year.
www.cnbc.com
December 3, 2025 at 4:26 PM
Is the AI Bubble about to burst? Microsoft stock sinks on report AI product sales are missing growth goals www.cnbc.com/2025/12/03/m...
@aobolaji.bsky.social Thanks for being my 2000th follower! Nice to connect on Blue Sky! We can connect on LinkedIn: www.linkedin.com/in/yuriquint...
Yuri Quintana, PhD, FACMI, FIAHSI, FAMIA - Beth Israel Deaconess Medical Center | LinkedIn
Yuri Quintana, Ph.D., FACMI, FIAHSI is an internationally recognized leader in digital… · Experience: Beth Israel Deaconess Medical Center · Education: University of Waterloo · Location: Greater Bosto...
www.linkedin.com
December 3, 2025 at 4:26 PM
@aobolaji.bsky.social Thanks for being my 2000th follower! Nice to connect on Blue Sky! We can connect on LinkedIn: www.linkedin.com/in/yuriquint...