Patrick Emedom-Nnamdi

I’m excited to share that our manuscript was accepted to Neurosurgery (@NeurosurgeryCNS)! In our latest work, we use data collected from smartphone-based sensors to evaluate mobility trends among patients with Glioblastoma (an aggressive form of brain cancer) after receiving a surgical resection of their tumor. Additionally, we performed an individual-level analysis and examined change in mobility after an initial 6-week treatment with concomitant temozolomide and radiotherapy followed by multiple cycles of adjuvant chemotherapy. Our work highlights the potential of using GPS-based features to objectively capture changes in patient quality-of-life during key moments of their treatment and recovery periods.

Come do a PhD with me at Columbia! My lab tackles basic problems in alignment, interpretability, safety, and capabilities of language systems. If you love adventuring in model internals and behaviors---to understand and improve---let's do it together! pic: a run in central park

Interpretability research has made only minor contributions to AI safety so far. What can we do to change that? (Clip from a longer talk; YouTube link in the thread):

🔥Veo 3 has emergent zero-shot learning and reasoning capabilities! This multitalented model can do a huge range of interesting tasks. It understands physical properties, can manipulate objects, and can even reason. Check out more examples in this thread!

I gave quite a personal talk yesterday at UQÀM on the history of semantics (and pragmatics), charting out a modern connectionist semantics in which meanings are tensors and LLMs are important investigative tools. YouTube link just below.

Big fan of Jeremy’s work on optimization—great to see his first Thinking Machines post!

Clinical AI in the LLM era has done a brilliant job optimizing the 30-minute doctor's visit. But what about the 100,000+ minutes a patient spends at home between appointments? Curious? Excited to share my first post on Olira. Check it out below!

The amount of compute that labs have is growing dramatically. For example, @OpenAI has 14x more compute than when it launched GPT-4. @xAI is growing compute extremely rapidly, and they doubled the compute available between the Grok 3 and Grok 4 launches. @AnthropicAI has the least compute, but they’ve been very efficient at converting compute into highly successful models. For example, Claude 3.5 was launched with about one-fifth of the compute Anthropic has now. Note: data excludes @GoogleDeepMind which likely has the most compute, though not clear how concentrated it is for LLM training. Source: @SemiAnalysis_

We're excited to start sharing daily trends and topics in outpatient monitoring and healthcare AI! Stay tuned for more!

Olira is officially live! 🚀 American healthcare is at a crossroads - nearly 60% of adults live with at least one chronic condition, the population over 65 will grow by 47% to 82 million by 2050, there’s a projected shortage of 124,000 physicians by 2034, and specialty fields are seeing an explosion of novel therapies that demand closer monitoring than ever before. Yet, traditional remote monitoring has become part of the problem - flooding providers with alerts but offering limited ways to act on data. Care teams are overwhelmed and patients who need support fall through the cracks. At Olira, our mission is to provide insights without the noise and support without the burden. We capture the complete patient story between visits and turn everyday data into actionable, clinical intelligence for clinicians. Our Focus: Bridging machine learning and clinical science, we use digital phenotyping to analyze passive sensor data and voice/text interactions - creating a clear, real-time picture of patients’ health for providers, while delivering personalized support to patients along their care journey. Our Technology: Our support layer starts with the device patients already own: their smartphone. Olira's application transforms continuous data into insights that seamlessly and safely integrate into clinical workflows, giving providers signals without the noise. It’s simple, secure, and scalable. We know that making a measured impact requires the best possible measures. That is why persistent evaluation and responsible systems design are the cornerstones of our entire technological foundation. Learn more at olira.ai. #healthcare #ai #technology

I’d also like to thank my PhD advisors, @jponnela and Junwei Lu, for their continued mentorship and for inspiring the journey that led here. And to the Blavatnik Fellowship at Harvard Business School, thank you for being my earliest supporter and helping turn this vision into a reality. For my father, and for the millions of people navigating chronic illness, we are building a more proactive and empathetic future for healthcare. The real work starts now. Learn more about our mission at olira.ai. If you're a clinician, researcher, or builder who believes in a future where no patient falls through the cracks, I would love to connect.

10 years ago, I lost my father to a sudden brain aneurysm. What felt like a bolt from the blue was, in reality, the culmination of chronic diseases that were incredibly difficult to manage alone. His story is a deeply personal example of a systemic challenge in healthcare. The gaps between doctor visits are vast, leaving patients to navigate complex conditions with little support and clinicians with a poor window into their patients' daily lives. How can we expect to create effective interventions when we're missing 99% of the story? This question fueled my decade-long academic journey through a PhD in Biostatistics at Harvard, where I dedicated my research to solving this very challenge. Designing algorithms to objectively capture a patient's function is a non-trivial task. Creating personalized support by hand is nearly impossible at scale. And integrating new digital tools into a clinician's workflow is a massive barrier. Traditional remote monitoring has tried to solve this, but often becomes part of the problem—flooding providers with alerts and noise. We need a better way. We need to turn everyday data into clinical intelligence. That’s why today, I am incredibly proud and humbled to announce the launch of Olira and our $2.1M in funding to build a new future for outpatient care. Olira is an AI-native platform that captures the complete patient story between visits. We start with devices patients already own—their smartphone—and use digital phenotyping to transform passive sensor data and active voice/text interactions into clear, proactive guidance for clinicians and personalized support for patients. Our mission is simple – Insights without the noise. Support without the burden. This journey is only possible with the support of an incredible founding team. I’m deeply grateful to our visionary investors: Precursor Ventures, Dria Ventures, Plug and Play Ventures, Black Tech Nation Ventures, and leading angels including @JeffDean and @jefrankle. We're also guided by the wisdom of our brilliant advisors Mark Fleming, MD., @ChrisGPotts, and Toby Eduardo Redshaw.

@ekzhang1 This is awesome @ekzhang1 !

We're thrilled to share Modal Notebooks: a new, powerful cloud-hosted GPU notebook. It has modern real-time collaborative editing and is backed by our AI infrastructure — swap GPUs in seconds. Modal Notebooks are generally available, and you can start using them now. 🧵

For context they're 7/9!

Started teaching my niece and nephew to program using Scratch - it seems to be incredibly fun and intuitive for them so far (which is awesome!). I'm curious if there's anything similar for getting them comfortable with early concepts of stats / ml, and perhaps (cautiously) creative AI tools.

🥹🥹 All vague-posting aside, super happy this model is finally out there & proud of everyone for making this happen 💖 let us know what you think!

1/ The US used to dominate the critical minerals space. But we outsourced all of it to China. Tax credit, executive orders and tariffs are trying to bring these supply chains by brute force But, only cost leadership will win. Here’s why:

My longtime collaborator Dave Patterson (long-time faculty at @UCBerkeley, @TheOfficialACM Turing Award winner, and fellow @LaudeInstitute board member) wrote a very good op-ed about how continued investing in basic science and technology research is essential for the U.S. Dave describes the government funding his labs received, but also the impact: "For decades, my research received support from the NSF’s Directorate for Computer and Information Science and Engineering, through grants and Ph.D. fellowships. All told, American taxpayers invested just under $100 million in the labs I helped lead. Accounting for inflation, technologies that came out of them went on to generate over $1 trillion in product sales. That is a 10,000-to-1 return on investment to the public and surely at least 1,000-to-1 return directly back to the government in taxes. Not from luck, but from decades of public investment." ... "Every American taxpayer is a silent shareholder in that success. If we walk away now, we lose not just future breakthroughs but also what we have already earned." Continuing to invest in basic research in the U.S. is what leads to these sorts of extraordinary outcomes. thehill.com/opinion/techno…

@pdhsu Pinged!

Ping me if you’re building this

I’ve started taking pictures of nearly every meal I eat to feed into a multimodal AI doctor that has context on my wearable, how I’m feeling, blood work, meds/supplements, genome variants interpreted with a reasoning model, etc It’s a very hacky prototype now but I’m surprised no one has built the definitive version of this yet. Clear network effects to be unlocked at scale, eg RCTs

Journal Link: doi.org/10.1214/24-AOA… Paper Link: patrickemedom.com/assets/pdf/AOA… Code: github.com/patricknnamdi/…

Many thanks to my PhD advisors @jponnela and Junwei Lu (Harvard University), and collaborator Timothy R. Smith (Brigham and Women's Hospital)

🎉 Excited to share that our paper was published in The Annals of Applied Statistics! "Nonparametric Additive Value Functions: Interpretable Reinforcement Learning with an Application to Surgical Recovery"