Joyce Wang

Augment and Invent ✨ 💡 In a recent substack post (https://lnkd.in/g4iCuVaY), the always insightful Rishad Tobaccowala presents "10 thoughts about AI, Humans and Work in 10 minutes". They are all worth pondering (although I would not invoke Suleyman in # 3) but one that struck a chord with me, Thought # 6, summarizes two important themes: human augmentation and business (re)invention. Here is the thought: "6. The simpler questions are about efficiency and effectiveness. The real question is more existential. AI will deliver efficiencies by getting things done faster and cheaper and as importantly more effectively by unleashing insights and freeing talent to work on areas where humans excel and let the math and pattern finding and the rest be done by the machine. But the smart companies are thinking about how to re-invent, re-imagine and re-think their business. In the world we live today with distributed and unbundled work, next generation technologies and multi-polar globalization why are more companies not re-imagining their businesses versus just focussing on making yesterdays model cheaper and more effective?" Obvious pockets of inefficiencies will be addressed with #AI in the short term by businesses, thereby quickly eliminating any competitive advantage (a point made very nicely by Martin Reeves in a recent HBR article https://lnkd.in/gsaHk4V2). Any sustainable advantage will have to come from new ideas about augmentation (that amplify a company's unique assets) and from novel ideas in general for businesses. So, first, augmentation. I have been for 25 years a proponent of thinking about "unleashing insights and freeing talent to work on areas where humans excel and let the math and pattern finding and the rest be done by the machine". What is missing from a lot of the discourse on augmentation is what form it should take, specifically, how do humans and AIs interface? The chatbot form factor has been surprisingly effective (surprising to me at least) at opening floodgates of creativity from all kinds of horizons. It can't be the last frontier of human-AI interfacing. That needs to be invented. Which takes us to the second part: (re)invention. The winners will use #genAI to create new products, services, experiences that are not even conceivable without AI. Business model disruption driven by AI is just beginning. When you product or service can be 80% replaced by a self-serve AI-based version at 0.001% of the cost, you have to seriously rethink where you create value. And as Rishad Tobaccowala writes in thought # 7, "A company’s best opportunities and threats are likely come from outside its immediate category and AI allows it to extend into new areas."

3

I just published > "Part 4 - Non Local Interactions in AGI through Weighted Choquard Equation" In the pursuit of building Artificial General Intelligence (AGI), integrating long-term memory and accumulated experiences into cognitive reasoning is crucial for developing systems that can adapt and learn like humans. Traditional AI models often fall short in this aspect, focusing more on pattern recognition than on experiential learning. The weighted Choquard Equation offers a sophisticated mathematical framework that can model these non-local interactions, enabling AGI to maintain a dynamic, experience-based memory system. This approach not only captures long-range dependencies and prioritizes significant experiences but also ensures stability and coherence in memory integration, providing a new paradigm for cognitive reasoning in AGI. #AGI #AI #Math #ArtificialIntelligence https://lnkd.in/gVykvyjY

3

Team of Physics, Math and CS engineers are solving complex molecular biology problems in AI. Almost no biologists 👏 Chai Discovery team is excited to release Chai-1, a new multi-modal foundation model for molecular structure prediction that performs at the state-of-the-art across a variety of tasks relevant to drug discovery. Chai-1 enables unified prediction of proteins, small molecules, DNA, RNA, covalent modifications, and more. Chai-1 model achieves a 77% success rate on the PoseBusters benchmark (vs. 76% by AlphaFold3), as well as an Cα LDDT of 0.849 on the CASP15 protein monomer structure prediction set (vs. 0.801 by ESM3-98B). #Math #AI #Biology https://lnkd.in/gTJEgVBY

4

Belated post, but my recent talk on how we're leveraging #Metaflow for disease trends forecasting at a lab in Santa Clara University is now featured in the #Metaflow Stories playlist by Outerbounds! #Metaflow is an open-source framework that makes it easy to build and manage #AI projects. Thanks Shri Javadekar and Outerbounds for the platform, and to everyone who participated in the discussion! The full presentation is available here https://lnkd.in/dmmTdnHZ On a side note, I'm open for Summer & Fall internship opportunities in ML and LLM space. If you've something in mind, hit me up! #ML #MLOps #MLEngineer

48

2 Comments

Mind-blowing quantum computing breakthrough! Google just unveiled their Willow chip that performed a calculation in 5 MINUTES that would take today's fastest supercomputers 10 septillion years (that's longer than the universe has existed!) The real game-changer: Willow solved a 30-year-old problem in quantum error correction - the more qubits they add, the FEWER errors occur. This is a huge step toward practical quantum computers that could revolutionize everything from drug discovery to AI. As a sci-fi fan, I can't help but dream about how quantum computing could help us understand the fabric of space-time. Who knows - maybe this is one step closer to that warp drive! Want to geek out about this together? Check out the full announcement:

4

A major mental health company just got slammed by the FTC for playing fast and loose with client data. And it could have been avoided. Here's the secret: if your company is collecting, sharing, or analyzing data, you should get IRB (Institutional Review Board) approval. IRBs don't just deal with research that will be published. You can get their help to keep things above board for almost anything data-related. If you’re a healthcare company, you will be collecting data. Everyone does. And even if your privacy policy covers data usage, you should still get IRB approval to analyze this data. They'll point out any ethical conflicts to avoid and help you prevent a major PR crisis later on. Approval from an IRB means that the study has been reviewed and that the procedures for collecting and sharing the data meets high ethical standards for protecting the privacy and rights of those who use your services.       IRBs help ensure your practices are HIPAA compliant, that you're handling PHI correctly, and use aggregated de-identified data from customers for writing reports and publications (which means you don’t need explicit consent from each individual, but you’re still protecting their privacy). You can hire an independent IRB to help you make sure that your methods are ethical and your data handling is secure. The IRB process might seem tedious, but there are no shortcuts to maintaining trust and credibility. It protects both your company and your customers. And trust me, presenting everything to an IRB board is a lot better than presenting everything to the FTC. #DataPrivacy #IndustryScience #FractionalCSO

98

13 Comments

📢 Academic Integrity: A Growing Concern (Part 2) Unlike traditional journals, preprint services (like arXiv, bioRxiv & medRxiv, EarthArXiv Preprint Service, Research Square) allow researchers to share their work without undergoing the typical peer review process. While this accelerates the sharing of ideas and collaboration, it also means there is less oversight, potentially allowing anyone to upload content without stringent checks. The integrity of citation metrics is crucial because these metrics play a significant role in how scientists are ranked and recognized globally. One such ranking is Stanford University and Elsevier top 2% Scientists list, which identifies scholars who are top-cited in their respective fields. This ranking is based on an analysis of citation impact across multiple scientific fields and subfields. Being included in such a list is a mark of distinction for researchers, and unethical practices like artificial citation manipulation can threaten the integrity of these rankings, misrepresenting the true impact of a scholar's work. 🛡️ Here are some steps we can take to address these issues: 📩 Contact Support: Reach out to support teams of these platforms with detailed information if you notice any misuse of citations. 🗂️ Document Incidents: Keep records of artificial papers and any communication with the platform, which can be useful if the issue escalates. 🏫 Inform Academic Institutions: Academic institutions can offer support and may help address such issues on a larger scale. 🧑‍⚖️ Consider Legal Advice: If misuse persists or causes reputational damage, exploring legal options might be necessary. #ResearchIntegrity #AcademicReputation #EthicalResearch #ScholarshipMatters #ScienceCommunication #HIndexGaming #IntegrityMatters #ProtectResearch #ResearchAccountability #ResponsibleResearch #FraudInResearch #PublishOrPerish #QualityResearch #AcademicFraud #ResearchEthicsMatter #ResearchQualityAssurance #TrustInAcademia #FakePapers #ResponsiblePublishing #IntegrityInScholarship

7

After decades of experience in building foundational AI models, the simplest rule of thumb I've observed that works empirically as a framework for hybrid modeling AI platforms, when applied in the following strategic order, is: 1) Start by extracting simple local features, prioritizing multi-scale extraction early on. 2) Then, focus on broad-range (dilated) interactions, ensuring high resolution by incorporating skip or residual connections. 3) Next, recalibrate features by modeling inter-dependencies (squeeze and excite). 4) Follow this by capturing long-range dependencies through some form of state management (e.g., multi-head attention). 5) Enhance expressivity through non-linear transformations (such as non-linear feedforward networks). 6) Finally, condense the representation using global pooling or dense layers. By applying these layers in multiple blocks through stacking, you will generally achieve effective numerical conditioning rapidly, minimizing the need for extensive ablation studies. This approach has consistently proven to be reliable. I stand by this rule all the time and works for me in general. #AI #ArtificialIntelligence #Engineering

4

A new Stanford Wu Tsai Neuroscience Institute podcast where Nicholas Weiler and I discuss the interplay between biological and artificial intelligence, the huge data and energy efficiency of brains over machines, explainable AI, and closed loop brain machine interfaces https://lnkd.in/gvNB-FEb

46

A framework for human evaluation of large language models in healthcare derived from literature review With generative artificial intelligence (GenAI), particularly large language models (LLMs), continuing to make inroads in healthcare, assessing LLMs with human evaluations is essential to assuring safety and effectiveness. This study reviews existing literature on human evaluation methodologies for LLMs in healthcare across various medical specialties and addresses factors such as evaluation dimensions, sample types and sizes, selection, and recruitment of evaluators, frameworks and metrics, evaluation process, and statistical analysis type. Our literature review of 142 studies shows gaps in reliability, generalizability, and applicability of current human evaluation practices. To overcome such significant obstacles to healthcare LLM developments and deployments, we propose QUEST, a comprehensive and practical framework for human evaluation of LLMs covering three phases of workflow: Planning, Implementation and Adjudication, and Scoring and Review. QUEST is designed with five proposed evaluation principles: Quality of Information, Understanding and Reasoning, Expression Style and Persona, Safety and Harm, and Trust and Confidence. https://lnkd.in/giHmFEbA

7

"Sequence modeling and design from molecular to genome scale with Evo" by Eric Nguyen, Michael Poli, Matthew Durrant, Armin Thomas,  Jeremy Sullivan, Madelena Ng, Ashley Lewis, Aman Patel, Aaron Lou, Stefano Ermon, Tina Hernandez-Boussard, Patrick Hsu, Brian Hie et al. "The genome is a sequence that completely encodes the DNA, RNA, and proteins that orchestrate the function of a whole organism. Advances in machine learning combined with massive datasets of whole genomes could enable a biological foundation model that accelerates the mechanistic understanding and generative design of complex molecular interactions. We report Evo, a genomic foundation model that enables prediction and generation tasks from the molecular to genome scale. Using an architecture based on advances in deep signal processing, we scale Evo to 7 billion parameters with a context length of 131 kilobases (kb) at single-nucleotide, byte resolution. Trained on 2.7M prokaryotic and phage genomes, Evo can generalize across the three fundamental modalities of the central dogma of molecular biology to perform zero-shot function prediction that is competitive with, or outperforms, leading domain-specific language models. Evo also excels at multi-element generation tasks, which we demonstrate by generating synthetic CRISPR-Cas molecular complexes and entire transposable systems for the first time. Using information learned over whole genomes, Evo can also predict gene essentiality at nucleotide resolution and can generate coding-rich sequences up to 650 kb in length, orders of magnitude longer than previous methods. Advances in multi-modal and multiscale learning with Evo provides a promising path toward improving our understanding and control of biology across multiple levels of complexity." Blog: https://lnkd.in/dDpfHZbd Paper: https://lnkd.in/dpaswkHR

43

2 Comments

We need to 'concretize' aging clocks to accelerate discovery in the biology of aging - Martin Borch Jensen and I just wrote a Norn Group call to action proposing the path for robustly validating aging clocks to the point of replacing lengthy lifespan studies. The vision? Accelerating pivotal experiments 2-10x, by extending preclinical lifespan studies to generate the data needed to benchmark the forecasting power of clocks. Full piece: https://lnkd.in/e97YZGZV Research on the biology of aging is held back by the lengthy timelines of experimental studies involving natural aging. The consequence is that we face an enormous hurdle - time - in developing life-saving therapeutics that target this biology to make a difference in prevalent chronic conditions (neurodegenerative disease, sarcopenia, heart disease, and many more). Biomarkers of the aging process (aging 'clocks') would unlock faster breakthroughs, laying the foundations for surrogate endpoints in human clinical trials, just like cholesterol became adopted as a surrogate endpoint for cardiovascular death. Many clocks have been proposed in the last 13 years, but no definite process for validating their predictive power exists. We envision the Clock Assessment Program (CAP) as a gold standard for benchmarking aging clocks, deployed to accelerate preclinical discovery by reducing the amount of time it takes to identify interventions that meaningfully shift survival curves. Instead of waiting 30 months or more for mouse experiments to finish, CAP-validated clocks will predict eventual endpoints (lifespan, frailty index at a given age, etc.) at earlier timepoints (e.g., 12 and 18 months of age) with sensitivity and specificity that clears a high bar. To no longer wait three years for each preclinical lifespan study, to not take up a decade to validate life-saving therapies that treat universal age-related diseases, we need a rigorous, visionary program to enable progress. Now is the time to concretize. We invite you to join us in tackling or sponsoring this initiative, and reach out to us for further discussion.

71

1 Comment

Many thanks to my co-panelists on "Impacts of AI’s Rapid Rise in the Life Sciences" and for the opportunity to share my thoughts on over a decade of supporting life sciences companies in protecting the investments they make in software and data science! Michelle Chen of Insilico Medicine, Hans Sauer of Biotechnology Innovation Organization, and myself, alongside our moderator Allison Schmitt, spoke about business impact of AI in the life sciences, IP protection strategies, and impact of current regulatory practices on use of AI in the invention process. The recording is available at the link below. Reach out if you'd like to continue the conversation! #ai #ml #aiml #artificialintelligence #techbio #computationalbio #bioinformatics #drugdiscovery #patent #ip #intellectualproperty

38

2 Comments

“For so many startups and even larger tech incumbents, the point at which they hit the shoulder in the S-curve is a mystery, and I suspect the failure to see it occurs much earlier. The good thing is that identifying the enemy sooner allows you to address it. We focus so much on product-market fit, but once companies have achieved some semblance of it, most should spend much more time on the problem of product-market unfit. For me, in strategic planning, the question in building my forecast was to flush out what I call the invisible asymptote: a ceiling that our growth curve would bump its head against if we continued down our current path. It's an important concept to understand for many people in a company, whether a CEO, a product person, or, as I was back then, a planner in finance”

2

1 Comment

🚀 Professor Hani Zaher (and Lab) have uncovered a groundbreaking mechanism in cellular stress response linked to ribosome speed! 🧬 Their recent paper reveals how cells activate stress responses by altering ribosome speed through the interplay of eIF4E and eIF4A factors. This discovery not only deepens our understanding of cellular biology but also opens new avenues for cancer research, potentially influencing therapeutic strategies. 🔬 #CellBiology #ResearchBreakthrough #CancerResearch #RibosomeSpeed #ScienceInnovation Read more: https://lnkd.in/e-KJnHYV

5

you can use it easily @ tripod-llm.vercel.app towards more transparent science reporting

13

I would rather fail while realizing a transformative idea than succeed in advancing an incremental idea. Transformative ideas will eventually succeed, be in someone else’s hands, and get realized. In 1990's, myself, Yosi Kost and Bob Langer launched Sontra Medical to develop a non-invasive wearable continuous glucose monitor (CGM) using #ultrasound. Another company, Cygnus was leading the field with a reverse #iontophoresis CGM championed by pioneers Richard Guy, Janet Tamada, Russell Potts. These were early CGM devices and wearables that challenged the status quo of diabetes management at the time. A dream was born. Cygnus and Sontra both demonstrated efficacy of their CGM product in patients. However, neither was able to make a commercially viable product. The dream of CGM and non-invasive wearables, however, lived on. Thirty years later, CGMs are widely used by patients and many non-invasive wearable metabolic sensors are currently in development. Scott Kellogg Michael Langer Tuan Elstrom Shikha Barman #dreamon #diabetes #wearables #cgm Photo: Article from MIT Technology Review nearly half of my life ago.

485

14 Comments