Bihan Jiang

Based on the provided articles, here are 3 promising AI startups with recent funding news: 1. Flint - Raised $5 million in seed funding (October 14, 2025) - Led by Accel with participation from Sheryl Sandberg's fund and Neo - Building an AI platform that autonomously builds and updates websites - Source: https://lnkd.in/dYMYVyWu 2. Leonardo.ai - Acquired by Canva (announced July 29, 2024) - Had previously raised over $38.8 million from investors including Smash Capital, Blackbird, Side Stage Ventures and others - Specializes in generative AI content creation with over 19 million registered users - Source: https://lnkd.in/d3J5DAMA 3. Coco Robotics - While specific funding isn't mentioned, the company announced launching a new physical AI research lab (October 14, 2025) - Partnering with OpenAI and led by UCLA professor Bolei Zhou - Focuses on autonomous last-mile delivery robots - Source: https://lnkd.in/dUbYZDCf

3

🚨 According to the World Health Organization’s June 2024 report, over 3 million people die each year due to alcohol and drug use, with men making up the majority of these deaths. In our latest episode, we talk with Faiza Khan Khattak, PhD, CTO of a startup developing a personalized therapy platform for Alcohol Use Disorder (AUD). This conversation is a must-listen for anyone interested in mental health innovation, women-led tech, and how AI can address urgent health crises. 🎧 Tune in to learn more about this inspiring initiative! #WiAIR #WomenInTech #MentalHealth #MedTech #AlcoholUseDisorder #AUD #HealthTech #TechForGood

7

1 Comment

🧠 AI Rights Historical Wrongs Researchers at Stanford and Princeton have fine-tuned a large language model to uncover racially discriminatory housing clauses buried in millions of historical property deeds — accelerating a process that could have taken a decade by hand. 📍 The problem: Many property deeds in Northern California — especially in Santa Clara County — still contain illegal racial covenants from as far back as the 1850s, banning people of color from owning or living in certain homes. 🔎 The solution: The team used OCR to digitize 5.2 million deed pages, then fine-tuned Mistral-7B via LoRA on a hand-labeled dataset of discriminatory phrases. The model learned to identify harmful language in context, avoiding false positives like the surnames "Black" or "White." 📊 Results: Detected 24,500 lots with racial clauses (~25% of homes in 1950) Found that 10 developers were responsible for 1/3 of all discriminatory clauses Model reviewed all pages in 6 days at a cost of $258 Equivalent manual effort: 10 years, $1.4M 💥 This work doesn't just use LLMs to build the future — it uses them to understand and repair the past. The open-sourced model now enables other U.S. counties to follow suit. #AI #MachineLearning #LLM #ResponsibleAI #Stanford #Princeton #HistoryMeetsAI #TechForGood #EthicalAI #SocialImpact #PropertyTech https://lnkd.in/dtec-zbx

4

📢 New Paper Published! Excited to share our latest work, "LitANFIS: Literal-aware Adaptive Neuro-Fuzzy Inference System to learn Conjunctive Normal Form," now published in Neurocomputing. This model brings negation into fuzzy rule learning, enabling more compact and interpretable systems. Thanks to Dr. Armin Salimi-Badr for his support. Check out the publication here: https://lnkd.in/eJ-WgnJz (Available until Feb. 06, 2025)

87

24 Comments

Titans + MIRAS: Helping AI have long-term memory From Ali Behrouz, Meisam Razaviyayn, Vahab Mirrokni Summary Google Research has introduced a new architecture (Titans) and a theoretical framework (Miras) to solve a critical bottleneck in modern AI: the trade-off between memory capacity and computational speed.   * Titans is a new model architecture that adds a "neural long-term memory" module to standard AI models. It allows the model to "learn" and update its memory in real-time as it processes data, rather than just reading a static window of text.   * Miras is the "blueprint" or framework that explains how these memory systems work mathematically. It unifies previous architectures (like Transformers and RNNs) under one theory and proposes new ways to optimize how models decide what to remember and what to forget. Analysis 1. Solving the "Memory Amnesia" Problem The biggest limitation of current LLMs is that they have a fixed context window. Once information slides out of that window, it is gone forever. Titans introduces Test-Time Training (TTT), meaning the model effectively "trains" itself on the fly. It updates its own internal parameters as you talk to it, allowing it to maintain a "gist" of a conversation indefinitely without needing to re-read the entire history every time. 2. Moving Beyond "Static" Compression Previous attempts to fix this (like Mamba or RWKV) tried to compress history into a fixed state. Titans proves that a Deep Neural Network is a better container for memory than a simple vector. By making the memory module itself a neural net, it has much higher "expressive power"—it can understand complex relationships rather than just storing keywords. 3. Efficiency vs. Accuracy The blog claims Titans outperforms GPT-4 on the BABILong benchmark (a test for finding needles in very large haystacks of text) while being much more efficient. If this scales, it could lead to AI assistants that can "read" entire libraries of technical documentation or legal codebases and reason about them instantly without the massive cloud costs currently associated with long-context Transformers. 4. A New Class of Models? Miras suggests we are moving away from the "Transformer Era" into a "Memory-Centric Era." By generalizing how memory works, Google is opening the door for models that are custom-built for specific memory tasks—some that never forget (for legal/medical use) and some that prioritize recent adaptability (for real-time chat). https://lnkd.in/gk8DJKJs

I recently completed the Product Faculty's AI Product Strategy program with Miqdad Jaffer, Satyajeet Salgar, Henry Shi, and Jennifer Liu, leaders across OpenAI, Google AI and Anthropic, a strong follow-on to the AI Product Management program I did with Miqdad last year. If AI Product Management taught me how to build and ship AI products, AI Product Strategy taught me why most of them fail. The answer is almost never the technology. Three ideas I keep coming back to: 1. Behavioural science is the real strategy. AI is the enabler. The goal is not to impress users with capability. It is to deliver outsized value consistently enough that the product earns a place in their daily workflow. 2. Inertia is a more powerful force than most product teams account for. Users do not abandon familiar habits for better products. They switch when the value of change clearly outweighs the comfort of staying put. 3. The moat is not the model. It is the learning. Features get copied fast. What compounds over time is the insight you build by iterating quickly around a specific user and a specific problem. Excited to keep turning strategy into adoption.

33

3 Comments

Looking forward to this session from Shaheer Tariq. He has a solid understanding of both the AI landscape and the needs of regulated sectors, and has a great ability to explain things simply..

12

𝐔𝐑𝐎𝐆 𝐢𝐬 𝐥𝐢𝐯𝐞. Cold emailing professors for research opportunities is practically a rite of passage at GIKI. It is also incredibly inefficient. You send an email, wait weeks for a reply, and often just hope for the best. That is exactly the gap Alishba Ramzan noticed during her UGRAD exchange. She saw how structured platforms like #UROP streamlined research in the US and realized we were missing that infrastructure here. Along with Muhammad Ahmed, she set out to build UROG to fix it. When I interviewed for the Technical Lead role, the vision was clear, but the canvas was blank. We needed a centralized hub where students could discover projects and professors could manage talent without the email clutter. 𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐭𝐡𝐢𝐬 𝐟𝐫𝐨𝐦 𝐬𝐜𝐫𝐚𝐭𝐜𝐡 𝐰𝐚𝐬 𝐚 𝐣𝐨𝐮𝐫𝐧𝐞𝐲. I chose a stack that could handle complexity while moving fast: Next.js 14 and TypeScript for the frontend, paired with Supabase for the backend. The hardest part wasn't the code itself, but the logic. Designing role-based dashboards that look different for a freshman student versus a tenured professor required a lot of architectural rethinking. There were plenty of late nights debugging generic "500 Internal Server Error" messages and rewriting database policies to ensure security. But seeing the system work is worth it. Right now, UROG handles project discovery, a full application management system with status tracking, and real-time notifications. You no longer have to guess if your application was seen. We are still iterating. My teammate Haseeb Syed is currently doing some heavy lifting to integrate real-time messaging so students and supervisors can chat directly within the app. It is not just about code. It is about saving everyone time.  And I must confess that the UI looks 𝘒𝘢𝘧𝘬𝘢𝘦𝘴𝘲𝘶𝘦. If you are a student or dev who has dealt with the chaos of manual applications, I would love to hear how you handled it. distrub the web: https://lnkd.in/dyT6ymS7 #GIKI #WebDevelopment #NextJS #Supabase #UROG #StudentResearch #Engineering #SoftwareDevelopment #React #dev #project #CS

37

4 Comments

ML ensembles benefit from the diverse predictions of their models, but what if incorrect models still outvote correct ones?  Our newest paper at DSN 2025, "ReMlX", uses explainable AI (XAI) to uncover what features each model focuses on - like edges or colours in images. This enables ensembles to reconsider their voting decisions based on features in safety-critical systems. https://lnkd.in/gFsf9ET3

24

A new study from Carnegie Mellon, “Does AI-Assisted Coding Deliver? A Difference-in-Differences Study of Cursor’s Impact on Software Projects” that looks at a more recent period, tracking code quality in projects using GenAI tools up to mid-2025, finds that AI briefly accelerates code generation, but the underlying code quality trends continue to move in the wrong direction. https://lnkd.in/gm6dxb8W

2

5 Comments

Excited to share what we've been building at Precepi 🚀 We've been working on the core AI and search infrastructure behind this platform designing retrieval systems, building data pipelines, and shipping production features alongside an incredible team 💪 Pharmaceutical labeling is a space where precision matters 💊. Building AI that meets that bar has been one of the most rewarding engineering challenges I've worked on. Proud to see this go public 🎉 More to come.

5