Surojit (Roj) Niyogi

I've looked at a lot of YC batches. The bio and health lineup in P26 is STACKED. https://lnkd.in/gQKxpe9t Adialante - Compact, mobile whole-body MRI to make cancer screening accessible at hundreds per scan FinalDose - A programmable drug that reads DNA inside the cell and destroys it if it's diseased, starting with previously undruggable cancer targets Voquill (YC P26) - An AI coworker for pathologists that learns your reporting style and produces sign-out ready reports in real time Clara - An AI primary care doctor with a licensed clinician reviewing every medical decision Lumius (YC P26) - Affordable, real-time 3D ultrasound, starting with vascular access

112

7 Comments

Yesterday we showed Qwen3.6-27B running at 105 tokens/sec entirely on an Apple M5 Max. We’re very happy with the result, but ultimately it is an infrastructure metric. The metric I personally care about most is time-to-verified-outcome: the time between a person expressing an intent and their computer completing the work and verifying the result. Users shouldn’t need to watch tokens appear or supervise every step. They should be able to express an intent, let their computer understand the context, reason and act locally, and get back a verified result. Everything we are building across models, inference, and product is about making that distance as close to zero as possible. This is what we mean by ambient AI: not models generating tokens continuously in the background, but intelligence that is continuously available - private, contextual, and immediate, and able to act when needed. The latest benchmarks validate an important part of the technological foundation. The next milestone is to turn that performance advantage into a repeatable consumer experience across a focused set of high-frequency personal and productivity workflows. We are now working to demonstrate how sub-second local intelligence changes the way people use a Mac: from opening applications and completing workflows step by step to expressing an intent and receiving a completed, verified result. Models, inference speed, and runtime are not the end product, they make this kind of product possible. Our ambition is to build the frontier AI lab for the device era.

166

26 Comments

"#OpenAI Scales Back Shopping Plans for #ChatGPT" https://lnkd.in/giyfGmRS "OpenAI staff had realized that while ChatGPT users were researching products to buy in the chatbot, they weren’t using the chatbot to actually help them make purchases" "The change means that instead of purchasing directly from a product listing, a shopper will either need to pay through a retailer’s app or be redirected to the retailer’s own site." ☝ is consistent with the vision for #agenticCommerce that Dhruv Bansal and i had shared in Retail TouchPoints back in October. For my #retail and #ecommerce tribe, the relevance battleground continues on 2 fronts 1) Pushing product, pricing and availability data into AI search for better discoverability 2) Continuing hyper-personalized context on incoming AI Search traffic when your shopper is "redirected to the retailer's own site". A generic category page or PDP will feel jarring when coming from a 1:1 chat experience Winning those battles is why I'm excited about what FERMÀT is building.

40

7 Comments

Groq–Nvidia’s ~$20B licensing deal has generated a lot of buzz over the past few days. I ran a small experiment: I asked Grok, Gemini, Claude, and ChatGPT to independently run Deep Research on Groq’s fair valuation as of Dec 28, 2025, as if they were investment bankers. Two rounds: 1) With full knowledge of Nvidia’s ~$20B offer* 2) Explicitly without that information (*Important caveat: the $20B figure was reported by CNBC citing one of Groq’s investors — it’s not an official statement from either Nvidia or Groq.) What happened? Most models anchored hard to the $20B headline in round one, landing at $18B–$20B. Once the anchor was removed, valuations fell sharply to $9B–$12B. Classic anchoring bias. ChatGPT was the outlier. It stayed relatively stable at ~$9B–$13B, resisting the anchor from the start. Key insights: 1️⃣ ChatGPT resisted anchoring bias. It treated the $20B number as an outlier and grounded its valuation in fundamentals. The other models implicitly baked the Nvidia deal into their conclusions. 2️⃣ Claude was the most self-aware. It explicitly admitted: “The $20B NVIDIA deal anchored my entire analysis upward—classic anchoring bias that any good banker should be aware of but often falls prey to.” 3️⃣ The “Monopoly Tax” observation. Claude also noted: “The $20B represents what a monopolist will pay to stay a monopolist.” My takeaway: AI models are getting remarkably good at market research & financial analysis. But like humans, they’re still vulnerable to cognitive biases—especially anchoring. 📄 Full analysis with all four Deep Research reports: https://lnkd.in/gvAuiZ3U Curious to hear others’ takes: Would you pay $20B for Groq? Also, how do you use Deep Research for tasks like financial or market analysis while actively steering it away from biases?

18

1 Comment

We just unlocked $300K+ worth of startup infrastructure for founders. Most talented founders never get to launch because they lack access to tools and infrastructure. AXI is changing that. Founders can now access $300,000+ in startup infrastructure completely free, including: AWS Activate – $10,000 credits Hostify – Premium hosting plan Vidmoat – 120 free credits + 15% off Studio plan Linear – 6 months free access PostHog – $50,000 infrastructure credits Cloudinary – Startup program access JetBrains – 6 months free + 50% discount Framer – 1 year Pro plan free Cuoral – 6 months free + 50% off for 6 months AWS Technical Support Package PentaTech – Technical consulting Notion Plus – 6 months free And many more resources founders need to launch without capital. Founders simply sign up and use simple prompts to access the infrastructure and resources they need to build. For example: “I’m building a fintech MVP and need cloud hosting, analytics, and developer tools.” AXI instantly recommends the right infrastructure, credits, and tools to get started. Getting started is simple: Sign up at useaxi.com Describe the tool you need using a simple prompt Instantly unlock the infrastructure and resources you need This is part of our mission at Ideation Axis Group to democratize startup infrastructure in Africa. You should not need capital to build your first product. You should only need an idea, ambition, and the right infrastructure. If you're building something, this is for you. Welcome to the future of launching startups.

57

7 Comments