Gaurav Ragtah

Introducing Agentic Culture — an open-source experiment in collective memory and in-context cultural accumulation for multi-agent systems. A hot topic from the recent Andrej karpathy // dwarkesh podcast is that LLMs don't yet have culture. >>> So we built the scaffolding for them to develop one. Agno Agents can now tap into a shared cultural memory — knowledge that persists beyond individual sessions, users or memories. Culture is collective, evolving context that shapes agent behavior over time. Now they can create, read, explore and learn from their culture. You can use the `CultureManager` class to create and manage `CulturalKnowledge` objects. These are stored in your database of choice (SQLite for now) and automatically retrieved as in-context cultural grounding for your agents. To give your agents access to shared culture, just set `add_culture_to_context=True`. That's it. Agents now learn from collective experience. Explore the code and README: https://lnkd.in/e7E3AcJX Agno is open source — come build with us: https://agno.link/gh

98

4 Comments

One of the biggest fears around AI is that it will replace humans. When I spoke with Naveen Jain, he reframed this conversation in a powerful way. AI doesn’t create knowledge out of thin air. It stands on the shoulders of millions of researchers, clinicians, and scientists. Every single day, more medical research is published than any human could read in a lifetime. AI can read it all. Humans can’t and were never designed to. That’s where AI truly shines: not replacing human wisdom, but amplifying it. Helping us see patterns, biomarkers, and insights that would otherwise remain invisible. Used correctly, AI doesn’t remove the human from healthcare, it brings us back to what humans do best: judgment, empathy, creativity, and care. 🎧 Listen to the full episode here: https://lnkd.in/eQjUtZKE

10

1 Comment

India is not behind on talent. It is behind on risk appetite. Building LLMs or foundational AI infra is not about capability. It is about capital. Not just any capital, but capital that is comfortable with long timelines, unclear outcomes, and massive burn. Most of India’s funding still chases near-term traction. SaaS, fintech, consumer. Sectors where returns are easier to model. But AI infra does not play by those rules. It demands belief before proof. The Middle East launching a sovereign fund to accelerate AI through public private bets is not just money. That is intent. India will not build foundational AI by hoping startups will figure it out alone. It needs conviction at the system level. And a willingness to back uncertainty at scale.

56

10 Comments

📖 Weekend Reading 📖 1️⃣ AI: Dwarkesh is not only a great interviewer, but also a very clear writer. His latest recaps his current thoughts on realistic timelines to AGI. What’s most interesting about this piece is that it provides a framework to compare humans against AI, and in doing so gives us a useful way of understanding how humans are so different. “The reason that the Fortune 500 aren’t using [LLMs] to transform their workflows isn’t because the management is too stodgy. Rather, I think it’s genuinely hard to get normal humanlike labor out of LLMs. And this has to do with some fundamental capabilities these models lack.” “The fundamental problem is that LLMs don’t get better over time the way a human would.” In other words, they’re not learning – which goes right to the heart of what makes us humans so different. “If AI progress totally stalls today, I think <25% of white collar employment goes away. Sure, many tasks will get automated.” Ie. Tasks vs. Jobs; agents vs. beings. https://lnkd.in/d8yFZMkQ Because labs are incentivized to release any innovations quickly, we’ll see a broken early version of continual learning (or test time training - whatever you want to call it) before we see something which truly learns like a human. 2️⃣ Fintech: Boston Consulting Group (BCG) and QED Investors published their third Global Fintech Report, arguing that fintech is only 3% of the way there. Key trends include agentic AI built into financial workflows, as well as onchain finance. Payments is the area fintech has penetrated revenue pools most, sitting at 14% https://lnkd.in/db46Vc3S 3️⃣ Economics: I’ve been on an #EconTalk kick this week, prompted by an excellent interview with Patrick McKenzie on credit card economics. This functions as a great primer if you’re looking to better understand credit cards – and payments more broadly. https://lnkd.in/dimakvRD There’s another pair of episodes with Doug Irwin that were excellent: the latest on tariffs, and a previous episode on the gold standard, which remains remarkably salient. Recommend listening to both as a pair. https://lnkd.in/dbztu7ft https://lnkd.in/dvKsuHMM #ai #agi #fintech #payments #llm #crypto #blockchain #tariffs #china #gold #bitcoin

12

1 Comment

The work done at Mira Murati's (former CTO at OpenAI) new company ,Thinking Machines Lab, might be the most interesting thing when it comes to LLMs in a long time (taking the LLM lifetime in consideration). They are trying to solve nondeterminism in LLM inference. One of the biggest problems with using LLMs in real cases. What they found:   • The big culprit isn't "GPU randomness", it's batch variance.   • Core ops can return slightly different numbers when batch size/slicing changes in production, so your output can depend on other users' requests.   • Reductions inside RMSNorm, MatMul, and Attention. What they built:   • Batch-invariant kernels so reduction order stays identical no matter how requests are batched.   • Fixed tiling/no split-K paths for matmul; attention that doesn’t change with cache layout or chunking; RMSNorm reductions kept per-example.   • A deterministic mode for vLLM that swaps these in as drop-ins. 𝗪𝗵𝘆 𝗶𝘁 𝗺𝗮𝘁𝘁𝗲𝗿𝘀:   • Trust & reproducibility for evals, debugging, and compliance.   • Stable A/B tests and benchmarks.   • Better caching and incident response (you can actually replay a bug). If you ship LLM apps and care about correctness and reproducibility, this is a big deal. This is still under development, but solving this will make generative AI much more useful in a lot of cases. The company got valued at $12 billion earlier this summer without any product or revenue, more or less only from Mira's track record. Now they showed that the company can deliver on that valuation. This is what happens when you take one of the smartest brains within AI and tries to solve a real problem. Link to Thinking Machines Lab's blog post: https://lnkd.in/dtX4vVj9

47

1 Comment

Shipped a new personal website / started writing as a forcing mechanism to reflect + democratise any learnings + have a home for research. Links to home and first blog on arXiv pre-print in comments. (thnx Opus 4.6)

32

6 Comments