bogdan nesvit

The LLM era is peaking. The SLM era is just getting started. Why care? 🐧 Tencent just dropped their 0.5B-parameter Hunyuan Small Language Model that can run offline on a smartphone or edge/constrained device — no cloud, no Wi-Fi, no “please hold while your AI is thinking.” 📈 It’s tiny compared to models like o4 (~200B) or o4-mini (~20B), yet it supports a 256K context window and both “fast” and “slow” thinking modes. Translation: it can produce high-quality outputs while living entirely on-device. Our AIOT portfolio companies like Atym, Mimiq and Tripolar Industries (stealth) are primed to take advantage of this wave. 🏁 Why SLMs > LLMs in certain domains: - Speed: Millisecond responses. - Offline operation: Works in connectivity deserts. - Privacy: No data leaves the device. - Focus: Perfect for specialized tasks. 🤖 Industry will be a killer app for SLMs: - Edge AI in factories: Local analysis of production data without risking IP leaks. - Aerospace & automotive: On-device AI guidance for additive manufacturing. - Frontline productivity: Real-time troubleshooting without a network tether. 🔥 Hot take: Within 3 years, most “AI in manufacturing” will not be powered by giant LLMs in the cloud — it’ll be nimble SLMs at the edge. The next AI arms race isn’t about who has the biggest model. It’s about who can make the smartest model that fits in your pocket. What’s your bet?

192

29 Comments

Founders need conviction capital and long-term alignment. Web3 is rapidly maturing as fintechs acquire into the crypto ecosystem. Web3 innovation haven’t slowed and capital is rapidly catching up to where the frontier use cases in Web3 x AI are.... Wonderful report from Jamie Burke, Rumi Morales, Pietro Negri and the Outlier Ventures team. Their growth playbook offers a clear framework for what it takes to support founders through hype cycles into durability: 1/ Grants are not a growth strategy. Founders chasing capital over users burn out before they break out. 2/ Post Web ≠ Web3 hype, We’re entering an intent-driven internet. The convergence of AI x crypto is foundational. AI agents, RWA tokenization, and DePIN are reshaping what scalable utility looks like. 3/ Domain expertise matters from your capital partners, especially those with deep vertical conviction and ecosystem design. At Decasonic, this is what we mean by "enhancement capital." We don’t just invest, we co-build alongside the frontier. And we back founders early before they scale escape velocity. If you’re building the next breakout in AI x Web3, don’t just ask who’s funding you. Ask who’s building with you.

68

5 Comments

$110M raised in 5 months.   A one-year-old startup just raised $85M in a Series A round led by Redpoint Ventures, with participation from 8VC, Shopify Ventures, Autotech Ventures, and others.   That tells us investors believe:   – Workflow-native AI (trained on logistics data + processes) creates defensible moats.   – Adoption curves are faster when AI feels like a teammate, not another SaaS login.   – The TAM is huge. Logistics is a trillion-dollar industry with margin pressure and labor bottlenecks.   Logistics is just one example. Similar vertical plays are emerging in:   Legal → Contract review agents. Finance → AI underwriters, wealth management agents. Construction → AI safety monitors and project schedulers. Healthcare → Ambient clinical documentation, revenue cycle AI.   Augment’s raise is a datapoint in the larger trend:   📌 Capital is moving from broad AI platforms into applied, industry-specific AI.   What AI are you building?   _______________ P.S. My newsletter (Join 20K+ subscribers): https://lnkd.in/dNVbGxXv My reading list: Reading List | https://lnkd.in/dS8XJ9_W

105

9 Comments

From Fusion Industry Association: - Fusion companies raised a total of $2.64 billion in the last 12 months, a 178% rise from last year’s growth. - Total funding for the 53 fusion companies now stands at $9.766 billion, a five-fold increase since 2021. :) :)

42

1 Comment

Remember how Midjourney and Sora blew your mind? Inception Labs brings that same "wow" factor to text and code with diffusion models. I'm proud to announce Mayfield led the investment  in Inception Labs last July, creators of Mercury - the first commercial-scale diffusion LLM that runs 10x faster and at 1/10th the cost of traditional LLMs. The founding team represents the perfect storm of talent: Stanford University professor Stefano Ermon who co-invented diffusion models, UCLA professor Aditya Grover, and Cornell University professor Volodymyr Kuleshov. After working together for over 10 years on AI research, they’ve embarked on the ambitious mission to rebuild the AI tech stack from the ground up fundamentally. Personally, partnering with Inception Labs feels like a homecoming. It brings me back to my Stanford days when my PhD research in video compression led to my first startup, VXtreme (acquired by Microsoft), the pioneer in video streaming. I’m witnessing the same pattern unfold again, but with significantly higher stakes. AI represents a once-in-a-lifetime opportunity to reshape how we work, live, and play. As I see it, Inception Labs' technology is an enabling layer that will propel #CollaborativeIntelligence to new heights. Their approach mirrors human thought processes more closely than traditional left-to-right text generation, resulting in faster, more accurate, real-time language generation with fewer hallucinations. This breakthrough has the potential to democratize AI. With lower costs and reduced computing requirements, Mercury can bring powerful language models to the edge, even to the palm of your hand. The possibilities are truly endless. Are you curious how this breakthrough happened? Watch my conversation with the Inception Labs cofounders, in which they share their journey from academic research to startup and their vision for revolutionizing AI with diffusion LLMs. What areas do you think will be most transformed by diffusion LLMs? #AI #DiffusionModels #Mayfield

954

31 Comments