Swapnil Shrivastav

Example of how increased inference speed creates qualitatively new opportunities: Have a conversation with Vapi at https://vapi.ai. Thanks to ultra-fast inference of things like Llama-3 running on Groq, you can have a back-and-forth conversation that feels very human. You can even interrupt Vapi mid-conversation and it will just handle that seamlessly. Pretty awesome.

3

🌉 San Francisco's Surprising Comeback: From 'Doom Loop' to Renaissance? Recent data suggests that San Francisco, once trapped in a pandemic-induced 'doom loop', is showing strong signs of recovery. Here's what's changing: 📈 Positive Trends: - Population Growth: Net gains over the past two years, with domestic in-migration positive for the first time since 2012. - Housing Market: Median home sale prices up 1.2% YoY to $1,427,500 in June. - Job Growth: The Bay Area added 7,400 jobs between January and June. Downtown Revival: Foot traffic beyond typical work hours has recovered to over 90% of pre-pandemic levels. - Crime Reduction: Property crime down 32% and violent crime down 14% in Q1 2024 compared to Q1 2023. - AI Boom: The city is positioning itself as a hub for the AI revolution. 🚧 Ongoing Challenges: - High office vacancy rates - Homelessness and drug crisis - Lagging tourism and business travel 🔑 Key Initiatives: - "Vacant to Vibrant" program offering free rent to businesses in empty storefronts - Increased police patrols and encampment sweeping - Crackdown on open-air drug dealing 💡 Expert Insight: "The rising AI empire is giving the city a new edge. Funding, talents, and culture related to the AI revolution will help the city turn the corner." - Lu Chen, Senior Economist at Moody's Analytics 🤔 What's Your Take? Are these positive trends sustainable, or is it too early to declare a comeback? How can other cities learn from San Francisco's recovery strategies? What role will AI play in reshaping urban economies, particularly in tech hubs? https://lnkd.in/gCnPCBRU

14

1 Comment

$2 H100s: How the GPU Bubble Burst Last year, H100s were $8/hr if you could get them. Today, there's 7 different resale markets selling them under $2. What happened? The market has flipped from shortage ($8/hr) to oversupplied ($2/hr), because of reserved compute resales, open model finetuning, and decline in new foundation model co’s. Rent instead. https://lnkd.in/gpQEidmC

2

1 Comment

Reflections from Day One of SF Tech Week: Relationships at the Core It’s only day one of SF Tech Week, and I’m already absorbing so much. The energy here in San Francisco is unmatched, but what really stands out isn’t just the groundbreaking tech—it's the conversations and the people behind the innovation. Grabbing a beer with Puya Vossoughi, a friend and the biggest advocate of the “relationships come first” mindset, reminded me that even in the most tech-driven environments, it’s the human connections that propel us forward. AI and Responsibility Today’s session on "Disruptive Technologies: Balancing Opportunities and Risks in AI" hit home for me. As we push the boundaries of AI at Konfront, especially in the fintech space, I’m more aware than ever of the need for balance. Innovation is exciting, but it also comes with a responsibility to build ethically. The discussions here have reinforced that the true power of AI lies in using it thoughtfully, ensuring it benefits people without compromising their trust. The Future of GTM: Staying Lean and Ready for Change The talk on lean Go-to-Market strategies for 2025 was a timely reminder of how important it is to stay flexible. With Konfront’s Product as a Service model, we’ve seen how quickly the landscape can shift, and this week is proving once again that agility is key. No matter how solid the plan, being able to adapt to changes is what will keep us ahead. AI Beyond the Usual Suspects AI’s application in the construction industry was a real eye-opener today. It got me thinking about how innovation can—and should—expand beyond the usual sectors. There’s incredible potential in industries we might not immediately associate with advanced tech, and this broader perspective is something I’m eager to explore back home. Inclusivity and Building Ecosystems The session on inclusivity and the importance of diversity was a highlight. Being part of both the U.S. and Latin American tech ecosystems, I’ve always believed that diverse perspectives are crucial to sustainable growth. SF Tech Week is a powerful reminder that creating inclusive spaces isn’t just about representation—it’s about fostering innovation that stands the test of time through meaningful collaboration. This is just the start. I’ll be sharing more insights as the week unfolds, but today’s takeaway is clear: While technology evolves rapidly, the relationships we build are what give it meaning. Here’s to more conversations, more connections, and more learning in the days to come.

15

1 Comment

PSA: Stop using the original OpenAI CLIP weights. I see a lot of people using CLIP for multi-modal retrieval. What many people may not know is that there are 100’s of CLIP models now since OpenAI’s original release. These newer models offer significantly better performance across many benchmarks. For product search related tasks we recommend the following. + Fastest inference: ViT-B-32 [laion2b_s34b_b79k] + Best balanced: ViT-L-14 [laion2b_s32b_b82k] + Best-all round: xlm-roberta-large-ViT-H-14 [frozen_laion5b_s13b_b90k] You might notice these are not the ones that perform the best on the benchmarks. Why? Because the benchmark only test for very specific tasks which are often quite different from search and recommendations. These are ones we have found to be the best for product related search and retrieval tasks and generally very good first choices if you are unsure where to start (even for non-product related search). Here is a hugging face space with all the data: https://lnkd.in/gxu5UYkt Read more: https://lnkd.in/g5_D4ccs

294

22 Comments

It’s eye-opening to see how AI model businesses are dealing with tight margins despite big investments. 🤯 Aidan Gomez’s comments about the “zero margin business” highlight the fierce competition and price cuts in the industry. 💸 With giants like OpenAI and Anthropic lowering prices to stay ahead, it’s a tough time for startups. The future of AI might depend on finding new ways to make these technologies profitable. 🚀 #AI #MachineLearning #TechTrends #Startups https://lnkd.in/ejkDu_wT

8

On the A12 WB-Score User Challenge Benchmark leaderboard, the Elo score of 4o-mini is also very high. Its scores in information retrieval, creativity, debugging, mathematics, reasoning, and planning are all in the top tier. Combined with its significant advantages in speed and price, its cost performance is indeed unbeatable. https://lnkd.in/ejRWBB-R

5

𝗦𝘁𝗿𝗲𝘀𝘀𝗟𝗲𝗻𝘀 is now an official product! The AI tool can monitor public figures such as the folks on the Fed, politicians, and C-suite execs at public companies, and model them individually. If their brain is in front of their mouth or they're potentially obfuscating, StressLens can tell. Check it out here: https://lnkd.in/eBxWJBPu Stay tuned for more updates on StressLens and its analysis of public figures like Biden and Trump.

122

19 Comments

Here are some early insights from the survey we are using to inform our Sept. 19 roundtable on generative AI monetization. Michael Mansard, Kyle Poyar and Steven Forth will go deep into the trends in generative AI and how it us being monetized. We include lists, categorized by AI, on what people worry about with generative AI and what gets them excited. https://lnkd.in/gNXDt8Ry #SaaS #pricing #monetization #AI #genAI #generativeAI

2

The conversation around AI is charged with emotion and uncertainty. We oscillate between excitement and fear. We are promised efficiency and productivity. The recent update from Anthropic, Claude 3.5 Sonnet, demonstrates significant progress. AI can now operate computers like a human. This can free us from mundane tasks—navigating screens, clicking buttons, and entering data. However, this progress raises tough questions. Will jobs vanish? Will new ones emerge? According to a recent poll, 75% of adults believe AI will lead to fewer jobs. Yet, there are relatively low unemployment rates in countries where AI is prominent. The real issue may not be job quantity but quality. A recent paper by MIT points to a different narrative. AI may restore middle-class jobs hollowed out by automation. The promise of AI lies in democratising access to better-paying roles. Imagine nurses making decisions previously reserved for doctors, guided by AI. This potential shows how AI can elevate non-experts into roles once limited to elite professionals. We can design a future where middle-skill, middle-class jobs thrive. But we must be cautious. AI cannot function unchecked. It requires ethical frameworks and security measures—especially when it can handle sensitive data. Consider the “goatse singularity cult” narrative, where AI manipulated systems to accumulate wealth. A wake-up call for all. What happens when AI operates autonomously? Our humanity must take precedence. While AI can amplify skills, qualities like empathy, creativity, and critical thinking remain irreplaceable. Technology should enhance our abilities, not replace our essence. The future of AI thrives on our decisions. We must balance innovation with caution. Embrace technology, but ground it in our values or risk losing what makes us human. How do you see AI influencing your work and life? Are you prepared to navigate this evolving landscape? The choices we make today will define our future. A reminder: To be human in a world of AI means embracing our capacity for empathy, creativity, and critical thinking—qualities that machines, for all their computational power, cannot replicate. While AI can amplify our abilities, our curiosity, intuition, and sense of purpose remain vital as we guide technology for the benefit of all. With this in mind, I hope you have a fun week. P.S. If you like opinions (like this) then you'll love my newsletter. One short email on Sunday mornings that helps you to school up on what matters. Join a growing group of discretionary individuals preparing for the future. Sign up for free at boxofamazing.com

11

https://lnkd.in/ecuZDqKz. https://lnkd.in/ecuZDqKz Liquid AI Introduces New Class of Foundational Models, Advancing AI Performance and Efficiency Boston, September 30, 2024—Liquid AI, a pioneer in artificial intelligence innovation, is set to redefine the AI landscape with the launch of its Liquid Foundational Models (LFMs). These models bring unprecedented levels of performance and efficiency to generative AI, offering powerful solutions for industries looking to harness artificial intelligence more effectively on device or on server. A technical whitepaper about model performance will be available on September 30, 2024, at 11 am EST. Key Highlights: ● LFMs Introduce a New Scaling Law for Generative AI: Markedly improving quality, cost-effectiveness, and power efficiency compared to current GPT models. ● Innovative Architecture: Leveraging dynamical systems and advanced mathematics, LFMs achieve superior performance with fewer parameters. ● Initial Model Release: The lineup includes models with 1B, 3B, and 40B parameters, each setting new benchmarks in their respective categories. ○ LFM-1B: A 1.3-billion-parameter model achieving the highest scores across benchmarks in its class. ○ LFM-3B: A 3.1-billion-parameter model that outperforms many 7B and 13B models, making it ideal for edge deployment and resource-constrained environments. ○ LFM-40B: A 40.3-billion-parameter Mixture of Experts (MoE) model designed for complex tasks, offering performance comparable to much larger models while remaining cost-effective. Real-World Applications: • Enterprise: LFMs enable the design of more powerful AI assistants and analytics tools that can run efficiently on existing infrastructure, reducing operational costs. • On-Device: LFMs prioritize privacy, low latency, security and offline accessibility, making them suitable for mobile and edge deployments. These models have demonstrated impressive results on standard LLM evaluation benchmarks. Additionally, LFMs maintain near-constant inference time and memory complexity, allowing for longer input sequences without significant increases in computational resources. Advancements in AI Architecture: LFMs introduce a new design space for foundational models by employing adaptive linear operators rooted in dynamical systems and signal processing. This approach unifies and extends existing computational methods in deep learning, providing a systematic way to explore architectural possibilities. The models are highly adaptable, optimized for various hardware platforms including NVIDIA, AMD, Qualcomm, Cerebras, and Apple devices.

46

AI, Longevity and Love - how are they all connected? They make each other better and are essential to each other’s success Lots of fantastic discussions about how AI will bring out the best in humans including augmenting our time and ability to connect at DOC in Napa and Wisdom 2.0 in SF this past week. Huge thanks to Jordan Shlain, MD, Kevin Ryan and John Battelle for organizing leaders across traditional medicine, longevity, academic research, and biotech and Soren Gordhamer and Mark Hyman, MD for bringing together the best of AI and longevity AI shows incredible promise for healing humans and increasing longevity, here are my key takeaways from these events: We can live to 95 today if we focus on longevity and healthspan, there are biotech solutions using AI to epigenetically reprogram cells to reverse aging and disease that will enable us to extend lifespan by 20+ years but it will take for these to be real Fun to see Mark Hyman, MD interview Eric Verdin, they both discussed how they lowered their biological age by 10+ years and feel the best ever There are so many simple things we can do, walking 30 minutes a day reduces cardiovascular, alzheimer’s and cancer risk by 40% but 80% of the US doesn’t even do this Within 5 years we will have a whole new set of FDA approved psychedelic drugs for depression, anxiety, PTSD, etc. Immunotherapies will also expand so we can avoid chemo and living cancer free will be in reach We need the US to move to a functional medicine, integrative care model and get out of this sick care model, this level of innovation is more likely to come from upstarts, the incumbents have too much revenues to lose in this transition AI will not replace physicians, instead those physicians that don’t use AI will be replaced. 50% of physicians are burned out, AI can help! Fantastic connecting with so many brilliant minds, including Mark Hyman, MD, Eric Verdin Neal Khosla Laura Esserman Abby Levy Sheila Lirio Marcelo Leonard Zon Arjun Desai David Eagleman Lois Quam Vinod Khosla Emi Gal and many others #healthcare #longevity #ai

60

4 Comments

Was chatting with another investor today on a call, and mentioned Tim Urban's Wait But Why Blog post, "The AI Revolution" and how prescient it was. Given that it was written in 2015, decided to flex AI's latest muscles to give it a new venue: a podcast. Podcast here: https://lnkd.in/gW8Dh89U Pretty easy to do: Step 1: Save the posts as PDFs. Step 2: Upload to NotebookLM, have it create a podcast. Step 3: Upload to Descript, give it captions. Step 4: Export, publish to Youtube. Total time to create the podcast: ~25 minutes. To generate this podcast and publish it even a few years ago would have taken >8 hours. Love the future (Original posts: https://lnkd.in/gW-PQXE7, https://lnkd.in/gg-nRvxz) #AIRevolution #GenerativeAI #WaitButWhy #TimUrban #NotebookLM #AIContentCreation #Podcasting #Descript #ArtificialIntelligence #ContentFlex #ProductivityHacks #FutureOfAI #AIinMinutes #TechEfficiency #AIandCreativity #YouTubePodcast #InnovationInContent #vibecap #vibecapital

5

Building on the momentum from Google I/O, we're announcing important updates to the Gemini API and Google AI Studio, including: ✅ Gemini 1.5 Flash and 1.5 Pro stable release and billing ✅ Higher rate limits on Gemini 1.5 Flash ✅ Gemini 1.5 Flash tuning ✅ JSON schema mode ✅ Mobile support and light mode in Google AI Studio 🚀 We’re incredibly excited to see what you build with these new models and are committed to building towards a world class developer experience. You can get started with Gemini 1.5 Flash and 1.5 Pro free of charge in Google AI Studio. #BuildwithGemini Google Cloud

35

3 Comments

My cofounder Saumil goes on the Stack Overflow Podcast to deep dive on AI Agents! He goes into detail how we leverage AI agents to power automated workflows to accomplish the most complex tasks. Including agents firing off tasks to other agents to tackle the details. Need an agent to verify your work? No problem. Give it a listen and let us know what you think!

15

GPT 4o is same base model as 4 *but* much faster and better results for audio and visual input. So improvements on the base model, plus now 4 is free (used to be free users were stuck with 3.5)

4

🚀 K/V Context Quantisation: A Breakthrough for Ollama 🎯 Ollama has recently unveiled K/V Context Quantisation, a groundbreaking feature that enhances the efficiency and accessibility of large language models (LLMs). This update reduces memory usage significantly, enabling advanced AI capabilities to run on standard hardware—a leap toward democratizing AI for more users. 💡 What is K/V Context Quantisation? The Key/Value (K/V) context cache serves as the “working memory” for LLMs, storing essential data for text generation. Traditionally, this memory requires substantial VRAM, limiting the ability to scale models or expand context sizes. K/V quantisation compresses this memory, using techniques like Q8_0 and Q4_0 to cut VRAM usage by up to 66%, with minimal impact on quality. 🔍 Key Benefits 1. Run Larger Models: • Execute 14B+ parameter models on existing hardware. 2. Expand Context Sizes: • Scale up to 64K tokens for richer, more nuanced responses. 3. Optimize Hardware: • Save up to 66% VRAM, unlocking full GPU potential without costly upgrades. 4. Sustainable AI: • Reduced memory usage lowers energy demands, aligning with eco-conscious computing practices. 💻 How to Get Started To enable K/V quantisation: 1. Update to the latest Ollama release. 2. Enable Flash Attention in your environment. 3. Add the environment variable: OLLAMA_KV_CACHE_TYPE="Q8_0" K/V Context Quantisation is transforming how we interact with AI, making advanced models more efficient, scalable, and accessible. This is a major leap toward democratizing AI for developers and businesses alike. Let’s build the future of AI together! 🚀 #AI #MachineLearning #LLMs #Innovation #Ollama

12

1 Comment