Articles by Ravi
Activity
30K followers
-
Ravi Mhatre shared thisAI capability is compounding faster than the industry's understanding of how to control it. The frontier is agents: models post-trained with RL to be relentlessly goal-oriented, then given long horizons to pursue goals autonomously. But RL optimizes for proxy rewards, not true human intent. And we still can't verify what goals a model has actually internalized. None of this is new to Dario Amodei. Below is a lecture he gave at Carnegie Mellon in 2016, as a researcher at Google Brain, years before Anthropic existed. Misalignment was his research agenda a decade before it made headlines. That matches my direct experience knowing him. Now he's gone further than anyone in the industry: committing Anthropic to embedded third-party evaluators with employee-level access, and calling for the frontier to be paced so safety work can keep up. A decade of technical work on exactly this problem is, in my view, the best background a frontier lab leader could have. The founders worth backing hold convictions before they're obvious. Dario is one of them. https://lnkd.in/gNvSwhP9 https://lnkd.in/ga9ruTUw
-
Ravi Mhatre shared thisToday, we’re excited to announce our investment in Covenant's $250M raise as they emerge from stealth. At Lightspeed, we believe the future of defense air systems will be driven by cost-optimized hardware. Co-founders Michael Kaufman and Ofir Dayan are answering the demand with Anthem, a long-range strike arsenal that offers heavy payload and long range. With 200+ flights already tested, Covenant’s full-system design aims to flip the traditional production process. Design and iteration through flight has been core to Covenant’s culture as they scaled from a flight every other week to multiple times per week. We are thrilled to partner with Michael, Ofir, Abby Denburg and the Covenant team on their mission to bring an affordable, scalable answer at the strategic tier of deterrence. CC: Raviraj Jain, Prathik H..Ravi Mhatre shared thisThe West has a deep precision strike gap. We built Covenant to help close it at scale. Today, we’re emerging from stealth and introducing Anthem: a heavy-payload deep precision strike solution designed for mass production. In two years, we’ve flown Anthem more than 200 times, secured approximately $150 million in orders including contracts with the U.S. Army, U.S. Navy, and key international allies - and raised $250 million backed by Andreessen Horowitz, Founders Fund, Lux Capital, Lightspeed, 8VC, Aleph and Altimeter. Beginning in 2027, we plan to deliver Anthem in the thousands from facilities in Dallas, Germany, and Israel. Now we deliver at scale. Come build with us: covenant.com
-
Ravi Mhatre shared thisNir Zuk founded Palo Alto Networks on the belief that security had become too fragmented. Over twenty years, he collapsed single-purpose security appliances into one single platform that helped define the industry. Now he's doing it again, this time at the data layer, with Cylake. We believe Nir is an exceptional founder, and we’re backing Cylake by being the lead on their $245M financing. Right now, large organizations can run well over 30 security vendors, each covering one function with its own data silo, dashboard, and bill. The landscape is often worse for regulated industries that can’t use cloud platforms because of air-gapping and data sovereignty requirements, leaving them with outdated, clunky tools. Cylake's answer lives inside the organization's own walls: a complete, appliance-based cybersecurity stack that provides all security functions. It comprises CPUs, GPUs, memory, and storage, and functions like a private cloud in the customer's data center, or in facilities managed by a managed service provider (MSP), where every stream of data flows into a single data lake, with functions switched on as needed. With AI adoption expanding at pace, cyberattacks becoming more sophisticated, and the cost of security measures increasing, Cylake not only allows companies to control their security landscape as a whole, but also helps ensure this can be done in a more cost-effective way. We believe that few founding teams could take this on. Nir is joined by Wilson Xu, who ran engineering at Palo Alto Networks for nearly a decade, and Ehud S., who co-founded SentinelOne. This is a team that has proven this formula once already, at category-defining scale, and we’re thrilled by their ambitions to do it again. Congratulations to Nir, Wilson, Udi, and the whole Cylake team. https://lnkd.in/g-dHH7j9 CC: Guru Chahal
-
Ravi Mhatre shared thisWelcome to the Lightspeed team Alex!Ravi Mhatre shared thisToday, I’m starting my next chapter as a Partner at Lightspeed. The day after GSB graduation, I met Ravi Mhatre for the 1st time after a casual meet with Bejul Somaia. After that, the conversation never really stopped. It became obvious pretty quickly that this was a once in a lifetime opportunity: the chance to learn from one of the best to ever do it, on a team at the frontier of AI. Bejul and Ravi, thank you for your conviction. Sebastian, Will, Faraz, Sachin, Guru, Raviraj, and the whole Lightspeed team: I can't wait. Shared values, different spikes. Lightspeed embodies ultra-competitive with a heart a gold. I'll be investing across stages in AI, from specialized models to AI apps to defense and reindustrialization to robotics. My goal is simple: help make Lightspeed the first call for the best founders in the world, when they're raising and when they have a problem. Then, doing everything in my power to help them fulfill their life’s mission. If you're building, or not even building yet, I'd love to meet you. Investors who want to collaborate too. Based in NYC, in SF all of September. Contact below. (203)832-5510 lapolice@lsvp.com X: @alex_lapolice
-
Ravi Mhatre shared thisSkild AI's S1 just demonstrated it can learn a new workflow from a single video in context, with no task-specific post-training. That is what the GPT-3 moment in robotics looks like and Skild has shown the world's first demonstration of it over long horizon human labor tasks. The case for a general-purpose robotic brain gets stronger as the same model adapts across more robots, tasks and environments, and in an exponential environment decade-long outcomes can arrive much sooner than you'd expect. If this kind of generalization keeps compounding, advances in the foundation model will carry into an ever wider range of physical work, and we move closer to ubiquitous general-purpose robots that can perform any task, anywhere, set up with nothing more than a simple video prompt. Exceptional progress from Deepak Pathakak, Abhinav Gupta and the entire Skild AI team. Extremely excited to be your partners on this journey. CC: Raviraj JainRavi Mhatre shared thisAnnouncing S1, our new foundation model that learns from one example. It can be taught 10-minute long tasks that it has never seen before, from one video prompt without any fine-tuning. How? S1 learns new tasks like a language model. You prompt it with a video demonstration, and it outputs robot actions to complete the task in any environment and in any embodiment. By composing skills learned in pretraining or creating new ones, it can make coffee, pot a plant, fry pancakes, and more – tasks never seen during pretraining. The first time S1 flipped a pancake, we assumed pancake flipping must have been in its pre-training data. We searched our whole pre-training data of a million hours and found no examples of flipping. S1 inferred the out-of-distribution task from one video prompt. S1 does not blindly replay the video demonstration. Instead, it displays common-sense understanding that goes beyond the video prompt. It can withstand perturbations and improvise upon mistakes, even if the human didn’t. At times, S1 executes with more precision than the human in the video prompt Compared to conventional VLAs, S1 is a step-change improvement. For known tasks, S1 can match the performance of language-prompted VLAs. For novel tasks, S1 exponentially outperforms any existing language-prompted VLAs as we scale the pre-training. We found that to match the accuracy that S1 can achieve with just one example of prompting, current VLA models would need to be post-trained with 50-100 hours of data collection followed by fine-tuning! We believe this is a promising direction towards establishing scaling laws for robotics. We're excited to deploy S1 with our limited industrial partners today, rolling out to more customers over the next few months. S1 is the first light on a new path towards general intelligence grounded in the physical world. Technical blog: https://lnkd.in/g7ANeDkb
-
Ravi Mhatre shared thisProud to deepen our partnership with Castelion and invest in their $1B Series C. Built by a team that scaled hardware production at SpaceX, Castelion is applying a similar playbook to hypersonic weapons: design to cost, iterate fast, produce at scale. Lightspeed led their Series A a year and a half ago, before Blackbeard was a fielded, production-representative strike system. Since then Bryon Hargis, Sean Pitt, Andrew Kreitz, CFAand the team have taken a clean-sheet hypersonic weapon from concept to program of record in 2.5 years. The deeper story: building that first system meant building the architecture and manufacturing base for high-rate production. Every new program the team takes on now starts from proven engineering rather than a reset. We believe that's a compounding advantage, and why what @Castelion can build is much broader today than when we first invested. Our conviction in this team has grown with it. CC Raviraj Jain, Prathik H.ain, Prathik H. Read more: https://lnkd.in/ee_27G9tExclusive | Carlyle Joins $1 Billion Investment Round in Missile Startup CastelionExclusive | Carlyle Joins $1 Billion Investment Round in Missile Startup Castelion
-
Ravi Mhatre posted thisJeff Dean has built the foundations that successive eras of modern computing have stood on, from web-scale infrastructure to scaling deep learning itself. Exceptional people inspire great talent and the founding team around him is among the strongest I have ever seen in over 25 years of investing. Lightspeed is honored to join them from the start of this journey.
-
Ravi Mhatre shared thisConversations about humanity's road to superintelligence are finally taking centre stage. The technology is part of it. But it is also a shift in thinking, a collective willingness to take seriously ideas that were, until recently, considered fringe. What we're building towards is an era of discovery that takes us forward. Flashes of brilliance that push the frontier of what's possible. Move 37 was an early glimpse. In 2016, AlphaGo played a move so unexpected it brought a professional Go player in the audience to tears. Not because of what it meant for the game, but because a machine had found something humans hadn't in thousands of years of play. Joining David Silver in conversation at Founders Forum Group's event this week was energising. Not just for what he's building at Ineffable Intelligence, but for how seriously he takes the responsibility of building these systems safely, so the benefits reach everyone. David, it was a privilege to share the stage with you and am looking forward to partnering with Ineffable on this journey.
-
Ravi Mhatre reposted thisRavi Mhatre reposted thisClaude Fable 5 is available today! It's a new moment for AI: a Mythos-class model, the most capable class of systems we've built, now safe for general use. It's already changed how we work internally, and I'm excited to see what you all do with it. Every request runs past safety classifiers trained to detect misuse in cybersecurity and biology. When one triggers, your request is answered by Opus 4.8 instead. More than 95% of sessions never see a fallback, and 1,000+ hours of external red-teaming produced no universal jailbreak. In terms of benchmarks, Fable 5 reached 80.3 on SWE-bench Pro (Opus 4.8 scores 69.2), 88 on Terminal-Bench 2.1. State-of-the-art on nearly every coding benchmark we tested. But the benchmarks undersell how truly capable it is. Fable holds quality deep into long, hard problems where most models degrade. It verifies its own work. It catches what others miss, things like root-cause bugs that no other model had surfaced. Base44 found it "much deeper and better at one-shotting full apps"; at Genspark it came out #1, winning head-to-head against every model they tested. Internally, writing code stopped being the slow part a while ago — Anthropic engineers on average shipped 8x as much code per quarter as they did compared to 2021-2025 — Fable pushes the bottleneck further toward verification and review. We're excited to make all of that available today for every use case outside bio and cyber. For API customers, here's how we've imagined fallbacks: pass a fallbacks parameter and the Messages API retries any blocked turn on Opus 4.8 server-side — even mid-stream, keeping the partial output. We think of this as a graceful handoff between models, and we'll be iterating on the design with the community. Moments like this are worth doing right. We're making sure it's safe, but the classifiers may be annoying at times. They're tuned conservatively, and false positives will keep coming down. Read more here: https://lnkd.in/giBEAAcP
-
Ravi Mhatre liked thisRavi Mhatre liked thisWe just shipped Claude Sonnet 5.5! Sonnet 5.5 is the model to help you get more work done, faster. It runs 30%+ faster than Sonnet 5 and typically needs far fewer tokens, so most work costs up to 30% less A good complement to the more complex work done by Opus 5.5, Sonnet 5.5 is strong at scoped everyday work tasks, fixing bugs, and has a great eye for design. Haiku 5.5 will round out the 5.5 family in the coming weeks! More in the blog: https://lnkd.in/gS7AU4RB
-
Ravi Mhatre liked thisRavi Mhatre liked thisAnthropic is calling for a global effort to pace AI’s progress and give experts additional time to better manage the risks of increasingly capable models. To make this effort as effective as it can be in helping humanity build AI safely, we think three things should happen. First, every AI company should commit to giving third-party evaluators employee-level access to verify safety, report problems, and guarantee compliance with the slowdown. We’re committing to doing this. Second, AI companies operating in democracies should work together, with government support where necessary, to set common safety standards and limits on runaway AI progress. Third, the U.S. and other democratic governments should try to coordinate with authoritarian governments pursuing advanced AI, recognizing that verification here is critical. None of this is easy, but we believe we need to try, and if we manage to secure the benefit of time, we need to use it well. I hope you’ll read more here: https://lnkd.in/gxkPj9FC
-
Ravi Mhatre reacted on thisRavi Mhatre reacted on thisWe grew from ~0 to $500M ARR, adding $250M last year alone while being EBITDA profitable. 500-word post on the most important growth tactic that worked for us. What got us from 0-$400M ARR in the US and works in every country: We run a 90 sec video trailer of an audio drama as an Ad and ask users to download the app if they're interested in the rest of the story. Our core insight after spending >$100M on user acquisition is: If the clickthrough rate (CTR) for an ad goes from 2% to 2.25%, our customer acquisition cost (CAC) decreases by ~ 30%. We remodelled our system around this insight and built an AI-first 2.25% CTR ad manufacturing machine that works in every country. For every new country, we take our hit shows -> use LLMs to extract and most intriguing moments -> Write a 5-min script combining all the best parts. The first 60 seconds has to have a hook every 5 seconds and needs to end with a crazy cliffhanger to force a download mid-scroll. If a marketing video is not hitting our benchmarks (2.5% CTR and 55% 3-second through-play), a creative director gets involved to change the hook or cliffhanger to get the numbers there. AI lets us make 1,000 Ads per show, and in total we do ~17.5k Ads per month. When a business creating scales from 1k to 10k Ads, the normal thing is for CAC to skyrocket. But with what I just shared, we 7-8x'd our User Acquisition budget without increasing our CAC materially. It took us 8 months to figure this out, but then the timeline from 0-$10M in every country got shorter and shorter: US revenue grew to $25M in 19 months, Germany to 21M in 11 months, France to $10M in 3 months, _______ Pocket Entertainment (Pocket FM + Pocket Saga) has become the largest AI entertainment platform. Our Bet: In ~3 years we will be able to produce Naruto-like series for $1000. Netflix has to spend $17B to find 100 Blockbusters per year. What happens when you can produce 1M high quality shows for $1B? The streaming war caused a $300B reallocation of market cap. The AI entertainment war may be a $1T+ reshuffling. Unlike other AI creator tools like Runway or Midjourney, we own both the supply and demand side of AI content. We've attracted 550k+ writers who are producing an annualized 2.5M hours of content every year. This pairs with 137B+ minutes streamed. Because of this, growth is accelerating as we scale further. We have many S-curves inflecting at the same time: - AI produces better and more ads -> scale faster in new countries - Writers + AI produce more shows -> more blockbusters - Localisation -> more reach per show - Audio to video format expansion -> larger TAM -> more creators Every aspect of our business is designed to improve another, and everything is aligned so that the number of blockbusters continues to rise. I'm excited for what's ahead. If you're interested in building at the intersection of ai, tech and entertainment, DM me.
-
Ravi Mhatre liked thisRavi Mhatre liked thisToday, we’re deepening our partnership with Mistral by participating in its Series D. When we first partnered with Arthur Mensch, Timothee Lacroix, and Guillaume Lample for Mistral’s inception round, we believed open models could give enterprises more choice and control over how they use AI. Fast forward to today, those needs now extend beyond the model. As AI moves into critical workflows, large enterprises and governments increasingly need to shape models around their own knowledge, deploy them within their security and governance requirements, and control where and how they run. Mistral now spans that full stack, pairing enterprise AI deployment with an AI cloud on owned infrastructure. Customers can train models around proprietary knowledge, build and run agents in production, and use Mistral’s infrastructure for training and inference. Its teams work directly alongside customers from the first use case through production. Proud to have backed Arthur, Timothée, Guillaume, and the Mistral team in every round since inception as they build AI that the world’s largest organizations can make their own. CC: James Ephrati, Ravi Mhatre, Antoine Moyroud
-
Ravi Mhatre liked thisRavi Mhatre liked thisWe’re thrilled to be the lead on Cylake’s $245M financing. For the past two decades, much of cybersecurity has been built around the cloud. Yet, some organizations with the most sensitive data were never able to make that leap. This includes governments, banks, and critical infrastructure operators. Cylake is building around that reality. Founded by Nir Zuk, the company's AI-native security platform runs on-premises, aiming to give organizations visibility and control over their data without requiring it to leave their designated environment. Twenty years ago, Nir founded Palo Alto Networks and helped reinvent the firewall and move cybersecurity towards the cloud. With Cylake, he’s returning to the organizations that couldn’t follow that shift, rebuilding the stack around their critical needs. Congratulations to Nir Zuk, Wilson Xu, Ehud S., and the entire Cylake team! CC Ravi Mhatre, Guru Chahal
-
Ravi Mhatre liked thisRavi Mhatre liked thisCyber threats are happening at AI speed, and your recovery plan must keep pace. Rubrik CEO, Chairman, and Co-Founder Bipul Sinha caught up with Jim Cramer on Mad Money to discuss #AI security, agentic cyber resilience, and the needs of our customers amid AI transformation. Catch the full interview 👉 https://rbrk.co/46IFsUr
-
Ravi Mhatre liked thisRavi Mhatre liked thisModern cybersecurity has an architectural problem. The organizations with the most to protect get the least protection. Their data cannot leave their control, and the industry’s best tools live in a cloud they cannot use. So they make do with less. Less data, less AI, less defense. We are rebuilding the entire cybersecurity stack to end that dilemma, starting the other way: your requirements, your premises, your data going nowhere. A complete cybersecurity platform that runs all cybersecurity functions, hardware included. Today, we raised another $245 million to keep building, with participation from Lightspeed, Picture Capital, and Redpoint. Read the announcement: https://lnkd.in/gwSnGxCT
Experience & Education
Honors & Awards
-
Midas List of Top Venture Capitalists
Forbes
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content