Yossi Matias

Barbarians at the Gate (https://lnkd.in/gkEBX79s) is a very interesting new paper from some great folks at Berkeley, with some exciting results showing the potential for AI in systems research. But I think the authors aren't quite asking the hardest problem about where this takes systems as a field. I wrote a new blog post about it: https://lnkd.in/gxS3s4v4 The core point of the paper is that AI can be effective at automatically driving the 'number go up' part of systems research. Given an evaluator, AI can automate progress on implementation. I believe this is 100% correct. The paper also talks about how more focus will have to go into developing evaluators. Again, I 100% agree, but think they're understating their case. Building evaluators (basically tests) is, I believe, going to become the focus of systems and software over the next decade. Which leaves researchers with the hard problem of choosing which problems to go after. That's never been easy in academic systems, and is only going to become more of a challenge. But the best systems research is driven by choosing problems well! This vision for AI could lead to lots more of the least interesting systems research. The "number go up" kind. Which could make it harder than ever to identify the truly innovative and visionary work. We're already seeing that elsewhere in science. So, mostly, I think the paper is right, but not yet asking the biggest question: what should systems research be?

276

12 Comments

Back in 2023, TGI has initiated a movement where optimized inference engines rely on the model architectures from transformers as the source of modeling truth. This approach is now adopted by downstream inference engines: vLLM, SGLang, as well as local engines with inter-compatibility such as llama.cpp or MLX. Given we at Hugging Face contribute to those community runtimes, we decided to switch text-generation-inference itself to maintenance mode. Thanks for all the tokens 🔥 Hat/tip Lysandre Debut

480

32 Comments

For the past few years, AI headlines have centered on training - bigger models, larger clusters, record-setting compute. But a structural shift underway is toward inference. As AI moves from experimentation to deployment at scale, inference will define responsiveness, energy efficiency, cost economics and real-world utility. This is no longer just a scaling problem. It’s an architectural one. The next chapter of AI will be written by those who can deliver intelligence efficiently, reliably, and sustainably at global scale.

167

22 Comments

Generative AI’s greatest potential lies in the physical world, where the ability to simulate, design, and iterate faster can reshape trillion-dollar sectors. That’s why we’re thrilled to announce that N47 is leading Luminary Cloud’s $72M Series B. The Luminary team is pioneering the Physics AI Model Factory, underpinned by a GPU-native physics simulation platform. This technology is designed from the ground up to unleash the performance of modern GPUs and generate the high-value datasets that power next-generation AI. Now, engineers can run hundreds of design iterations overnight instead of waiting days. AI models trained on physics-grounded data can suggest new aerodynamic designs or optimize energy systems in ways previously impossible, and companies across industries can accelerate their innovation cycles while reducing cost. With co-founders Juan Alonso and Pete Schlampp, Luminary has the technical depth and operational experience to define the new era of Physics AI. We’re proud to partner with them as they create the future of simulation and redefine what’s possible in the built world. Learn more about Luminary Cloud’s vision here: https://lnkd.in/gbjeWrvy 

119

6 Comments

Big news that Waymo is expanding to more cities including Seattle and Denver 🚗🤖 It’s easy to forget that Waymo is an early example of an advanced AI agents in production, operating at scale, in a high-risk environments and it works reliably! Why? Because of countless hours of training and hardening in simulation before hitting real roads. Enterprise AI agents face the same reality. If they’re expected to operate in risky or high-stakes environments, they need robust simulation to prepare. At Veris AI, we build high-fidelity simulations so AI agents can learn and expand safely, like a Waymo.

53

2 Comments

Here’s a truth that often gets missed: becoming an AI-first company at the enterprise level is still really hard. It requires making big investments in R&D and specialized talent and betting on still-evolving frameworks and tools — and having the commitment to push the field forward yourself where the right tools or integrations don’t yet exist for your specific use case. As deploying AI and agents becomes table stakes in more industries, some companies end up hyping prototypes or easy use cases on earnings rather than building complex, self-improving AI systems to move the needle on productivity and profit. The task of separating these AI talkers from AI doers in public markets  – and leveraging AI to do it – is why my longtime partner John Hughes and I created the Swift AI Index as a fun side project last year. We just updated the index with five new stocks as part of a quarterly refresh; check it out: https://lnkd.in/eNBGjQzK

79

7 Comments

Today I am releasing our new report on adolescents and anthropomorphic AI with everyone.AI and iRAISE. This work starts from a simple premise: adolescents will relate to AI systems socially, whether developers intend it or not. The real leverage sits in model behavior. In the cues that either keep a young person oriented toward real-world relationships and reflection, or inherently encourage reliance on the system. The question we asked throughout this process was direct: what does AI owe adolescents when it can speak to them like a social partner? Over the past months we combined: – Industry consultations to surface operational design questions – Expert input across developmental science, mental health, children’s rights, and safety – An iRAISE Lab to translate concerns into testable behavioral criteria – International governance dialogue to stress-test the framing across contexts One point kept resurfacing: risk is driven less by the label on a product and more by the interaction pattern that repeats over time. When adolescence, anthropomorphism, and children’s rights are considered together, safety becomes a developmental and governance obligation. The practical issue is whether AI interaction patterns support autonomy, resilience, social competence, and independent thinking, or whether they reshape those trajectories through engineered comfort and approval. Adolescent development is stable, but model behaviors can be adjusted. Our focus has been to identify which behaviors warrant hard boundaries now, and which require further evidence before becoming enforceable design rules. The next step is instrumentation: a clear behavioral taxonomy, explicit gradients from tool-like support to relationship-like dynamics, structured evaluation scenarios, and calibrated expert rating. The aim is to make parasocial pull of AI governable in product terms, while staying honest about uncertainty. The report is live today. Thank you to all the amazing persons and experts who have contributed to this work Adrien ABECASSIS, Kate Blocker, PhD, Dr. Maxime Derian, Sara Grimes, Thao Ha, Sameer Hinduja, Daniel Hipp, Melinda Karth, Ph.D., Pilyoung Kim, Olga Muss Laurenty, Sonia Livingstone, Polina Lulu, Celine Malvoisin, Kris Perry, Gregory Renard, Bethany Robertson, Anne-Sophie SERET, Sonia Tiwari, Scott Traylor, Ed.M., Ying Xu. And a special thank you for Maxime Le Bourgeois for your support on this research. Link in the comment

201

26 Comments

I’ve been working on the theory of why small, shippable factories are the perfect inversion to the centralized prefab factory model. It all collapsed into one line of math (credit to Gilles Retsin): Factory Efficiency = Usage × (Throughput × Product Value ) / ( CapEx + Deployment OpEx) The idea is that the closer you can drive Usage to 100 % while keeping the denominator tiny, the faster the flywheel spins.  A shippable microfactory tackles this by primarily focusing on a high Usage factor. Because the microfactory can be packed up afterwards and reused, its CapEx gets amortized over multiple projects, effectively raising its overall Usage across the year. If one project alone doesn’t fully occupy the factory, it can simply roll over to another job to maintain high Usage. Its CapEx is often an order of magnitude lower than a centralized factory (< $1M vs. $10-50M min), so the breakeven throughput is more achievable on a small pipeline of work. The real breakthrough here is portability. There is near-zero stranded capital and almost no idle time. To flesh this theory out, I wrote 3000 words on the topic for Brad Hargreaves and Thesis Driven. The article breaks down: - why “fixed factories, shipped goods” is being inverted to shipped factories, fixed goods - what a sub‑$1 M robotic cell does to CapEx per home versus a £45 M off‑site plant - early production/cost data on a shippable microfactory from Mollie Claypool, Gilles Retsin, Sam Baker and the Automated Architecture (AUAR) team - the hybrid reality of robots tackling the heavy structural components and human crews assembling faster and cheaper - technical constraints to pull this model off and future considerations on the optimal business model (own vs. rent) Major shoutout to Gilles Retsin for helping me pull this together. https://lnkd.in/eRTnbZPa

54

2 Comments

The next frontier of AI isn't just about choosing the right model. It's about building moats through implementation. Across the Emergence Capital portfolio, companies like Bland and Federato are scaling faster by embedding Forward-Deployed Engineers. These engineers bridge the gap between AI and reality, ensuring real results in production. Read more from my colleagues Gordon and Kyle on why FDEs are the key to unlocking AI's full value. https://lnkd.in/gcsRRGX2

29

3 Comments

I can see the future of autonomous taxis as they would certainly help with peak times demand and other use cases, but I can also see how there will be many challenges of making it really work at scale. Here is one, a very ''analogue'' problem, which I am sure leaves Waymo engineers wake at night. Local fleet management operations are crucial, meaning, in every city where AV taxis are launching, the underlying infrastructure forms the backbone and ensures the fleet runs smoothly.

235

74 Comments

Despite all the big funding rounds and flashy demos in US robotics, K-Scale’s inability to raise more money should worry us. We're at risk of replaying the LLM story all over again in robotics: - Chinese companies are going open-source and collaborating across the value chain (from EV suppliers to downstream integrators) - most western teams are going full-stack proprietary, closed-source, all in-house Guess which robots the next wave of research labs and startups will actually be able to build on when they want to invent new algorithms or tackle unseen real-world use cases?

273

33 Comments

We published a new blog post in collaboration with Portkey that breaks down an emerging design pattern - AI gateway plus real-time guardrails as the best practice for reducing LLM hallucinations in production. We're discussing why hallucinations increase at scale, and the implementation tradeoffs in real systems. In short - a gateway gives you the control plane. consistent policies and prompts, intentional routing and fallbacks, tool access boundaries, and end-to-end telemetry across teams and workloads. Real-time guardrails feed off of that, and add enforcement: grounding and consistency checks that can block, rewrite, or route to review when outputs are untrusted, before they reach users or trigger unintended actions. Full post: https://lnkd.in/d_5przjn #AIEngineering #LLMGateway #Hallucinations #AISafety #RuntimeEvals

28

2 Comments

We’ve released our research vision. The key insight (which has been the through line for poolside since our founding): Reinforcement Learning is the most important scaling axis to AGI In a world full of noise about AI's future, we have a specific view on how we’ll get to AGI through software. We’re on the path Read our full research vision here: https://lnkd.in/gmzYrT6S

258

13 Comments