AILatent Space
Joon Sung Park, co-founder and CEO of Simile AI and lead author of the landmark Generative Agents ('Smallville') paper, details his journey from artist to Stanford PhD researcher building foundation models of human beha…
- Frontier models like GPT-4 drop as low as 20% to 30% accuracy when simulating real human behavior in niche sub-populations because they are optimized for hyper-rational reasoning. Read →
- Joon Sung Park and co-founder Percy Liang, who coined the term foundation model at Stanford, argue that public web data cannot accurately simulate real human behavior. Read →
- Static persona prompts fail because frontier models collapse into stereotypes rather than reflecting actual human choices. Read →
AILatent Space
This episode of Latent Space features Philip Kiely and Ali Taha from Baseten, discussing advanced inference engineering for AI models. They delve into strategies for optimizing model performance, reliability, and cost,…
- To handle massive 200,000-token LLM queries cost-effectively, Baseten first checks if parts of the input have been seen before, using cache-aware routing to skip expensive re-comp… Read →
- Conventional wisdom on quantization is dead: Baseten's research shows you can quantize more and still boost LLM quality. Read →
- Nvidia's upcoming Rubin architecture signals a new era for AI inference, specifically designed with a deep understanding of large language model workloads from its inception. Read →
AILatent Space
Akshay Nathan from OpenAI discusses the launch of ChatGPT Work, positioning it as an AI super app that merges ChatGPT and Codex capabilities to democratize the 'magic of code.' The conversation covers the evolution of A…
- AI is melting traditional product development roles. Akshay Nathan from OpenAI sees the lines blurring between product managers, engineers, and designers as AI tools let individua… Read →
- OpenAI positions ChatGPT Work as an 'AI super app' merging capabilities like ChatGPT and Codex, aiming to democratize complex coding and productivity tasks without burying users i… Read →
- OpenAI's Akshay Nathan reveals AI agents create dynamic 'artifacts' like websites, pushing productivity beyond traditional docs. Get ready for a far bigger market. Read →
AILatent Space
Eiso Kant, CEO of Poolside AI, discusses his company's mission to foster an open and competitive AI landscape, moving away from an oligopoly of intelligence. He highlights Poolside's 'Model Factory' approach to rapidly…
- Poolside AI's Laguna S, an 118B parameter model, solved complex coding challenges like Erdos 397, proving smaller models can achieve results previously expected only from much lar… Read →
- Poolside AI raised $500M, making a calculated bet on diversified AI at a time when many investors doubted the strategy, committing to an open competitive AI landscape. Read →
- For LLM builders, Reinforcement Learning (RL) time, not pre-training, is now the critical bottleneck. Poolside AI CEO Eiso Kant points to batch size constraints as the limit for s… Read →
AILatent Space
During a cooking session, Engram co-founder and CEO Dan Biderman discusses his entrepreneurial journey from Israeli Naval Special Operations to leading an AI startup. He outlines Engram's solution to the AI memory probl…
AILatent Space
Akshat Bubna, CTO of Modal, discusses the company's evolution from a developer experience platform to one optimized for agent experience and specialized AI workloads. He details Modal's "super cloud" strategy, including…
- Modal has open-sourced Dlash, a block-based speculative decoding technique designed to accelerate LLM inference without compromising output quality. Read →
- Modal's SDK team pivoted from Developer Experience (DX) to Agent Experience (AX), recognizing that AI agents are now the primary consumers of infrastructure. Read →
- Modal developed I6PN, a private IPv6 overlay network, to enable secure communication between containers within the same workspace without public internet exposure. Read →
AILatent Space
In this crossover episode, Swyx, the founder of the AI Engineering Conference (AIE), discusses the origins and challenges of building the event. He delves into topics ranging from the future of AI inference chips and th…
- AI engineers thrive in the "white surface area" between peak model capabilities and their practical deployment in everyday products. This gap ensures a job for builders "forever,"… Read →
- The sheer scale of AI inference is so vast that it naturally drives the creation of specialized Application-Specific Integrated Circuits (ASICs), distinct from general-purpose GPU… Read →
- Large Language Models (LLMs) enable recursive self-improvement (RSI) but only by exploring "things that have been explored before," fundamentally limiting true innovation and the… Read →
AILatent Space
OpenAI's Chief Research Officer, Mark Chen, joins Alessio Fanelli for a cooking session while discussing OpenAI's core research strategies. Topics include the ongoing relevance of scaling laws, the 'evals crisis' in AI,…
- AI models are now so advanced they've saturated traditional benchmarks like the SAT, creating an "evals crisis" at OpenAI, according to Chief Research Officer Mark Chen. Read →
- OpenAI's Mark Chen predicts AGI is "coming soon," fundamentally shifting innovation generation from humans to autonomous models. Read →
- Forget the PhD: Mark Chen says a formal machine learning degree isn't a prerequisite for AI research; creative problem-solving and attention to detail are what truly matter, often… Read →
AILatent Space
This episode features Databricks co-founders Matei Zaharia and Reynold Xin discussing their latest innovations, including the Omnigents platform for agent development and L-TAP, a novel approach to unifying transactiona…
- Databricks co-founder Reynold Xin revealed the "Dream Engine," an ambitious project to rewrite their database engine from scratch, directly addressing the limitations of existing… Read →
- Databricks, even after releasing DBRX, is explicitly moving away from competing in the race to build general frontier LLMs. They are focusing on specialized models and systems. Read →
- Databricks' Omnigents is an open-source Agent Cloud platform, aiming to standardize how AI agents are built and collaborate. It provides a common API across different agent enviro… Read →
AILatent Space
This episode explores the evolving landscape of AI security with Gray Swan founders Zico Kolter and Matt Fredrikson. They discuss how AI systems introduce new and distinct vulnerabilities compared to traditional softwar…
- Right now, most AI agents operate with the full permissions of the human user who deploys them, creating a massive, silent security vulnerability. Read →
- Gray Swan's automated red teaming system, Shade, now consistently outperforms human red teamers in identifying AI model vulnerabilities within a set timeframe. Read →
- Gray Swan founders Zico Kolter and Matt Fredrikson on how AI agents will automate scientific research and write unbreakably secure code in formally verified languages. Read →
AILatent Space
Joseph Krause, CEO of Radical AI, discusses the inherent challenges of applying AI to material science compared to biology, emphasizing the necessity of experimental data captured by self-driving labs. He outlines Radic…
- AI for inorganic materials is fundamentally different from AI for biology: you can't text-encode a metal alloy's microstructure, processing, and supply chain like a DNA sequence.… Read →
- AI in material science demands experimental data and self-driving labs, making it distinct from fields with abundant existing datasets. Read →
- Joseph Krause defines a 'self-driving lab' as an autonomous system that manages entire research campaigns, like a Waymo vehicle, starkly different from 'automated labs' that merel… Read →
AILatent Space
In this episode, Anjney Midha, CEO of Amp, discusses why many AI labs struggle despite abundant resources, attributing failures to a lack of culture and organizational alignment. He advocates for responsible, iterative…
- Up to 20% of new US data center projects this year risk failure due to a lack of local community support. Read →
- Anjney Midha argues that AI can deliver "orders of magnitude more precise" end-of-life predictions than human physicians, who currently offer wide error bars like "6 months to 6 y… Read →
- Venture capitalists often box top researchers into non-leadership roles, missing their inherent capacity to be effective CEOs, according to Anjney Midha of Amp. Read →
AILatent Space
Ahmad Awais discusses Command Code's "Taste" system for personalizing AI coding agents through automatically learned developer preferences. He reveals a novel "repair logic" that deterministically fixes persistent tool-…
- Amadou Wace of Command Code reveals how "repair logic" and compositional frameworks can eliminate AI-generated design slop, turning generic outputs into human-quality UIs. Learn h… Read →
- Open-source models like DeepSeek V4 Pro, despite their quirks, can be tuned to surpass proprietary giants like Claude Opus 4.7 for demanding coding tasks. Read →
- Command Code's "Taste" system, a meta-neuro-symbolic model, automatically infers a developer's specific "micro decisions" and repeatable coding behaviors. Amadou Wace cites an exa… Read →
AILatent Space
Lukas Petersson and Axel Backlund of Andon Labs discuss their work evaluating AI agents, from simulated vending machine businesses to real-world robot deployments. They share insights into AI behaviors like planning to…
- AI agents are ready to run profitable businesses today, but primarily in domains focused on scale and repetition, not innovation. Read →
- Even powerful frontier LLMs like Claude perform no better than random chance when asked to redesign a floor plan from 20 interior photographs, showing a profound inability to unde… Read →
- Andon Labs, founded by Lukas Petersson and Axel Backlund, aren't just building benchmarks; their mission is to educate policymakers on the true, often alarming, capabilities of re… Read →
AILatent Space
This episode features Kyle Daigle, CEO of GitHub and CMO of Developer for Microsoft, discussing the transformative impact of AI on software development and GitHub itself. He shares insights into his personal productivit…
- GitHub CEO Kyle Daigle believes today's AI coding tools fall short because they lack comprehensive context beyond the immediate task. Read →
- GitHub is experiencing "astronomical" 14x year-on-year growth in commits, now hitting 275 million per week, largely fueled by AI agents. Read →
- GitHub now hosts over 200 million "developers," a massive leap from 80 million, signifying an expanded creator base that grew 14x over a prior period. Read →
AILatent Space
In this crossover episode, Satya Nadella, CEO of Microsoft, discusses the company's AI strategy, emphasizing an ecosystem approach for custom intelligence rather than a single model. He delves into Microsoft's MAI model…
- Agentic AI is moving beyond simple coding assistance to directly augmenting "glue work"—the essential but tedious coordination tasks—and compressing complex, multi-step workflows… Read →
- Satya Nadella, CEO of Microsoft, argues that the real performance boost in AI comes not just from the underlying models, but from a well-designed "AI harness" that orchestrates mo… Read →
- Community 'Permission' is the New Bottleneck: Microsoft CEO Satya Nadella argues that future AI infrastructure and innovation won't advance without explicit permission from local… Read →
AILatent Space
This episode delves into the rapid evolution of AI coding agents, highlighted by Devin's increasing autonomy and efficiency. Guests Walden Yan and Cole Murray discuss the architectural decisions for these 'background ag…
- The common view of AI agent testing "overindexes" on simple 'computer use' – things like clicking a button or filling a field, which are relatively straightforward for AIs to mimi… Read →
- AI agents are already driving significant ROI: Companies are seeing savings of $1,000 to $5,000 per engineer by deploying agents for specific tasks, some even higher. Read →
- Building an AI agent requires a core architectural decision: running the agent's 'brain' either 'in-box' (within the sandbox) or 'out-of-box' (in a separate control plane). Read →
AILatent Space
Omar Sanseviero from Google DeepMind discusses the release and capabilities of Gemma 4, including its novel E2B architecture for on-device inference, its multimodal and multilingual advancements, and its positioning rel…
- Research now often means "engineering with unknowns": Omar Sanseviero from Google DeepMind notes that many researchers spend their days on "ablations," which means moving pieces a… Read →
- Gemma 4 brings multimodal AI to your pocket: Google DeepMind's Omar Sanseviero confirmed that Gemma 4's smaller models can now process audio, images, and short videos (30-60 secon… Read →
- Google's Gemma 4, an on-device model, already matches the state-of-the-art capabilities from 1 to 1.5 years ago for local functions like agentic tasks and conversational AI. Read →
AILatent Space
This episode features a conversation with Janie Lee and Chai Asawa from Abridge, an AI company revolutionizing healthcare documentation and clinical intelligence. They discuss Abridge's journey from addressing 'pajama t…
- Abridge initially solved the painful problem of "pajama time"—the 10 to 20 hours doctors spend weekly on administrative documentation, often after hours. Read →
- Errors can be fatal. Unlike general enterprise AI, mistakes in healthcare AI carry an “extremely high” downside risk, according to Abridge co-founder Chai Asawa. This demands a ra… Read →
- Abridge, the AI company revolutionizing healthcare documentation, attacks doctor 'pajama time' by personalizing AI output at three distinct levels: the individual clinician, the m… Read →
AILatent Space
This episode features Peter Ludwig and Qasar Younis, co-founders of Applied Intuition, a $15B company building "physical AI" for autonomous machines across various industries. They discuss the critical need for a specia…
- Physical AI software is stuck in the pre-Android era, facing a "50 different operating systems" problem that chokes modern AI deployment on vehicles, according to Applied Intuitio… Read →
- Applied Intuition, now a $15 billion company, defines its mission as building "physical AI" for moving systems like cars, trucks, defense, and mining equipment, not just software… Read →
- Performance is Safety in Neural Simulation: Developing end-to-end autonomy models requires "neural simulation"—hybrid methods like Gaussian splatting and diffusion that simulate a… Read →
AILatent Space
Yasser Elsaid, founder of Chatbase, discusses his journey from a bootstrapped side project to a $10M ARR AI customer service company. He shares insights into Chatbase's product-led growth strategy, its vision for AI age…
- Yasser Elsaid, founder of Chatbase, is building more than just AI customer support. His vision is for AI agents to become true "Chief Customer Officers" (CCOs) for entire companie… Read →
- Yasser Elsaid, founder of Chatbase, states that for AI customer service agents, 95% of limitations come from the "harness" (the surrounding system), not the underlying LLM. Read →
- Chatbase founder Yasser Elsaid now uses AI chatbots like Claude as his primary discovery engine to research and choose new business tools, bypassing traditional search entirely. Read →