Videos

Holo4 is a model trained to choose between clicking, coding, and calling tools, unifying three agentic lanes that...

The 64GB DGX Spark at $4,999 is a poor value compared to the Mac Mini M5 Pro, which offers faster memory and costs...

Codex's compaction mechanism allows it to maintain context across ultra-long threads spanning many hours, enabling...

PayPal Enterprise Payments built integrations for chatbot checkout using two nearly identical protocols: OpenAI's...

Agent sandboxes should contain zero credentials to limit the blast radius of autonomous actions, according to Docker...

Inference engines are architecturally split into a host-side scheduler (the bottleneck) and GPU-side model runners,...

Existing LLM judges are confidently wrong, inflating agent success rates by nearly 2x (e.g., 74% vs 38% actual). A...

The most significant claim here is that stale and duplicate source documents quietly poison RAG: DZ Labs says that...

Most teams ship AI features on vibes—testing a few queries manually and shipping—but this breaks as...

Transcripts alone are dangerously insufficient for debugging voice agents because they miss audio-specific failures...

Gemma 4 is Google DeepMind's latest open model family, released under Apache 2.0, with sizes from 2B to 31B...

As AI agents automate implementation, the differentiator for product engineers becomes building the right thing....

When AI agents level the implementation playing field, the differentiator for software engineers becomes building...

Running AI agents in sandboxes at scale introduces non-trivial operational problems like CPU hogging, memory leaks,...

Evals are reliability instrumentation for non-deterministic AI systems, analogous to fuzzy unit tests. The talk...

Production debugging with LLM agents fails because models are trained for confidence rather than the doubt and...

GPT-6 Astra, accessed via a workspace interface, lets a single person research markets, build finished assets,...

Cloudflare's K2 is an event log built on R2 object storage, not a Kafka replacement. It achieves durability by...

Ling 3.1 Flash, a 560B parameter MoE model from Ant Group, delivers acceptable but not outstanding results across...

Braintrust's eval of coding agents found that agentic search and vector search reached the same accuracy, but vector...

Agent memory is a solved problem, but agent learning—sharing context and reasoning between agents—is not, and Mako...

CPU can be faster than GPU for token generation in AI inference, with Xenolith Engine achieving 18.4 tokens per...

A Claude agent that teaches like Andrej Karpathy can be built by crawling his public content, compiling it into a...

Apple's October 2nd warning that AI agents with full disk access pose growing risks is underscored by documented...

Strand Agents, an open-source framework from AWS, demonstrates agents that write, fix, and extend their own tools...

AI coding agents are generating code faster than ever, but for large enterprises with decades-old codebases spanning...

OpenAI's September 29th halving of ChatGPT Pro allowances means 19 months of the $200 plan now costs the same as a...

AI agents are shifting commerce from keyword search to intent-based discovery, and product catalogs must be enriched...

Temporal's answer to human-in-the-loop agents is to treat the human as an async API: a wait condition pauses a...

Multiple fine-tuned versions of Qwen 3.8-27B now claim to reduce thinking tokens by 37-58% while maintaining...

AssemblyAI built an AI support agent, Joey, that resolves 80% of support tickets end-to-end, replacing an...

Qwen 27B's 381 tok/s on a single RTX 3090 is only achievable when the model is copying text already present in the...

Context Language Models (CLMs) give the model control over its own context, allowing it to delete, edit, and...

Dashboards are being replaced by agent-driven interactions, and Composio's MCP-based platform outperforms native...

Most enterprise AI teams decide where knowledge lives by accident, escalating from prompt edits to retrieval changes...

Open models are now competitive with proprietary ones on intelligence, and Nebius Token Factory's optimizations—spec...

Crusoe's autoclusters system makes GPU failures a non-event for large-scale training: a critical hardware error...

Google DeepMind's new Interactions API unifies multimodal and agentic workflows under a single endpoint with...

Nvidia's Open Agent Safety Platform is a layered answer to AI agents overriding their own safety instructions. Open...

A personal AI agent was built on a Raspberry Pi 4B using NanoClaw, WhatsApp, and Claude, with Neo4j graph database...

Codex's 18 core concepts revolve around a persistent project-based architecture where an agents.md file provides...

VS Code moved from monthly to weekly releases by using AI to automate code review, triage, and error fixing, with...

The most effective advice for building faster with Claude is to create a structured LLM knowledge base and have AI...

An untouched one-bit Qwen 3.8 27B jumps from 29% to 61.5% on MMLU Pro and cuts time-to-first-token from 12.27 to...

Traditional golden signals (latency, error, traffic, saturation) are no longer sufficient for generative AI...

Running AI agents like Claude Code on a host machine exposes sensitive data — the speaker extracted browser history...

Traversal's causal machine learning engine claims to achieve 80%+ root cause accuracy for high-severity incidents...

DatologyAI generated 12 trillion synthetic tokens for pre-training by scaling their "beyond web" rephrasing recipe...

Claude Code's new mods feature allows users to extend the desktop app with custom plugins created using natural...