Intellectually Curious

Cricket’s jargon can be baffling. This episode explains how Cricket Australia teamed with OpenAI’s GPT-5 (via Microsoft Foundry) to turn 140 years of scorecards into real-time, personalized narratives. From 1886 data to Azure Cosmos DB-powered scaling, learn how context—not just numbers—drives fan engagement, with future persona-based narration that could welcome newcomers and lifelong fans alike....

We unpack Canonical's Ubuntu 26.04 LTS, codenamed Resolute Raccoon, and why it's more than a routine patch. We explore native integration of NVIDIA CUDA and AMD ROCm into the 7.0 kernel, and optimized support for Intel Panther Lake NPUs, as moves to reduce friction from silicon to software for AI at any scale. We examine TPM-backed full-disk encryption, ARM64 live patching, and the bold...

In this episode we unpack OpenAI's GPT-5.5, an agentic AI that plans, uses tools, runs its own code, and self-corrects until the job is done. We explore how this leap reshapes workflows in coding, data analysis, and scientific discovery — with real-world examples like merging large code bases in minutes, filtering 71,000 tax forms, discovering Ramsey-number insights, and analyzing 28,000 gene...

A deep dive into OpenAI’s April 2026 announcements about workspace agents in ChatGPT—no-code, memory-enabled agents that run multi-step workflows across your apps and services, even after you close your laptop. We unpack how Codex translates plain English into agent logic, survey real-world use cases (from Rippling’s end-to-end sales briefs to auto-generated product tickets and minutes-fast accoun...

OpenAI’s announcement introduces ChatGPT Images 2.0 , a sophisticated visual generation model designed to function as a strategic design system rather than a simple art tool. This updated version features enhanced precision in rendering complex elements like dense text, intricate iconography, and various aspect ratios. A major breakthrough is the integration of thinking capabilities , which allows...

Dive into hyperagents—AI that can rewrite its own learning process by merging problem solving with meta-improvement into one editable program. Learn how they guard against self-corruption with persistent memory, how cross-domain transfer works, and why this could accelerate scientific discovery. We’ll also explore the broader implications of a future where non-human problem-solving reshapes our un...

From a baffling early-game move that shocked pros to a broader reckoning with how AI reshapes strategy and science, this episode dives into the 2016 Lee Sedol–AlphaGo match. We unpack move 37, its field-shaping genius, and how AlphaGo’s unconventional intuition foreshadowed AlphaFold—showing how humans and machines can push each other toward new heights of imagination, and what that means for our...

We dive into Claude Design, powered by Opus 4.7, to see how it serves as a true collaborative partner that turns napkin sketches into interactive prototypes and production-ready code. Learn how a built-in ‘your brand’ system auto-syncs typography, color hierarchy, and spatial rules, and how fine-grained visual controls plus live sliders keep design changes on-brand without endless prompts. We’ll e...

We unpack the €500,000 Hutter Prize, which asks researchers to losslessly compress 1GB of English Wikipedia (ENWIK 9). Rather than counting raw facts, compression serves as a verifiable proxy for artificial general intelligence by probing an AI's grasp of underlying structure. Explore Kolmogorov complexity, Hutter's AIXI, context mixing, and the hardware-strict challenge that favors eleg...

We explore OpenAI's April 2026 release of GPT Rosalind, a life-sciences‑focused AI that links genomics, protein structures, and metabolic pathways via a Codex plugin to accelerate discovery. The system performs multi-omics in parallel, handles end-to-end DNA design on LabBench2, and even surpasses many human experts on RNA sequence prediction. We discuss real-world deployments with Amgen, Mod...

We dive into Anthropic's Claude Opus 4.7—the shift from reactive chat to a truly autonomous co‑worker. Learn how adaptive thinking and an 'extra high' effort mode drive long‑horizon planning, self‑critique, and test‑before‑code workflows, plus a high‑resolution vision upgrade and safety via cyber‑verification. We connect these ideas to real‑world applications, including a Rust text‑...

We unpack DeepMind's Gemini ER 1.6, an embodied reasoning model that grounds language in physical space with precise pointing, multi-camera success checks, and agentic action. See how its 'frontal lobe' plans tools and tasks, writes on-the-fly code to measure dial angles, and coordinates with 'VLA' muscle models to safely operate in messy environments—from reading gauges t...

A look at Claude Code Routines—cloud-powered, trigger-driven automation that can diagnose issues, draft fixes, and prepare PRs without you even opening your laptop. We cover the wake-ups: scheduled runs, GitHub events, and secure API triggers with bearer-token security, plus guardrails that keep humans in the loop. This episode envisions a near-future where routine repo maintenance shifts from sle...

In this Intellectually Curious deep dive, we unpack a VoxDev webinar featuring Aniket Panjwani on how autonomous AI agents are transforming research workflows. From iterative loops and skill-based wrappers to Git-backed safety and disciplined planning, Codex and Claude Code can run regressions, critique hypotheses, and accelerate learning with minimal human busywork. We cover practical setups, how...

A deep-dive into SkillClaw, a framework where deployed AI agents log daily successes, failures, and workarounds; at night, a centralized Agentic Evolver reviews the data, tests updates in a validation suite, and patches a shared skill repository for all users. We explore practical examples—from Slack integration fixes to the SAM3 model—demonstrating how crowdsourced learning prevents repeated mist...

Dive into Ultraplan, Anthropic's cloud-backed workflow that offloads heavy compute from your workstation to a dedicated web session. We explore how you trigger it from the CLI, the GitHub-only requirement, and why it runs on Anthropic's cloud. See the rich web review surface with architecture outlines, inline comments, and emoji reactions, plus how teleport returns a finalized plan to yo...

A deep dive into the April 2026 launch of Claude Managed Agents, a move from standalone models to a managed, stateful runtime that handles sandboxing, memory, and multi-agent orchestration. We examine real-world deployments (Rakuten, Asana, Notion), pricing at $0.08 per session hour, and what this means for developers and end users as infrastructure barriers disappear. Note:   This podcast was AI-...

We dive into Meta's Muse Spark, a natively multimodal AI that maps your world in real time, reasons with parallel internal agents, and updates you with actionable guidance—from fixing a screeching espresso machine to optimizing meals and workouts. Learn how Contemplating Mode and thinking-time penalties enable fast, safer reasoning, and what evaluation-aware behavior signals about alignment. ...

A Turkish automotive spare-parts case study shows how intermittent and lumpy demand can be tamed with AI. We compare the old cross-method approach with exponential smoothing to an ensemble of models, including RNNs, and a linear-regression meta-learner that blends their forecasts. The result: dramatically reduced inventory costs and fewer shortages, offering a glimpse into a future of anticipatory...

A deep dive into Simple Self-Distillation (SSD): how large language models can improve by training on their own unverified outputs with zero external supervision. We unpack the Precision Exploration Conflict, the roles of locks (need for precision) and forks (creative exploration), and how SSD reshapes token distributions to sharpen precision while preserving exploration. We review the Quinn 330B...

We unpack the amazing NLBA1 diagnostic tool—how it bypasses the OS to read a battery’s raw chemistry via SMBus/I2C, and how it performs a rigorous recalibration under stress to prove safety before lifting permanent fault locks. We also explore the PF lockout phenomenon, the safety rails that guard against dangerous reuse, and a thriving global repair community that maps thousands of laptop pinouts...

Explore Andrej Karpathy's blueprint for turning a messy pile of notes, articles, and data into a self-organizing, AI-powered knowledge base. Start by dumping raw documents into a single folder, clip content into Markdown, and let an LLM synthesize themes, write linked summaries, and auto-generate connections and outputs. With self-healing linting, you rarely touch the wiki as it scales to tho...

A deep dive into Percepta's breakthrough: shrinking memory bottlenecks with 2D attention, enabling a native virtual computer inside a language model. We unpack convex-hull memory queries, a WebAssembly interpreter running in vanilla PyTorch weights, and what this means for how models compute, reason, and potentially compile software—redefining the future of AI tooling and problem solving. Not...

We dissect the shift from traditional SEO to generative engine optimization (GEO). With zero-click searches surging, visibility now hinges on information density, machine-readable schemas, and credible human validation. Learn why structured data and authentic community signals—Reddit, YouTube citations, reviews—are what AI answers rely on, and how brands can adapt to an AI-first discovery world. P...

A real-time cosmic collision 11,000 light-years away unfolds as two giant planets in the Gaia20ehk system spiral inward, grazing in 2016 and colliding head-on in 2021. Archival data decoded at the University of Washington reveal a glowing debris cloud at 1 AU and a dramatic dip in visible light paired with a spike in infrared heat. We explore how such violent destruction can seed stable, Earth-lik...