Demos lie. An AI agent can look flawless in testing and then consistently produce confident, plausible, wrong answers the moment it touches real work. This episode of Development digs into the single most common — and most expensive — failure mode teams encounter when deploying AI agents inside their businesses: the context problem. It's not about the model. It's about what the agent actually knows when it has to make a decision.The episode covers what context really means for an AI agent

David and Mike reflect on a period of change and opportunity for mathematics education in Kenya. They discuss the upcoming launch of a new MSc in Mathematical Innovation, lessons from bringing open textbooks into classrooms, and the importance of putting teachers at the centre of educational innovation. Recorded just before the first STACK Conference to be held in Africa, their conversation also explores how bringing together local stakeholders and an international community could contribute to a

Most proposal teams treat the federal government's 30-day minimum response window as a schedule — spending time from the front until it runs out. This episode of Development examines what changes when you reverse that logic entirely, drawing on this worked example of budgeting a 30-day RFP response backwards. The math is simple; the implications for how teams allocate drafting, review, and decision time are anything but.The episode walks through two versions of the same 30-day window — one

We explore how knowledge graphs give AI a structured, bi-temporal memory and how task graphs with a diamond structure curb error amplification in AI swarms. From tamper-proof ledgers to isolated verifiers, this episode outlines a practical blueprint for reliable, scalable AI collaboration. Note: This podcast was AI-generated, and sometimes AI can make mistakes. Please double-check any critical information. Sponsored by Embersilk LLC

750 output tokens per second, same model, no accuracy trade-off. Gemini 3.7 Flash halves the price, through December 31. Fast and cheap, both with fine print. We go through what Ultrafast actually unlocks, the January price cliff buried in the Gemini announcement, why multi-agent failures cannot be predicted from single-agent testing, and the open plugin standard six competitors agreed on. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus

Hosted on Acast. See acast.com/privacy for more information.

Not all programming languages are created equal — and in 2025, the gap between choosing the right one and the wrong one can mean the difference between a system that scales and one that fails. This episode of Development cuts through the noise to examine the languages actually shaping modern software, drawing on the complete 2025 programming language guide to deliver a clear-eyed look at where each language fits, what it costs, and what it's worth.Here's what the episode covers:C and C++ —

Explore rendezvous hashing (highest random weight hashing), the 1996 idea from University of Michigan researchers that lets millions of independent clients decide where to send tasks without communicating first. Learn how hashing a task with every server yields a single winner, how the approach remains stable when servers fail (minimal disruption), and how it compares to consistent hashing. We’ll also see real-world deployments in GitHub, Apache Kafka, and cloud storage, and discuss what this...

Grok 4.6 matches GPT-5.6 Sol at half the price, but crossing 200,000 tokens reprices everything. A Swedish startup doubled to thirteen billion in eight months on cheap model APIs. Same week, opposite signals. We go through the pricing cliff and what it means for agents, where value is actually concentrating in the stack, tens of thousands of workers in India recording daily tasks to train physical robots, and why Hinton, Fei-Fei Li, and Ng landed in the same place on open development despite

Exploring Zuckerberg's Aug 2026 essay 'The Future is for Everyone,' this episode argues that AI will move from a passive tool to proactive partners that plan, execute, and optimize multi-step goals. We unpack how agentic systems could handle tasks—from calendars and shopping to real-time monitoring—while expanding opportunity, enabling new kinds of work, and boosting local communities through open models and infrastructure. A roadmap to a more creative, inclusive future. Note: This pod...

Recording Student Thinking During a Mathematics Discussion with Nicole Garcia ROUNDING UP: SEASON 4 | EPISODE 24 Learning to record students' mathematical thinking might best be described as "on-the-job training" with a great deal of trial and error and a lot of practice.  In this rereleased episode from Season 1, Nicole Garcia from the University of Michigan talks about the practice of recording student thinking and offers insight on this challenging but crucial practice. BIOGRAPHY Nicole Garcia is

Text us your thoughts!As we kick off the back-to-school season, we wanted to spend some time exploring new ways to bring debate and argumentation into your math classroom. Last year, we shared a few discussion and debate routines to help get meaningful conversations started. This year, we’re taking things a step further by highlighting how brilliant educators are building deeper activities around mathematical argumentation and debate. In today’s episode, we’re diving into a routine called Math

Send us Fan Mail How have pre-prints and ArXiv changed the landscape of mathematical publishing? And how is AI changing ArXiv? Plus Magazine editors Rachel Thomas and Marianne Freiberger sat down with Professor Ramin Zabih, one of the leaders of ArXiv, when he came to visit the INI for our AI for Maths and Open Science in April. Watch the talk he gave here: https://www.youtube.com/watch?v=RyT0Z94kvGU --- Living Proof is the podcast of the Isaac Newton Institute for Mathematical Sc...

We explore how Claude, an AI, dramatically advanced the Riemann zeta problem by proving that about 67.25% of its nontrivial zeros lie on the critical line. From a wall of dead ends to a human prompt that sparked 60 coordinated sub-agents, the episode follows the move to a Montgomery–Taylor window, a rank-trace inequality, and a formally verified Lean4 proof. It’s a vivid case study in AI–human collaboration turning grinding insight into rigorous math—and a glimpse of what collaborative discov...

Neuroscientist Stuart Firestein (Columbia University) joins Breaking Math to make an extravagant claim: uncertainty isn't a weakness in science — it's the defining feature that makes progress possible. In this episode, we break down why the "one right answer" myth is one of the most damaging ideas in science, why real experts are often the most uncertain people in the room, and why authority and expertise pull in opposite directions, covering two fundamentally different kinds of probability, why

A system of word equations is called quadratic if no variable occurs more than twice in it. There is an interesting simple algorithm to solve quadratic systems of word equations, which I talk through in this episode. My source is Chapter 12 of "Algebraic Combinatorics on Words" by Lothaire.

Most small and mid-sized businesses carry a SaaS stack that feels both essential and absurd at the same time. Some of those subscriptions are load-bearing walls; others are doing something a weekend script could handle. The hard part isn't knowing that the bloat exists — it's knowing where to cut first, and in what order, without breaking anything that actually matters. This episode of Development walks through a structured audit methodology designed for operators who want answers, not

Football franchises run on data. Jono Luk of Sumer Sports explains how AI is changing who wins — on the field and in the stands. Full Show Notes: Most sports AI coverage focuses on stats. Jono Luk, Chief Product Officer at Sumer Sports, is thinking about something harder: measuring what a player decided, not just what they did. Sumer's computer vision tracks all 22 players up to 60 times per second, generating probabilities for every moment of every play — not to describe what happened, but t...

We unpack how continued pre-training turns a general AI into a coding and math specialist. From Meta's CodeLlama to DeepSeek's findings on code-based learning and Nvidia's synthetic debates, we explore model souping, ultra-long contexts (131k tokens), and why training on code can sharpen logic and mathematical reasoning. We discuss what this means for solving real-world scientific and engineering challenges—and what human-style conversation can unlock next in AI. Note: This podcast was...

Agent routing that cuts costs 74% in testing lands the same day researchers show encrypted reasoning traces can be stolen with two standard API calls. Expanding capability, expanding attack surface. We go through Nvidia's Nemotron and Switchyard, the ELLIS Institute paper and what to do before any provider ships a fix, Microsoft's MAI-Code-1.1-Flash at 73% lower cost, Mojo hitting 1.0 with backward compatibility and memory safety, and three quick ones: Gemini at one billion monthly users, Brad