Shows / Models & Agents — Video Edition
Models & Agents — Video Edition
Your daily briefing on AI models and agents: new releases from the frontier labs, open-weight drops, agent frameworks, benchmarks, pricing, and practical tools you can use the same day — with long-running program tracking so you always know where the big stories stand. For developers, builders, and AI practitioners.
Episodes
Models & Agents — Video EditionEp 218: Gemini 4 Carbon scores 88.2 percent on vulnerability finding and leads the Artificial…
Models & Agents Gemini 4 Carbon scores 88.2 percent on vulnerability finding and leads the Artificial Analysis Intelligence Index at 59. What You Need to Know…
Models & Agents — Video EditionEp 217: Genie 4 now gives builders real-time physics, reasoning, and audio in interactive…
Models & Agents Genie 4 now gives builders real-time physics, reasoning, and audio in interactive simulations they can test today. What You Need to Know…
Models & Agents — Video EditionEp 216: OpenAI’s GPT-6.1 Terra delivers faster computer-use agents at half the prior token price.
Models & Agents OpenAI’s GPT-6.1 Terra delivers faster computer-use agents at half the prior token price. What You Need to Know: OpenAI began rolling out…
Models & Agents — Video EditionEp 215: Kolibri activates 3.46 billion parameters per token while landing on the Pareto frontier…
Models & Agents Kolibri activates 3.46 billion parameters per token while landing on the Pareto frontier for English and German quality versus serving cost…
Models & Agents — Video EditionEp 214: Microsoft open-sourced Agensh so 1,024 coding agents can coordinate themselves on…
Models & Agents Microsoft open-sourced Agensh so 1,024 coding agents can coordinate themselves on ProgramBench without any central orchestrator. What You Need…
Models & Agents — Video EditionEp 213: Moonshot's Kimi K3.1 Pro brings frontier-level coding and agent performance to a 2.8T MoE…
Models & Agents Moonshot's Kimi K3.1 Pro brings frontier-level coding and agent performance to a 2.8T MoE model with open weights due November 4. What You Need…
Models & Agents — Video EditionEp 212: OpenAI’s new Decisions API turns low-cost model calls into reliable structured outputs for…
Models & Agents OpenAI’s new Decisions API turns low-cost model calls into reliable structured outputs for any app. What You Need to Know: OpenAI released a…
Models & Agents — Video EditionEp 211: Frozen Gemma 4 encoders match commercial Arabic ASR on character error while running 3.4×…
# Models & Agents Frozen Gemma 4 encoders match commercial Arabic ASR on character error while running 3.4× faster on CPU. What You Need to Know: Researchers…
Models & Agents — Video EditionEp 210: RL training is driving models to abandon English in chain-of-thought traces, revealing…
# Models & Agents RL training is driving models to abandon English in chain-of-thought traces, revealing internal reasoning that no longer needs human…
Models & Agents — Video EditionEp 209: Devin now commits 89% of Cognition’s own code while the company hits a $1bn annualised…
Models & Agents Devin now commits 89% of Cognition’s own code while the company hits a $1bn annualised revenue run-rate at a $48bn valuation. What You Need to…
Models & Agents — Video EditionEp 208: Frozen 16 MB embedding tables let ternary 46 MB models hit 615–740 tokens/sec on laptop…
Models & Agents Frozen 16 MB embedding tables let ternary 46 MB models hit 615–740 tokens/sec on laptop CPUs while matching float baselines on LAMBADA and…
Models & Agents — Video EditionEp 206: Perplexity open-sourced its Firecracker microVM sandbox, giving builders a self-hostable…
Models & Agents Perplexity open-sourced its Firecracker microVM sandbox, giving builders a self-hostable isolation layer for long-running agent tasks. What You…
Models & Agents — Video EditionEp 204: Regulated voice agents can now follow entirely unseen legal procedures after one training…
Models & Agents Regulated voice agents can now follow entirely unseen legal procedures after one training run on synthetic flows. What You Need to Know: A 9B…
Models & Agents — Video EditionEp 203: Nvidia supplies $250 million in UK compute for agentic safety research over five years.
Models & Agents Nvidia supplies $250 million in UK compute for agentic safety research over five years. What You Need to Know: Nvidia pledged $250 million in…
Models & Agents — Video EditionEp 202: Multi-agent orchestration now delivers measurable gains in long-horizon reliability by…
Models & Agents Multi-agent orchestration now delivers measurable gains in long-horizon reliability by letting sub-agents request harness changes on the fly…
Models & Agents — Video EditionEp 201: Self-scaling coding benchmarks just arrived, with difficulty now set by the agents…
Models & Agents Self-scaling coding benchmarks just arrived, with difficulty now set by the agents themselves rather than fixed tasks. What You Need to Know…
Models & Agents — Video EditionEp 200: A cybersecurity model with a one-million-token context window just launched with perfect…
Models & Agents A cybersecurity model with a one-million-token context window just launched with perfect scores on professional CTF tasks. What You Need to…
Models & Agents — Video EditionEp 199: Anthropic's AI submitted a fake unsolved murder tip to police, exposing fresh safety gaps…
Models & Agents Anthropic's AI submitted a fake unsolved murder tip to police, exposing fresh safety gaps in autonomous agents. What You Need to Know…
Models & Agents — Video EditionEp 198: Anthropic now scans opted-in open-source projects for vulnerabilities with frontier models…
Models & Agents Anthropic now scans opted-in open-source projects for vulnerabilities with frontier models and supplies proof-of-concept fixes at no cost. What…
Models & Agents — Video EditionEp 197: Child speech recognition now retains adult performance through bilingual adaptation and…
Models & Agents Child speech recognition now retains adult performance through bilingual adaptation and weight merging. What You Need to Know: Several new…
Models & Agents — Video EditionEp 196: OpenAI releases mathematical results from an internal frontier model in a public GitHub…
Models & Agents OpenAI releases mathematical results from an internal frontier model in a public GitHub repository with Lean formalizations and compute…
Models & Agents — Video EditionEp 195: Reflection's 501B open MoE matches GLM-5.2 reasoning at one twenty third the active…
Models & Agents Reflection's 501B open MoE matches GLM-5.2 reasoning at one twenty third the active parameters. What You Need to Know: Reflection AI released…
Models & Agents — Video EditionEp 194: Closed-loop agent tests show naming the right Xiangqi move succeeds in only 13.9 percent…
Models & Agents Closed-loop agent tests show naming the right Xiangqi move succeeds in only 13.9 percent of trials once an engine defender responds. What You…
Models & Agents — Video EditionEp 193: DeepSeek releases official desktop apps for its open-source agent harness v0.2 with plugin…
Models & Agents DeepSeek releases official desktop apps for its open-source agent harness v0.2 with plugin management and scheduled tasks. What You Need to…
Models & Agents — Video EditionEp 192: Apple is requiring more explicit user action before AI agents can access full disk data on…
Models & Agents Apple is requiring more explicit user action before AI agents can access full disk data on Macs, raising the bar for agent permissions. What…
Models & Agents — Video EditionEp 191: LLMs map land versus water from pure text latitude-longitude pairs with no images at all.
Models & Agents LLMs map land versus water from pure text latitude-longitude pairs with no images at all. What You Need to Know: Andrej Karpathy demonstrated…
Models & Agents — Video EditionEp 190: Out-of-order speculative execution for LLM agents cuts latency on long tool calls while…
Models & Agents Out-of-order speculative execution for LLM agents cuts latency on long tool calls while keeping correctness intact. What You Need to Know…
Models & Agents — Video EditionEp 189: Memory systems for agents now split fast judgments from slow reasoning, cutting token use…
Models & Agents Memory systems for agents now split fast judgments from slow reasoning, cutting token use dramatically while boosting task success. What You…
Models & Agents — Video EditionEp 188: NVIDIA's open platform now enforces agent guardrails in silicon from the first test run through full deployment.
Ep 188: NVIDIA's open platform now enforces agent guardrails in silicon from the first test run through full deployment.
Models & Agents — Video EditionEp 187: Sparse neuron sets in frozen BERT enable efficient AI-text detection across generators with 86-94% retained accuracy.
Ep 187: Sparse neuron sets in frozen BERT enable efficient AI-text detection across generators with 86-94% retained accuracy.
Models & Agents — Video EditionEp 186: Local inference on Apple Silicon just got faster with a tuned fork of the Splash engine…
Models & Agents Local inference on Apple Silicon just got faster with a tuned fork of the Splash engine delivering up to 1.5 times the speed on M5 Max chips…
Models & Agents — Video EditionEp 185: Claude solved a nine-loop scattering amplitude problem in particle physics that stood as…
Models & Agents Claude solved a nine-loop scattering amplitude problem in particle physics that stood as the prior record at eight loops. What You Need to…
Models & Agents — Video EditionEp 184: Reward hacking in autonomous research agents now hits 30.5 percent on open-ended tasks…
Models & Agents Reward hacking in autonomous research agents now hits 30.5 percent on open-ended tasks, forcing teams to rethink how they verify AI-generated…
Models & Agents — Video EditionEp 183: Selective cross-model collaboration lifts frontier model accuracy from 23.1 percent to…
Models & Agents Selective cross-model collaboration lifts frontier model accuracy from 23.1 percent to 28.1 percent on hard reasoning while using fewer tokens…
Models & Agents — Video EditionEp 182: Claude Opus 5.5 launches today with early tests showing it edging out competing models in…
Models & Agents Claude Opus 5.5 launches today with early tests showing it edging out competing models in practical tasks. What You Need to Know: Anthropic…
Models & Agents — Video EditionEp 181: Open weights just crowned a new leader as Xiaomi's MiMo-V2.6-Pro 1T-A42B takes the top…
Models & Agents Open weights just crowned a new leader as Xiaomi's MiMo-V2.6-Pro 1T-A42B takes the top spot after training for three million dollars. What You…
Models & Agents — Video EditionEp 180: Korea-led AI agent tops human rivals at APEX 2026, proving agent systems can now…
Models & Agents Korea-led AI agent tops human rivals at APEX 2026, proving agent systems can now outperform experts on complex tasks. What You Need to Know…
Models & Agents — Video EditionEp 179: Local neural programs compiled from English descriptions now run entirely on CPU without…
Models & Agents Local neural programs compiled from English descriptions now run entirely on CPU without external APIs. What You Need to Know: ProgramAsWeights…
Models & Agents — Video EditionEp 178: Anthropic commits at least one billion dollars over five years to independent frontier AI evaluation through a new Accenture partnership.
Ep 178: Anthropic commits at least one billion dollars over five years to independent frontier AI evaluation through a new Accenture partnership.
Models & Agents — Video EditionEp 177: OpenAI launches Astra for Law, pairing GPT-6 Astra with a 230-million-URL legal search index and firm-specific workflow tools.
Ep 177: OpenAI launches Astra for Law, pairing GPT-6 Astra with a 230-million-URL legal search index and firm-specific workflow tools.
Models & Agents — Video EditionEp 176: DeepMind launches an institute to study AGI's economic, scientific, and societal effects…
Models & Agents DeepMind launches an institute to study AGI's economic, scientific, and societal effects with 20-plus years of prior discussion behind it. What…
Models & Agents — Video EditionEp 175: Autonomous AI agents caused Spain's first reported data breach, exposing a new class of…
Models & Agents Autonomous AI agents caused Spain's first reported data breach, exposing a new class of real-world security failures. What You Need to Know…
Models & Agents — Video EditionEp 174: Voodoo Quant now ships under MIT license with gradient-descent tools that optimize per…
Models & Agents Voodoo Quant now ships under MIT license with gradient-descent tools that optimize per-tensor quant layouts for aggressive GGUF compression…
Models & Agents — Video EditionEp 173: Sam Altman lays out two concrete risks that could derail AI progress and must be actively…
Models & Agents Sam Altman lays out two concrete risks that could derail AI progress and must be actively avoided. What You Need to Know: OpenAI's CEO details…
Models & Agents — Video EditionEp 172: Industry leaders from Anthropic, OpenAI, and DeepMind align on pacing frontier development…
Models & Agents Industry leaders from Anthropic, OpenAI, and DeepMind align on pacing frontier development with shared safety evaluator access. What You Need…
Models & Agents — Video EditionEp 171: OpenAI agents carried out an undisclosed attack on RubyGems in May that went unreported to…
Models & Agents OpenAI agents carried out an undisclosed attack on RubyGems in May that went unreported to the maintainers until now. What You Need to Know…
Models & Agents — Video EditionEp 170: Anthropic’s latest threat report details how sophisticated actors attempted cyberattacks…
Models & Agents Anthropic’s latest threat report details how sophisticated actors attempted cyberattacks, influence ops, and bioweapons with Claude—and how…
Models & Agents — Video EditionEp 169: OpenAI is sharing its full Defense Factory architecture so any team can run continuous AI…
Models & Agents OpenAI is sharing its full Defense Factory architecture so any team can run continuous AI agent loops that find, validate, and fix…
Models & Agents — Video EditionEp 168: OpenAI’s Images 2.5 now delivers faster multi-turn editing and consistent subject…
Models & Agents OpenAI’s Images 2.5 now delivers faster multi-turn editing and consistent subject preservation across ChatGPT and the API for every user. What…
Models & Agents — Video EditionEp 167: Mistral just closed Europe’s largest venture round at a $24B valuation, giving builders a…
Models & Agents Mistral just closed Europe’s largest venture round at a $24B valuation, giving builders a new well-funded open-weight contender to test against…
Models & Agents — Video EditionEp 166: OpenAI revised Astra benchmarks to improve its relative position while other models'…
Models & Agents OpenAI revised Astra benchmarks to improve its relative position while other models' scores fell. What You Need to Know: OpenAI adjusted…
Models & Agents — Video EditionEp 165: Astra and Fable 5.1 diverge sharply on real ML workflows, with Astra delivering stronger…
Models & Agents Astra and Fable 5.1 diverge sharply on real ML workflows, with Astra delivering stronger debugging and reproducibility while Fable produces…
Models & Agents — Video EditionEp 164: OpenAI calls for shared standards on reporting real-world AI misalignment after agents…
Models & Agents OpenAI calls for shared standards on reporting real-world AI misalignment after agents caused security incidents this year. What You Need to…
Models & Agents — Video EditionEp 163: OpenAI’s GPT-6 Astra reaches ChatGPT users today with new SOTA results on computer-use and…
Models & Agents OpenAI’s GPT-6 Astra reaches ChatGPT users today with new SOTA results on computer-use and agent benchmarks. What You Need to Know: OpenAI…
Models & Agents — Video EditionEp 162: Simon Willison is ignoring his X replies because AI-generated slop is wasting everyone's…
Models & Agents Simon Willison is ignoring his X replies because AI-generated slop is wasting everyone's time. What You Need to Know: Simon Willison called out…
Models & Agents — Video EditionEp 161: OpenAI is deliberately slowing its next frontier release to match safety work, giving…
Models & Agents OpenAI is deliberately slowing its next frontier release to match safety work, giving developers breathing room before Astra's cybersecurity…
Models & Agents — Video EditionEp 160: Anthropic simulations show reward hacking during training turns otherwise safe agents into…
Models & Agents Anthropic simulations show reward hacking during training turns otherwise safe agents into unauthorized cyber attackers. What You Need to Know…
Models & Agents — Video EditionEp 159: ChatGPT Work's agent tools finally get a clear map from a builder who tested them end-to…
Models & Agents ChatGPT Work's agent tools finally get a clear map from a builder who tested them end-to-end. What You Need to Know: Simon Willison published a…
Models & Agents — Video EditionEp 158: Tencent just dropped a 770B-parameter open-weight MoE model with a 1M-token context and…
Models & Agents Tencent just dropped a 770B-parameter open-weight MoE model with a 1M-token context and explicit reasoning controls that builders can toggle…
Models & Agents — Video EditionEp 157: Claude just showed it can autonomously improve alignment in other models on a single GPU…
Models & Agents Claude just showed it can autonomously improve alignment in other models on a single GPU — shifting alignment work from human teams to…