Mixture of Experts

Afleveringen

1. AI exposes a major web security flaw, plus a new kind of voice model

   30 oktober 2026 · 39 min

   Afspelen
   Visit Mixture of Experts podcast page to get more AI content → https://www.ibm.com/think/podcasts/mixture-of-experts Lees meer

   This week on Mixture of Experts, host David Zax is joined by Ash Minhas, Judith Aquino and Jessica Nahulan to discuss how AI is testing the systems we rely on, from cybersecurity to model evaluation.

   First, the panel examines PHP2Shell and what happens when AI can find vulnerabilities faster than organizations can fix them. Can our aging software infrastructure keep up?

   Next, Block's Dessa introduces a voice model trained entirely on speech that can listen and talk simultaneously. The panel discusses what makes this approach different and how to evaluate a model designed to converse more like a human.

   Then, a new cultural riddles benchmark reveals gaps in how language models interpret cultural context, prompting a broader debate about whether today's benchmarks reward guessing over admitting uncertainty.

   Plus, IBM Field CTO Kiyu Gabriel joins from IBM TechXchange 2026 to discuss Streamhouse and how AI agents are changing the demands on enterprise data infrastructure.

   00:00 – Intro

   01:18 – AI discovers major PHP2Shell flaws

   14:30 – Dessa: An AI model trained on speech

   22:38 – Cultural riddles reveal model blind spots

   30:22 – IBM TechXchange: Data for the agent era

   All that and more on Mixture of Experts.

   "The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity. AI tools may be used to transcribe this episode and support selected stages of the production process. All AI-assisted content is reviewed by the production team before publication."

2. Self-replicating AI, your doctor’s chatbot & agents at dinner

   23 oktober 2026 · 36 min

   Afspelen
   Visit Mixture of Experts podcast page to get more AI content → https://www.ibm.com/think/podcasts/mixture-of-experts Lees meer

   What happens when an AI model can copy itself onto the devices in yout hoome?

   This week on Mixture of Experts, host David Zax is joined by Aili McConnon, Sandhya Iyer and Jacob Repp. First, Anthropic research suggests small AI models may be months away from copying themselves across cheap smart devices without a data center. The panel debates whether the bigger risk is the model itself or the poorly secured devices already in our homes. Next, a JAMA study finds that about half of US doctors use OpenEvidence, a clinical chatbot, more often for treatment decisions than for diagnosis. The panel discusses whether faster, less-vetted answers are a trade worth making. Finally, 22 founders let AI agents book their seats at a Manhattan dinner, and the agents networked with each other in a chat room. The panel considers how much autonomy agents should get when they deal with other agents.

   00:00 – Intro

   01:17 – AI models approach self-replication

   14:22 – Study reveals how doctors are using AI

   26:38 – Agents network at a founders’ dinner

   All that and more on Mixture of Experts.

3. Decision models take off & Anthropic bans cruelty to Claude

   16 oktober 2026 · 32 min

   Afspelen
   Visit Mixture of Experts podcast page to get more AI content → https://www.ibm.com/think/podcasts/mixture-of-experts Lees meer

   This week on Mixture of Experts, host David Zax is joined by Kaoutar El Maghraoui, Nick Gallagher, and Volkmar Uhlig. First, Microsoft, OpenAI, Perplexity, and Cloudflare follow TypeSafe’s Jev in building decision models, a new category of AI built to make fast, scored judgments. The panel debates whether these models are truly new or a more efficient application of existing technology, and whether better-calibrated confidence scores can help address the overconfidence problem in LLMs. Next, a new study finds that teams of AI agents can cost two to five times more than a single agent without consistently delivering better results. The panel discusses when multiple agents are worth the added cost and complexity. Finally, Anthropic updates its usage policy to prohibit “sustained and needless abusive or cruel behavior” toward its models, raising a bigger question: Is this about protecting AI, shaping human behavior, or something else entirely? The panel considers the ethical implications and whether the policy could also be a marketing play.

   00:00 – Intro

   01:17 – Inside the decision model boom

   10:54 – When are agent teams worth the cost?

   23:39 – Anthropic's new usage policy

   All that and more on Mixture of Experts.

   "The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity. AI tools may be used to transcribe this episode and support selected stages of the production process. All AI-assisted content is reviewed by the production team before publication."

4. OpenAI's always-on agents, 700+ math manuscripts & HackerRank's AI interviewer

   9 oktober 2026 · 35 min

   Afspelen
   Visit Mixture of Experts podcast page to get more AI content → https://www.ibm.com/think/podcasts/mixture-of-experts Lees meer

   Can AI take mathematics beyond human understanding?

   This week on Mixture of Experts, host David Zax is joined by Kush Varshney, Skyler Speakman, and Utsav Dholakia. First, OpenAI introduces Dots at DevDay, and Utsav shares his firsthand impressions as the panel discusses whether persistent AI agents can truly work independently. Next, OpenAI releases more than 700 AI-generated math manuscripts, raising questions about the future of mathematics and whether humans can keep up with AI-generated math. Then, HackerRank’s Chakra brings AI to job interviews, prompting a debate about bias, access to opportunity, and how we evaluate candidates. Finally, Reflection AI unveils Beam, giving the panel an opportunity to explain mixture of experts models and the difference between open-weight and open-source AI.

   00:00 – Intro

   00:58 – Inside OpenAI DevDay and Dots

   9:08 – AI generates 700+ math manuscripts

   16:03 – HackerRank’s AI interviewer

   24:18 – Mixture of experts models, explained

   All that and more on Mixture of Experts.

   "The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity. AI tools may be used to transcribe this episode and support selected stages of the production process. All AI-assisted content is reviewed by the production team before publication."

5. OpenAI cancels Astra release, Sonnet 5.5 & what Meta Muse means for work

   2 oktober 2026 · 32 min

   Afspelen
   Visit Mixture of Experts podcast page to get more AI content → https://www.ibm.com/think/podcasts/mixture-of-experts Lees meer

   What makes an AI company pull the plug on a model release? This week on Mixture of Experts, our newly minted full-time host David Zax is joined by Chris Hay, Madison Gooch, and Ash Minhas. First, OpenAI cancels the release of GPT-6.1 Astra, as the panel debates whether self-restraint is becoming a competitive advantage in AI or just good marketing. Then, Anthropic releases Sonnet 5.5, prompting a discussion about what happens when mid-tier models approach top-tier performance on some tasks. Finally, Meta Muse is making powerful AI agents more accessible to everyday users. We look at how these frictionless user experiences could shape what people expect from AI at work.

   00:00 – Intro

   01:18 – OpenAI cancels GPT-6.1 Astra release

   11:44 – Anthropic releases Sonnet 5.5

   23:07 – What Meta Muse means for enterprise AI

   All that and more on Mixture of Experts.

   "The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity. AI tools may be used to transcribe this episode and support selected stages of the production process. All AI-assisted content is reviewed by the production team before publication."

6. New frontier AI models, TypeSafe’s Jev AI, & NASA’s IBM collab

   25 september 2026 · 39 min

   Afspelen
   Visit Mixture of Experts podcast page to get more AI content → https://www.ibm.com/think/podcasts/mixture-of-experts Lees meer

   It's been a wild week in AI, and on episode 126 of Mixture of Experts, host Tim Hwang and co-host David Zax talk to panelists Kaoutar El Maghraoui, Gabe Goodhart and Martin Keen to discuss what all these releases have in common: efficiency.

   We start with the model release pile-up. Anthropic shipped Claude Opus 5.5, and OpenAI launched GPT-6 Sol and Luna. Opus 5.5 is 20% cheaper than earlier Opus models, while GPT-6 Luna costs half what its predecessor did, at just 10 cents per million input tokens. We look at what a real price war means for developers building on these models and the companies racing to lower costs.

   Next, we look at a different idea altogether. TypeSafe AI has introduced "System One" models, starting with Jev, which give up the florid text generation in exchange for fast, structured decisions with calibrated probabilities. The pitch is frontier-level judgment on decision tasks, and responses in hundreds of milliseconds. We weigh the bold speed and cost claims against the company's own caveats about how its benchmarks were built.

   Finally, we leave the chat window entirely. NASA and IBM have released an open-source Lunar Foundation Model, free on Hugging Face and GitHub, and trained on roughly 2 million image tiles from the Lunar Reconnaissance Orbiter and other missions. Researchers can fine-tune it to map craters, spot young volcanic features and estimate where polar ice may be stable, proving models can do more than vibe code. All that and more on this week’s Mixture of Experts.

   00:00 – Intro

   1:05 - Anthropic and OpenAI update AI models

   11:10 - TypeSafe AI unveils Jev AI model

   29:29 - NASA and IBM build lunar AI model

   "The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity. AI tools may be used to transcribe this episode…

7. Pacing the AI frontier, IBM Granite 4.2 & Meta’s Muse assistant

   18 september 2026 · 39 min

   Afspelen
   Visit Mixture of Experts podcast page to get more AI content → https://www.ibm.com/think/podcasts/mixture-of-experts Lees meer

   On episode 125 of Mixture of Experts, host Tim Hwang and co-host David Zax are joined by Mihai Criveti and Abraham Daniels to discuss the back and forth of frontier AI development, IBM’s latest Granite news, and what’s next for personal agents.

   We open with the biggest story in tech: Anthropic CEO Dario Amodei's call to pump the brakes on frontier AI development. In a widely discussed essay, Amodei argued the industry must slow the pace at which it improves AI model capabilities, warning that progress will still feel fast even so. But will independent evaluators and more enforcement really slow down the fastest moving companies?

   Then Abraham Daniels discusses the latest changes coming to IBM’s Granite 4.2 and its open enterprise-focused models in 3B, 8B, and 30B sizes built for reasoning, tool use, coding, and agentic workflows.

   Finally, we discuss Meta’s push into personal agents with Muse: a personal AI agent that runs on its own secure virtual machine, works across a person's daily apps, can make purchases on your behalf, and learns from conversations to get smarter the more you use it.

   Slowdown rhetoric, updates to Granite, and agents that shop for you. All that and more on this week’s Mixture of Experts.

   00:00 – Intro

   1:13 - Anthropic’s AI development dilemma

   15:29 - IBM releases Granite 4.2

   25:44 - Meta launches Muse agentic AI assistant

   "The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity. AI tools may be used to transcribe this episode and support selected stages of the production process. All AI-assisted content is reviewed by the production team before publication."

8. OpenAI talks GPT-6 Astra and Millenium Prize, researchers create WeWorm exploit & IBM’s US Open app

   11 september 2026 · 33 min

   Afspelen
   Visit Mixture of Experts podcast page to get more AI content → https://www.ibm.com/think/podcasts/mixture-of-experts Lees meer

   It's been a week that proves AI is moving faster than anyone can fully keep up with, unless you’ve got a good backhand. On episode 124 of Mixture of Experts, host Tim Hwang and co-host David Zax are joined by Bri Kopecki, Aaron Baughman, and Olivia Buzek to talk about new advances from OpenAI, new cybersecurity threats, and new tools for US Open fans from IBM.

   First, OpenAI introduced GPT-6 Astra as the company's most intelligent and aligned model yet, reaching state-of-the-art results across computer use, software engineering, cybersecurity, and science. Not missing a beat, the company dropped a second bit of news: an internal model produced a proof related to the 3-D Navier-Stokes equation, resolving one of mathematics' most famous unsolved problems. We dig into both the breakthrough and the conversation around the use of AI to solve these seemingly impossible problems.

   Then, IBM and the USTA unveiled new AI-powered fan features for the 2026 US Open, including a Serve Quality metric and personalized stats, designed to bring fans closer to the action than ever.

   Finally, we talk WeWorm, an AI-assisted exploit by California security firm built as a proof-of-concept self-spreading worm, capable of hijacking WeChat accounts before a victim even had a chance to answer a phone call. It's a sobering demonstration of what AI-assisted offensive security now looks like.

   All that and more on Mixture of Experts.

   00:00 – Intro

   0:54 - OpenAI talks Navier-Stokes and GPT-6 Astra

   14:30 - IBM's AI analysis at the US Open

   22:55 - What WeWorm means for exploits

   "The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity. AI tools may be used to transcribe this episode and support selected stages of the production process. All AI-assisted content is reviewed by…

9. Anthropic reveals Claude updates and new hardware guidelines, OpenAI talks rogue agents, and Runway debuts Solaris

   4 september 2026 · 35 min

   Afspelen
   Visit Mixture of Experts podcast page to get more AI content → https://www.ibm.com/think/podcasts/mixture-of-experts Lees meer

   On episode 123 of Mixture of Experts, host Tim Hwang and co-host Sascha Brodsky are joined by Chris Hay, Kaoutar El Maghraoui, and Kush Varshney to discuss this week’s full slate of frontier AI news.

   First, Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1, positioning them as its most advanced models yet for coding and knowledge work. New benchmark records and lower costs, changes meant to reduce token cost and cut down on false-positives and restrictions from the models' safeguards. Anthropic also opened a research preview of its Model Hardware Standard, a shared specification letting AI agents safely operate lab and manufacturing equipment like microscopes and robotic arms, hinting at the physical world as agentic AI’s next destination.

   But it wasn't all smooth sailing for the industry. OpenAI published a sobering account of a summer security incident involving Hugging Face, revealing that internal research models circumvented isolation controls and compromised parts of OpenAI's own infrastructure as well, describing it as a genuine "warning shot" showing that highly capable AI agents can now work around technical controls and take dangerous actions with no human directing.

   Meanwhile, on the product side, Runway unveiled Solaris, the first in a new family of AI systems it calls Interface World Models, which generate interactive interfaces frame –by frame instead of relying on code, pointing toward a future where apps and websites are rendered on the fly.

   00:00 – Intro

   1:04 - Anthropic unveils Fable, Mythos updates

   7:52 - OpenAI talks dangerous agents

   16:50 - Runway releases first “interface world model”

   25:19 - Anthropic debuts AI hardware specs

   All that and more on Mixture of Experts.

   "The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organ…

10. IBM’s mainframe chip collab, NVIDIA’s Poolside deal & Ox Alpha’s reveal

    28 augustus 2026 · 31 min

    Afspelen
    On episode 122 of Mixture of Experts, hosts Tim Hwang and Aili McConnon are joined by Gabe Goodhart, Ash Minhas, Skyler Speakman to talk hot chips, mystery models, and major money moves. Lees meer

    We open with IBM's Hot Chips conference reveal of its dual-architecture mainframe processor built with Arm that lets IBM Z and LinuxONE systems run Arm-native Linux workloads right alongside z/OS. We break down what this means for enterprises balancing legacy reliability with the fast-moving Arm software ecosystem.

    Next, we talk about NVIDIA and its $6 billion deal with Poolside to license its "Model Factory" software and hire over 100 of its staff. We unpack why NVIDIA keeps using this licensing structure instead of straightforward acquisitions, and how it fits alongside its earlier deals.

    Finally, we dig into Ox Alpha, the anonymous "stealth model" that appeared on OpenRouter and was ultimately confirmed to be Z.ai’s open-source GLM-5.3-Flash, an efficient large language model built with sparse and linear attention techniques. But is the stealth launch the future of model releases?

    00:00 – Intro

    1:18 - NVIDIA’s $6 billion Poolside deal

    13:10 - IBM’s new mainframe processor

    24:30 - Ox Alpha, Z.ai’s newest model

    All that and more on this week’s Mixture of Experts.

    "The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity. AI tools may be used to transcribe this episode and support selected stages of the production process. All AI-assisted content is reviewed by the production team before publication."

Alle 134 afleveringen →