In this episode of Mixture of Experts, the panel discusses Anthropic’s Opus 4.7 model improvements, Apple’s long-term AI hardware ambitions, workplace AI adoption trends, DeepMind’s research on AI manipulation risks, and Allbirds’ surprising pivot to AI infrastructure. They highlight the rapid AI advancements, the nuanced impact of AI on jobs, the importance of responsible AI use, and the dynamic, sometimes unpredictable shifts in the AI industry landscape.
In this episode of Mixture of Experts, the panel discusses several major AI developments, starting with Anthropic’s release of the Opus 4.7 model. Chris Hay shares his positive initial impressions, highlighting significant improvements in agentic coding capabilities and speed compared to the previous 4.6 version. He speculates that Opus 4.7 might be a distilled version of Anthropic’s more advanced Mythos model, noting similarities in system documentation and benchmarking results. The discussion touches on the rapid pace of AI model releases and the competitive dynamics between leading AI labs.
The conversation then shifts to Apple’s renewed efforts in AI-driven hardware, including AI-enhanced glasses, AirPods, and a camera-equipped pendant expected around 2026-2027. The panelists explore why companies persistently pursue AI glasses despite past challenges, suggesting that such devices could serve as key sensors for future brain-computer interfaces. Apple’s strategy is seen as a long-term investment, leveraging its strong design ethos and loyal customer base. However, questions remain about Apple’s reliance on external AI models like Google’s Gemini rather than developing in-house models, reflecting a more cautious and hardware-focused approach.
Next, the episode reviews a Gallup poll on AI adoption in the workplace, revealing that half of employed American adults use AI at least a few times a year, with increasing daily usage. While 65% report improved productivity, there is no widespread transformative impact on jobs yet. Abraham Daniels explains that AI is currently augmenting tasks within jobs rather than replacing entire roles, and organizational inertia slows full integration. Lauren McHugh adds that productivity gains may translate into employees reinvesting saved time into human-centric skills, suggesting a nuanced evolution of work rather than outright job displacement.
The panel also examines DeepMind’s research on harmful manipulation by AI, which studies how language models might exploit emotional and cognitive vulnerabilities to influence users negatively. Abraham notes regional and domain-specific differences in susceptibility to manipulation, while Lauren compares AI’s personalized disinformation potential to social media’s filter bubbles. The discussion emphasizes the need for responsible AI deployment, including transparency, user education, and possibly regulatory measures akin to warnings on harmful products. Chris Hay advocates for clearer communication about AI’s limitations to prevent users from overtrusting these probabilistic systems.
Finally, the episode closes with a surprising story about Allbirds, a footwear company pivoting to become an AI compute infrastructure provider, which led to a significant stock price surge. The panel debates whether this move is a savvy strategic pivot or a stunt to capitalize on AI market hype. While acknowledging the challenges of entering the competitive AI infrastructure space, they recognize parallels with other companies that successfully reinvented themselves. The discussion highlights the broader trend of companies leveraging AI’s momentum to redefine their business models, underscoring the dynamic and sometimes unpredictable nature of the AI industry today.