Listen to today’s podcast: https://www.youtube.com/channel/UC-nqwUyvLDEvs7bV985k-gQ
Welcome back to the show. Today’s podcast episode was created from the following stories: a mix of product updates, policy shifts, research breakthroughs, and market shakeups shaping how we build and use AI right now.
I tried ChatGPT-5.1’s personality presets—cynical and quirky are my favorites
By Katherine Li | Published: November 16, 2025
OpenAI rolled out GPT-5.1 with seven personality presets and sharper instruction-following after backlash over model changes. Early testing suggests accuracy is largely consistent across personas, but tone and engagement vary — raising questions about user attachment and perceived agency. Customization can rebuild trust, yet it may also heighten frustration when models change again.
La privacidad era el as en la manga de Apple en la era de la IA. Google acaba de arrebatárselo
By Javier Pastor | Published: November 17, 2025
Google unveiled Private AI Compute, a fortified cloud that keeps prompts and outputs isolated — even from Google — using Titanium Intelligence Enclaves and remote attestation. It answers Apple’s Private Cloud Compute, though Google’s third-party verification is less open than Apple’s verifiable transparency approach. If it delivers, privacy could become a true competitive feature for AI assistants, especially with Gemini set to power the next Siri.
AI生成曲にはラベルが必要？ 97%がAI生成音楽を聴き分けられず
By ECE YILDIRIM – Gizmodo US | Published: November 16, 2025
DeezerとIpsosの調査では、97%の人がAI生成曲と人間制作の曲を聞き分けられず、8割がラベル表示を望むと回答。プラットフォームはクレジットや表記の標準化に動く一方、著作権訴訟や政策論争も加速。透明性、プロビナンス、ライセンシングがストリーミングの新常識になりそうです。
Managers managing managers is out. Managers managing AI agents is in
By Lakshmi Varanasi | Published: November 16, 2025
AI agents that autonomously execute tasks are giving rise to a new role: the agent manager. Vercel even consolidated a 10-person sales team around an agent trained on its top performer, shifting the job toward technical orchestration, QA, and workflow design. Expect companies to upskill existing staff first, reshaping middle management rather than eliminating it.
华为将发布AI领域突破性技术，有望解决算力资源利用效率难题
By Unknown | Published: November 16, 2025
华为预告一项通过软件创新显著提升算力利用率的技术，将GPU、NPU等资源的利用率从30–40%提升至约70%。方案据称可统一管理英伟达、昇腾等多方算力，屏蔽硬件差异，为训练与推理释放更多产能。在算力受限的市场环境下，这类效率提升与新增硬件同样具有战略意义。
A single beam of light runs AI with supercomputer power
By Aalto University | Published: November 16, 2025
Aalto University researchers demonstrated single-shot tensor computing that encodes data in light and completes operations as light passes through an optical system — no electronics required. The method could be integrated into photonic chips within a few years, promising dramatic speed and energy savings for AI workloads. It is early, but points toward a post-silicon path for accelerating deep learning.
Vintage large language models
By Unknown | Published: Not provided
This talk explores training models only on data available up to a specific year to backtest forecasting and scientific discovery. The concept faces challenges like data leakage, volume, and cost, with proposals such as synthetic data bootstrapping and chronological training forks. It could become a useful framework for epistemic AI research and for simulating counterfactual intellectual histories.
Is Perplexity the first AI unicorn to fail?
By MD ZAID ANWAR | Published: July 3, 2025
A critical analysis argues Perplexity looks like a wrapper squeezed by giants with superior models and distribution, pointing to risky vanity metrics from a massive free Airtel bundle and reliance on third-party models. Meanwhile, OpenAI and Google have integrated AI search into products with built-in reach, eroding Perplexity’s early edge. The broader takeaway: without model ownership or a distribution moat, AI search startups face brutal headwinds.
