Daily AI News Podcast · Alice Mallory and Bob Trent

Daily AI News Podcast: December 11, 2024 - Google Asks FTC to Break Up OpenAI / Microsoft Deal

·22 min·1 clip

Recorded Future says AI voiceovers were used in a campaign aimed at weakening European support for Ukraine.

1. Daily AI News Podcast on December 11, 2024 centers on OpenAI’s Canvas, Microsoft’s cloud deal, ServiceNow’s FastLLM, MIT bias research, and an Eleven Labs voice-cloning report. 2. The hosts frame the episode as a daily news roundup, and both speakers guide the discussion by summarizing the stories and reacting to the implications for AI development. 3. The episode asks what changes when AI becomes a writing partner, a cloud-power battleground, and a tool for both efficiency and manipulation. 4. OpenAI’s Canvas is described as a major departure from the standard chat window and as a more visual, interactive way to use ChatGPT. 5. The hosts say Canvas works like a digital whiteboard where people can edit text, write code, and receive contextual feedback in real time. 6. They mention shortcuts that can change writing length, debug code, and add emojis, which they present as part of the collaborative workflow. 7. OpenAI’s training approach for Canvas is said to use synthetic data to teach the model when to offer suggestions and how to interact naturally. 8. The cloud story says Google has approached the FTC because Microsoft’s deal gives it exclusive cloud hosting rights for OpenAI technology. 9. The hosts also mention that Amazon is among the competitors worried about competition and pricing in the cloud market. 10. ServiceNow’s FastLLM is presented as an open-sourced framework that claims to speed AI model training by 20%. 11. The hosts explain that the framework uses breadth-first pipeline parallelism to distribute work across multiple GPUs. 12. They also say FastLLM reduces memory fragmentation, which helps training run faster and more smoothly. 13. MIT researchers are said to have found a way to reduce bias in AI models without losing accuracy. 14. Their method targets specific data points that contribute to bias against minority groups instead of broadly balancing the data. 15. Another MIT Center for Constructive Communication study is described as finding left-leaning bias in reward models even when they are exposed to factual information. 16. The hosts say that bias became stronger as the models grew larger and more complex, especially on topics like climate change, energy policy, and labor unions. 17. The episode’s tone is fast-moving and explanatory, with the hosts translating technical terms like pipeline parallelism, memory fragmentation, and reward models into plain language. 18. The format alternates between news briefing and back-and-forth commentary, with repeated questions about competition, transparency, and misuse. 19. People tracking OpenAI, cloud competition, and AI safety will get the most from this episode. 20. Listeners wanting a deep technical interview or long-form debate may skip it.

As heard by us

A brisk AI roundup that starts with Canvas and ends with the politics of trust.

Daily AI News Podcast covers a familiar AI news cycle, but it does it cleanly. It starts with OpenAI's Canvas and the practical question of what changes when chat tools begin to feel like collaborators, then turns to deepfakes, detection, public skepticism, and the chance of…

Read the full review in PlayNext →

Why you'd press play

Want a fast AI-news rundown that connects Canvas, deepfakes, and bias?

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Daily AI News Podcast