AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Google introduces models Gemini 1.5 Flash and Pro for public use in their Cloud, offering faster performance and lower latency compared to GPT 3.5 Turbo. These models compete on price with features like context caching to store and reuse information.
Meta is nearing the release of their Lama Free 400 billion model, matching GPT-4 on major benchmarks. Despite potential usage limits, its impending launch on platforms like WhatsApp points to significant advancements in large-scale language models.
Runway unveils Gen 3 Alpha AI Video offering improved video generation capabilities with faster speeds and better results, marking a significant upgrade from its previous version. Users can now access more advanced features and editing options.
Google announces the inclusion of AI features in their Pixel 9 smartphone, allowing users to access tools like Adme, AI image generation, and Pixel Screenshot, enhancing user experience on the device. This move aligns with Apple's strategy of integrating AI features into consumer products.
The addition architecture in AI demonstrates scaling curves with steeper performance improvements compared to traditional transformer models. The curve indicates faster performance gains as more computational power is added to the model. Despite extrapolating from limited data points, the architecture intersects current open source models' scales, suggesting potential performance advantages if further validated.
An evaluation of AI agents underscores the significance of considering costs alongside accuracy when gauging performance. The study argues for standardized benchmarking practices that incorporate actual costs and failures modes like overfitting on single tasks. The research highlights limitations in current agent architectures' cost-effectiveness and advocates for comprehensive assessments beyond just accuracy measurements.
A new strategy in reinforcement learning addresses the issue of forgetting previously obtained knowledge during further training. By incorporating a weighted average rewarded policies approach, the method enhances optimization while retaining original information, ensuring a balance between adapting to new tasks and preserving existing knowledge. The technique involves multiple model iterations trained on reward feedback, merging insights to form an aligned and optimized model without deviating from the original behavior.
Our 173rd episode with a summary and discussion of last week's big AI news!
With hosts Andrey Kurenkov (https://twitter.com/andrey_kurenkov) and Jeremie Harris (https://twitter.com/jeremiecharris)
See full episode notes here.
Read out our text newsletter and comment on the podcast at https://lastweekin.ai/
If you would like to become a sponsor for the newsletter, podcast, or both, please fill out this form.
Email us your questions and feedback at contact@lastweekinai.com and/or hello@gladstone.ai
In this episode of Last Week in AI, we explore the latest advancements and debates in the AI field, including Google's release of Gemini 1.5, Meta's upcoming LLaMA 3, and Runway's Gen 3 Alpha video model. We discuss emerging AI features, legal disputes over data usage, and China's competition in AI. The conversation spans innovative research developments, cost considerations of AI architectures, and policy changes like the U.S. Supreme Court striking down Chevron deference. We also cover U.S. export controls on AI chips to China, workforce development in the semiconductor industry, and Bridgewater's new AI-driven financial fund, evaluating the broader financial and regulatory impacts of AI technologies.Timestamps + links:
Listen to all your favourite podcasts with AI-powered features
Listen to the best highlights from the podcasts you love and dive into the full episode
Hear something you like? Tap your headphones to save it with AI-generated key takeaways
Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more
Listen to all your favourite podcasts with AI-powered features
Listen to the best highlights from the podcasts you love and dive into the full episode