#154 - Google Gemini, Waymo Collision, Smaug-72B, EU AI Act final text, image watermarks

Our 154th episode with a summary and discussion of last week's big AI news!

Read out our text newsletter and comment on the podcast at https://lastweekin.ai/

Email us your questions and feedback at [email protected] and/or [email protected]

Correction: Andrey mentioned "State space machines", he meant "State space models"

Timestamps + links:

• (00:00:00) Intro / Banter

• Tools & Apps • (00:02:06) Google Releases Gemini, an A.I.-Driven Chatbot and Voice Assistant • (00:05:56) Copilot gets a big redesign and a new way to edit your AI-generated images • (00:09:40) Arc Search's AI responses launched as an unfettered experience with no guardrails • (00:12:40) Brilliant Labs’s Frame glasses serve as multimodal AI assistant • (00:15:30) Stability AI launches SVD 1.1, a diffusion model for more consistent AI videos • (00:16:18) OpenAI launches ChatGPT app for Apple Vision Pro •

• Applications & Business • (00:19:04) A Waymo robotaxi hit a cyclist in San Francisco – here’s what happened • (00:23:30) Canon plans to disrupt chipmaking with low-cost “stamp” machine • (00:27:33) US industry group calls for multilateral chip export controls to address disadvantage over Korea, other allies • (00:30:10) U.S. blocks shipment of 24 Nvidia AI GPUs to China over concerns about self-driving truck company • (00:32:24) Nvidia reportedly selects Intel Foundry Services for GPU packaging production — could produce over 300,000 H100 GPUs per month •

• Projects & Open Source • (00:37:23) Allen Institute for AI launches open and transparent OLMo large language model • (00:42:46) Meet ‘Smaug-72B’: The new king of open-source AI • (00:47:02) Introducing Qwen1.5 • (00:50:54) Hugging Face launches open source AI assistant maker to rival OpenAI’s custom GPTs • (00:53:20) Apple releases ‘MGIE’, a revolutionary AI model for instruction-based image editing •

• Research & Advancements • (00:54:30) Learning Universal Predictors • (01:01:00) Can Mamba Learn How to Learn? A Comparative Study on In-Context Learning Tasks • (01:04:40) MusicRL: Aligning Music Generation to Human Preferences • (01:05:47) FP6-LLM: Efficiently Serving Large Language Models Through FP6-CentricAlgorithm-System Co-Design • (01:09:06) AgentBoard: An Analytical Evaluation Board of Multi-turn LLM Agents • (01:12:36) Specialized Language Models with Cheap Inference from Limited Domain Data •

• Policy & Safety • (01:13:22) EU’s AI Act passes last big hurdle on the way to adoption • (01:17:04) Building an early warning system for LLM-aided biological threat creation • (01:23:44) FCC votes to ban scam robocalls that use AI-generated voices • (01:24:33) Biden administration names a director of the new AI Safety Institute • (01:26:23) OpenAI's GPT-4 finally meets its match: Scots Gaelic smashes safety guardrails •

• Synthetic Media & Art • (01:28:04) AI poisoning tool Nightshade received 250,000 downloads in 5 days: ‘beyond anything we imagined’ • (01:30:24) Labeling AI-Generated Images on Facebook, Instagram and Threads • (01:33:06) OpenAI is adding new watermarks to DALL-E 3 • (01:34:38) Following lawsuit, rep admits “AI” George Carlin was human-written •

• (01:36:20) Outro

Release date

Lydbog: 11. februar 2024