#180 - Ideogram v2, Imagen 3, AI in 2030, Agent Q, SB 1047
Last Week in AI
Exploring Sparse Autoencoders for Enhanced Model Interpretability
This chapter delves into the technical details of sparse autoencoders and their significance in understanding large language models. It argues that further training on these models can enhance interpretability and address challenges like deceptive alignment.
Our 180th 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)
If you would like to get a sneak peek and help test Andrey's generative AI application, go to Astrocade.com to join the waitlist and the discord.
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
Episode Highlights:
- Ideogram AI's new features, Google's Imagine 3, Dream Machine 1.5, and Runway's Gen3 Alpha Turbo model advancements.
- Perplexity's integration of Flux image generation models and code interpreter updates for enhanced search results.
- Exploration of the feasibility and investment needed for scaling advanced AI models like GPT-4 and Agent Q architecture enhancements.
- Analysis of California's AI regulation bill SB1047 and legal issues related to synthetic media, copyright, and online personhood credentials.
Timestamps + Links:
- (00:00:00) Intro / Banter
- (00:01:08) Response to Listener Comments / Corrections
- Tools & Apps
- (00:03:58) Ideogram AI expands its features with v2 model and color palette options
- (00:07:48) Google Releases Powerful AI Image Generator You Can Use for Free
- (00:11:41) Perplexity adds Flux.1 model for Pro users alongside Playground v3 update
- (00:13:58) Luma drops Dream Machine 1.5 — here’s what’s new
- (00:17:49) Runway’s Gen-3 Alpha Turbo is here and can make AI videos faster than you can type
- (00:20:21) Perplexity’s latest update improves code interpreter, charts included
- Applications & Business
- (00:24:14) AMD buying server maker ZT Systems for $4.9 billion as chipmakers strengthen AI capabilities
- (00:28:55) Ars Technica content is now available in OpenAI services
- (00:34:08) Anysphere, a GitHub Copilot rival, has raised $60M Series A at $400M valuation from a16z, Thrive, sources say
- 00:38:32 Stability AI appoints new Chief Technology Officer
- (00:41:45) Cruise’s robotaxis are coming to the Uber app in 2025
- Projects & Open Source
- (00:44:16) AI21 Introduces the Jamba Model Family: The most powerful and efficient long-context models for the enterprise
- (00:53:47) Microsoft reveals Phi-3.5 — this new small AI model outperforms Gemini and GPT-4o
- (00:57:33) Nvidia’s Llama-3.1-Minitron 4B is a small language model that punches above its weight
- (01:00:58) Open source Dracarys models ignite generative AI fired coding
- Research & Advancements
- Policy & Safety
- (01:38:20) California weakens bill to prevent AI disasters before final vote, taking advice from Anthropic
- (01:48:14) Personhood credentials: Artificial intelligence and the value of privacy-preserving tools to distinguish who is real online
- (01:52:44) Showing SAE Latents Are Not Atomic Using Meta-SAEs
- Synthetic Media & Art
- (02:01:43) Outro
The AI-powered Podcast Player
