Elon Musk accuses OpenAI of anti-competitive behavior, including discouraging investment in competitors like XAI, misusing sensitive information, and engaging in self-dealing, which he claims are violations of antitrust rules.
Amazon's Nova family includes four text-generating models (Micro, Lite, Pro, Premiere) and multimodal models for image and video generation. These models are significantly cheaper than competitors like Claude and Anthropic, making them attractive for many use cases, especially for tasks that don't require top-tier performance.
Llama 3.3 is a 70 billion parameter model that performs on par with the larger 405 billion parameter Llama 3.1 model while being much smaller and cheaper. Meta achieved this through post-training techniques, showcasing significant progress in condensing model performance.
Adding ads to ChatGPT could help OpenAI monetize its large user base (300 million weekly active users) more effectively. However, it may also lead to concerns about censorship and prioritizing advertiser interests over user satisfaction, similar to criticisms faced by social media platforms.
Tenstorrent's main challenge is competing with NVIDIA, which has an annual release cadence for new GPUs. Tenstorrent is still on a two-year cadence, making it harder to keep up with NVIDIA's rapid innovation in the AI chip market.
Genie 2.0 is an AI model capable of generating interactive 3D worlds from a single image and text description. It differs from Genie 1, which generated 2D video game-like environments. Genie 2.0 can create consistent worlds with different perspectives and simulate interactions like bursting balloons or opening doors.
AI safety researchers like Rosie Campbell and Miles Brundage are leaving OpenAI due to concerns about the company's trajectory and focus on building AGI without sufficient emphasis on ensuring its safety and alignment with human values.
The Densing Law of LLMs introduces 'capacity density' as a metric to evaluate the quality of LLMs. It shows that open-source LLMs have been improving, with smaller models achieving better performance relative to their size, indicating progress in efficient model training and compression techniques.
The MONET model uses a mixture of monosemantic experts, where each expert corresponds to a specific concept (e.g., chemical compounds, programming languages). This approach improves interpretability by allowing researchers to identify and isolate specific concepts within the model, making it easier to understand how the model processes information.
China's export restrictions on critical minerals like gallium and germanium could impact U.S. semiconductor manufacturing, as these minerals are essential for components like power delivery systems and high-speed communication interfaces in AI chips. The U.S. is heavily dependent on Chinese supplies for these materials.
Our 192nd episode with a summary and discussion of last week's* big AI news! *and sometimes last last week's
Note: this one was recorded on 12/04 , so the news is a bit outdated...
Hosted by Andrey Kurenkov) and Jeremie Harris). Feel free to email us your questions and feedback at [email protected] )and/or [email protected])
Read out our text newsletter and comment on the podcast at https://lastweekin.ai/).
Sponsors:
The Generator -) An interdisciplinary AI lab empowering innovators from all fields to bring visionary ideas to life by harnessing the capabilities of artificial intelligence.
The AI safety book “)Uncontrollable") which is not a doomer book, but instead lays out the reasonable case for AI safety and what we can do about it. Max TEGMARK said that “Uncontrollable” is a captivating, balanced, and remarkably up-to-date book on the most important issue of our time" - find it on Amazon today!)
If you would like to become a sponsor for the newsletter, podcast, or both, please fill out this form).
Timestamps + Links:
(00:00:00) Intro / Banter
(00:02:34) Sponsor Break
Tools & Apps
(00:04:19) OpenAI confirms new $200 monthly subscription, which includes its o1 reasoning model)
(00:10:40) Amazon announces Nova, a new family of multimodal AI models)
(00:17:13) ElevenLabs launches GenFM to turn user content into AI-powered podcasts)
(00:20:21) Google’s new generative AI video model is now available)
Applications & Business
(00:23:56) Elon Musk files for injunction to halt OpenAI’s transition to a for-profit)
(00:29:40) Amazon Is Building a Mega AI Supercomputer With Anthropic)
(00:34:15) It Sounds an Awful Lot Like OpenAI Is Adding Ads to ChatGPT)
(00:38:23) A16z in Talks to Lead $200 Million Round in Black Forest Labs, Startup Behind AI Images on Grok)
(00:41:10) Bezos Backs AI Chipmaker Vying With Nvidia at $2.6 Billion Value)
Projects & Open Source
(00:45:25) Meta unveils a new, more efficient Llama model)
(00:50:00) Alibaba releases an ‘open’ challenger to OpenAI’s o1 reasoning model)
(00:55:21) DeMo: Decoupled Momentum Optimization)
(00:57:01) PRIME Intellect Releases INTELLECT-1 (Instruct + Base): The First 10B Parameter Language Model Collaboratively Trained Across the Globe)
(01:03:03) Tencent Launches HunyuanVideo, an Open-Source AI Video Model)
Research & Advancements
(01:09:23) DeepMind’s Genie 2 can generate interactive worlds that look like video games)
(01:16:43) Language Models are Hidden Reasoners: Unlocking Latent Reasoning Capabilities via Self-Rewarding)
(01:20:40) Densing Law of LLMs)
(01:25:59) Monet: Mixture of Monosemantic Experts for Transformers)
Policy & Safety
(01:30:56) Commerce Strengthens Export Controls to Restrict China’s Capability to Produce Advanced Semiconductors for Military Applications)
(01:37:33) China retaliates against latest US chip restrictions)
(01:40:52) OpenAI Is Working With Anduril to Supply the US Military With AI)
(01:43:24) On Targeted Manipulation and Deception when Optimizing LLMs for User Feedback)
(01:47:52) AI Safety Researcher Quits OpenAI, Saying Its Trajectory Alarms Her)
(01:51:52) Meta Claims AI Content Was Less than 1% of Election Misinformation)
(01:55:05) Outro