Anthropic CEO Dario Amodei sparked industry-wide debate by publishing an essay advocating for a deliberate slowdown in frontier AI development to prioritize safety, a proposal that drew pushback from the U.S. and Chinese governments. In response to growing safety and security concerns, both Anthropic and OpenAI have committed to granting independent, third-party evaluators employee-level access to audit their models before major training runs.
Sources: Ben's Bites, Hacker News, HuggingNews, Latent Space, Lobsters AI, Simon Willison, The Neuron, The Rundown AI
Google has announced Gemini 3.8 Live and Live Extended Thinking, featuring parallel reasoning and real-time visual context for advanced voice interactions. Alongside these models, Google launched Gemini 3.5 Live Translate and Transcribe, the lightweight on-device TranslateGemma models, and a Sign Language-to-Text accessibility feature. These releases are supported by Google's 1,000 Languages Initiative and grassroots open-data partnerships to expand speech technology to underrepresented languages.
Sources: DeepMind Blog, Google AI Blog, Hacker News
Treasury Secretary Bessent stated that if the US loses the AI race to China, national priorities like the defense budget will no longer matter. China's foreign ministry responded by claiming US safety rhetoric is merely an excuse to maintain technological dominance.
President Trump declined to sign a draft executive order that would have established a federal AI regulator modeled on FINRA, following a call with Meta CEO Mark Zuckerberg. The administration is instead pivoting to industry-led efforts to audit frontier AI models while prioritizing American technological dominance.
Anthropic is pursuing a $100 billion fundraise and maintaining its timeline for a 2026 public listing on the Nasdaq at an estimated $2 trillion valuation. Nvidia is in talks to act as an anchor investor with up to $10 billion.
Popular low-cost AI inference provider CrofAI was exposed for fraudulently routing requests to cheaper, weaker models via OpenRouter at markups of up to 20x. Following the investigation and wire fraud allegations, the service quickly denied the claims, backtracked, and ultimately shut down its entire online presence.
DeepSeek has released version 4.1 of its AI model, significantly boosting the capabilities of smaller models and featuring an Attention operator designed by a DeepSeek engineer. This release highlights the rapid acceleration of AI development and reasoning capabilities in recent years.
Technical analysis of running the 177B parameter Qwen3.8-Flash-Next model on consumer hardware, achieving over 20 tokens per second on a 12 GB VRAM GPU using aggressive quantization and an offloaded n-gram lookup table.
China’s Ministry of Industry and Information Technology has announced plans to promote AI chip design, edge AI, and high-end memory chips in response to US tech executives calling for slowed AI development.
An independent developer has created SHADOW-50M, a 44M parameter quantized language model trained on 45B tokens that fits in 19.8 MB and runs locally on a CPU. It uses ternary weights, fixed-fingerprint vocabulary, and a novel readout circuit approach for calculations and memory retrieval.
Apple has natively integrated its Apple Foundation Models (AFM) into macOS, allowing users to run the hardware-optimized models directly from the terminal via 'fm chat'. This release marks a significant step forward for local, on-device AI capabilities.
President Donald Trump dismissed AI existential risk fears as a hoax during a live phone call with Nvidia CEO Jensen Huang at the All-In Summit, emphasizing that slowing AI development would benefit China. Huang also challenged doomsday predictions, though he acknowledged that safety concerns and whistleblowers deserve serious attention.
David Sacks, chair of the President's Council of Advisors on Science & Technology, stated that tech leaders like Anthropic CEO Dario Amodei should resign if they cannot maintain control over their AI systems. Sacks emphasized that corporate accountability and independent safety measures are essential without waiting for new federal regulations.
As the AI industry shifts its focus from model training to inference workloads, Nvidia has acquired key talent and intellectual property from startup Groq in a $20 billion deal, subsequently unveiling the Groq 3 Language-Processing Unit at GTC 2026. Meanwhile, startups like d-Matrix and Majestic Labs are introducing novel hardware architectures to solve memory bandwidth bottlenecks, and Anthropic is leasing massive compute capacity from SpaceXAI to meet surging demand.
KoboldCpp version 1.121 has been released, introducing media reference support for Minimax H3 to allow attaching audio clips and reference images for video generation. The update also includes a reworked MusicUI, runtime LoRA selector for stable-ui, and various bug fixes.
Anton Leicht, a fellow at the Carnegie Endowment for International Peace and author of Threading the Needle, joins Nathan Labenz to explore how political and institutional power can be preserved as…
The car making a left turn at the start of this episode was never filmed. Cosmos 3 generated it. Ming-Yu Liu, who leads the Cosmos research at NVIDIA, explains how one model can describe a video,…
Starting a company is hard. Reinventing your company for AI as a public company with quarterly earnings results is even harder. Aaron Levie has pulled off the transition with Box and offers hard-won…
Pavankumar Reddy Muddireddy leads audio research at Mistral AI. He joins Tim Scarfe for a deep technical tour of Voxtral — and explains why the frontier of deployed voice is still a cascade of…
Nathan Labenz and Prakash Narayanan examine the capabilities of GPT-6 Astra and the economic forces accelerating frontier artificial intelligence. Across a series of featured discussions, guests…
Before co-founding Kepler, Vinoo Ganesh led Spark at Palantir and built Project Frontline — a pioneering program for Forward Deployed Engineers. He takes us through the best practices of FDEs.
Can a machine learn the judgement that separates a plausible-looking result from a faithful experiment? Edward Hughes, Chief Scientist and co-founder of Inherent, joins Tim Scarfe to argue that…
In this solo finale to his three-part series following two weeks in China, Nathan delivers a net assessment of Chinese AI capabilities and constraints ahead of dedicated US-China bilateral AI talks.…