Claude code needs a mobile app where it can remotely let me approve things it’s doing and monitor the progress Cc @anthropic.com

It’s funny they say we don’t know what denisovans look like when I feel we can guess based on their geography and the modern humans in those areas, what some of their prominent features were arstechnica.com/science/2025...

Google DeepMind has updated TimesFM to 2.5. A pre-trained model for times-series forecasting, which performs great in zero-shot out-of-the-box - 200M params (down from 500M) - 16k context (up from 2k) Model: huggingface.co/google/times...

Open-source Hunyuan3D‑Omni lands as a playful but serious rival to commercial 3D control systems. 🚀 Built on Hunyuan3D 2.1, it acts like a “ControlNet for 3D,” fusing multi‑condition inputs to fix single‑view occlusions and messy geometry.

Microsoft just announced two new Office features: • Agent Mode in Excel/Word to auto-generate or check content, and • Office Agent (Anthropic-powered) to spin up slides/docs from Copilot chat. Honestly shocked these weren’t already there, feels like the most obvious Office + GenAI combo.

Sonnet 4.5 Better than Opus 4.1 on almost every benchmark Still the classic Sonnet prices, $3/$15

This is interesting about Anthropic's Claude Sonnet 4.5 - Apollo Research is stating that Claude Sonnet 4.5 has a high awareness when it is being tested/evaluated (58% vs. 22% for Opus-4.1). - GosuCoder is reporting that they cannot find any difference between Sonnet 4 vs. Sonnet 4.5.

Pre-owned electric cars and trucks in America are now almost as cheap as used gas-powered vehicles, on average, and are selling even faster, according to Cox Automotive, a services and technology company. Gift link for Kyle Stock's @bloomberg.com story: www.bloomberg.com/news/article...

DeepSeek drops V3.2-Exp: faster long-context, lower bill. Yes, both. Built atop V3.1-Terminus, the experimental V3.2-Exp debuts DeepSeek Sparse Attention for fine-grained sparsity—aiming to keep output quality while cutting compute on long inputs. Oct 15, 2025 (UTC).

Wrote up my initial impressions of the brand new Claude Sonnet 4.5 - I think it may live up to Anthropic's claims of being the "best coding model in the world", for the next few weeks at least! simonwillison.net/2025/Sep/29/...

This is the President explicitly threatening to withhold federal funds from New York City if voters elect a candidate he opposes. This is an unlawful attempt to coerce voters through federal financial threats and a textbook example of voter intimidation to influence the outcome of a local election.

Doing a deep dive into Apple MLX tonight Biggest challenge is Swift… as a long time C# and TypeScript coder, the syntactic sugar makes things harder to figure out for a beginner Thankfully we have Ai now to explain and translate the concepts in Xcode

Meet HunyuanImage 3.0—a native multimodal, MoE‑Transfusion system that fuses Diffusion with an LLM. It claims closed‑model parity while staying open, built for creators who need crisp typography, complex prompts, and fast workflows.

Just got Alexa Plus early access I’m kinda liking it— pretty futuristic

I try to avoid politics on this account but this is a confluence of social media and dark web and ai and conspiracy and propaganda I can only ever remember seeing before in anime And the media can’t cover it because Congress/public could never digest how deranged this is

79. American citizen.

Google DeepMind's paper on Veo3 "Video models are zero-shot learners and reasoners" "Veo 3 shows emergent zero-shot abilities across many visual tasks, indicating that video models are on a path to becoming vision foundation models—just like LLMs became foundation models for language."

This has been talked about a lot already but I just want to call out one salient fact: one million gallons of water is about what 10,000 people use in a day. If you put 10,000 people in the same area as this now-rejected data center, we'd be looking at typical urban mixed-use.

BYD's All-Electric Hypercar Hits 308 MPH, Becomes Fastest Car in Production https://hardware.slashdot.org/story/25/09/27/0441242/byds-all-electric-hypercar-hits-308-mph-becomes-fastest-car-in-production?utm_source=rss1.0mainlinkanon&utm_medium=feed

mathematicians: This is the Schlüsman-Katsuhira transform. It took researchers over 250 years to discover. Only 30 people in the world know how it works. tech artists: This is the Schlüsman-Katsuhira transform, or as we call it in the biz, "ol' Schlussy". We use it for realistic cheese textures :)

Suno drops a first-of-its-kind generative audio workstation—Studio—built for creators blending AI with classic music-making. 🎛

How loud does it crinkle…

One of the reasons this will be valuable for academics is that it provides a very clear recipe for pretraining ~3B-scale models on custom corpora.

Sam Altman’s description of ChatGPT Pulse is practically a sales pitch for an ad network. You tell ChatGPT your favorite vacation destinations and later on it gives you “updates” about these destinations? 🤔 Sounds like ads.

49% of high school girls said their mental health was not good most of the time or always during the COVID-19 outbreak – double the share of boys who said this (24%). www.pewresearch.org/short-reads/...

Some big implications for journalism, too, where reporters often mistake trending posts as a reasonable reflection of public opinion.

Well don't I feel stupid

I felt very moved by this too. Not because we yet have a cure, but because of the reminder of the passionate and humane commitment that goes into such research, contrasted with the vicious idiocy with which medical research is being treated right now by the US administration.

A single model built like an LLM that can adapt to any configuration of servos and limbs and handle failures This is really cool work. Something like this could become an expert tradesman, because humans use tools like an extension of ourselves

It’s funny I used to think movies and TV shows kept the bad guys in masks because it’s easy to just reuse actors when they’re masked But ICE has shown me the bad guys wear masks to hide their shame and guilt and cowardice

The Maquis and Bajoran Resistance storylines in DS9 are some of the best in the series, and Andor contains a lot of the nuance and questions of what made them excellent

The most unrealistic post of Andor is that anyone would care about a heartfelt Senate speech

Are we experiencing a 'GPT moment' in vision? In our new preprint, we show that generative video models can solve a wide range of tasks across the entire vision stack without being explicitly trained for it. 🌐 video-zero-shot.github.io 1/n

🔥Veo 3 has emergent zero-shot learning and reasoning capabilities! This multitalented model can do a huge range of interesting tasks. It understands physical properties, can manipulate objects, and can even reason. arxiv.org/abs/2509.20328 video-zero-shot.github.io Examples in this thread!

Google finally added image input to the Gemini actions for iOS shortcuts— this opens up a lot of scripting possibilities ChatGPT still missing these features

Suno just dropped v5, and it’s rewriting the rules of AI music creation! Meet the world’s best music model—now live for Pro and Premier users. Expect richer audio, lifelike vocals, and creative control that’ll make you rethink what’s possible. Try it now: suno.com/invite/@luok...

🚀 Next-level video creativity just got turbocharged! Kling AI has unveiled its 2.5 Turbo Video Model, promising a leap in creative power for video generation. With enhanced realism, support for both anime and real-life styles, and a more affordable price, it’s set to shake up the AI video scene.

The world needs a new Guitar Hero now more than ever

Meet Hailuo AI Agent — the next-gen creative sidekick that turns your ideas into visuals, fast! Just say what’s on your mind and watch it come alive on an infinite canvas. Free for one week: Try images, videos, and audio creation in one place.

The novel EVOREASONER framework enhances LLMs' dynamic question-answering capabilities by incorporating temporal reasoning and knowledge graph evolution, markedly improving performance and enabling smaller models to compete with vastly larger ones. https://arxiv.org/abs/2509.15464

Designating Antifa as a “domestic terrorist organization” is legally meaningless. But ominous. Designating narcos as “terrorists organizations” didn’t authorize use of force, but nonetheless paved the way for lethal strikes in the Caribbean on “terrorists.” www.whitehouse.gov/presidential...

I tried to capture everything that's strange about the TikTok deal. A giveaway to Trump's allies and another step toward US media being controlled by his allies www.platformer.news/tiktok-deal-...

Qwen drops Image Edit: apache 2 that feels on par with Nano Banana qwen.ai/blog?id=7a90... Qwen3-Omni: unified image, text, audio and video, like GPT-4o huggingface.co/Qwen/Qwen3-O... Qwen3-TTS: multi-timbre, multi-lingual TTS qwen.ai/blog?id=b426...

I hope nvidia knows what they're doing because some of their recent moves seem like accounting no-nos

Unitree G1 has mastered more quirky skills😱 With its new “Anti-Gravity” mode, G1 can stay stable through wild moves—and if it takes a tumble, it pops right back up. Stability boost: Any action sequence, instant recovery.

I just tried out Wan 2.2 Animate, and the results are so convincing it’s hard to believe they’re AI-generated.

America needs this Even since 2020 there’s been a drastic collapse in how much the middle class owns in the country, and we basically just need to do redistribution

I've had a programming project going for years that has been hung up on a section I did not know how to approach. Not insurmountable but needing more investment than I currently have. I put my project into Codex and within 10 minutes, it was perfected. Everything else I want to do is now available.

Introducing the SuperStation one. An open-source PS1 FPGA gaming console that supports original games, memory cards, and controllers. Load games from a disk or a backup. Region free. Supports all MiSTer FPGA cores, including N64 & Sega Saturn. Learn more: retroremake.co

Unitree doing some amazing work on RL for robotics www.reddit.com/r/singularit...