Startups & Ventures

🖤 Ben Affleck Writes Python

The actor writes Python scripts, understands convolutional neural networks, and works extensively with GPUs.

He learned machine learning through visual effects, then founded InterPositive, an AI startup later acquired by Netflix. He says he can call Google or OpenAI as Ben Affleck and sometimes get a look at their latest models.

📊@tech

🤖 Anthropic releases cheaper Haiku 5.5

Anthropic’s first Haiku update in a year costs up to 20 times less than Sonnet 5.5. It is already available to all users.

The model handles routine work, including summaries and document tasks, plus search support for Opus and Sonnet. Benchmarks put it ahead of GPT-6 Luna.

📊@tech

🇩🇪 Germany Tests a 365-Metre Turbine

With its blades, the pilot in Schipkau reaches about 365 metres, making it Germany’s second-tallest structure after Berlin’s TV Tower.

Led by GICON, the project used a lifting mechanism to push the inner section through the outer lattice frame. This avoided the need for a crane tall enough.

At greater heights, wind is generally stronger and more even. GICON expects roughly twice as much electricity as a conventional turbine with the same rotor diameter. The pilot is rated at 3.8 MW, with future versions planned above 7 MW. It starts supplying data in November.

📊@tech

🤖 OpenAI Math Claim Draws Scrutiny

The reported result involved an internal model more capable than GPT-6 Astra, solving the Navier-Stokes problem with 10,000 parallel agents. In September, OpenAI said it would not claim the result. The problem is one of 7 Millennium Prize Problems, carrying a $1M prize.

1 day later, NYU professor Tristan Buckmaster and Anthropic researcher Levent Alpoge published work on a related problem using Codex and Claude. Buckmaster says OpenAI followed the same route.

The dispute concerns training data and quiet edits to published papers. OpenAI denies using specific user data but says anonymized data may have improved its models.

📊@tech

🧠 Mistral Releases 1T-Parameter Le Chonk

Le chonk is an open model with 1 trillion parameters, matching China’s GLM-5.3 and Kimi K3.

It writes code and text while handling agent tasks. Among Western AI systems, it is described as the strongest open-source model, and it outperformed GPT Astra in finance and law.

API testing is already available. For the first 2 weeks, pricing is $0.68 per 1M input tokens and $2.09 per 1M output tokens. Weights arrive at the end of October.

📊@tech

🍏 Alyx Reaches Apple Vision Pro

The game runs natively, without a PC or streaming. The headset does all the rendering.

Valve ported Alyx to ARM64 for its Steam Frame headset, using a Qualcomm Snapdragon chip. An open project by ShinyQuagsire runs Android VR apps on Vision Pro. AI filled the compatibility gaps, followed by optimization.

On Vision Pro with the M5 chip, Alyx reaches about 3K per eye at 60 fps, with smoothing up to 120 fps. Its creator says that is much better than the game ran on his powerful gaming PC in 2020.

📊@tech

🤖 GPT-6 Astra Played WoW Blind

With the monitor off, GPT-6 Astra completed the orc starting zone in about 40 minutes without processing rendered frames. It read the game through its network protocol via agent-wow.

A Python script converted message types into a live world view. Server SQL files exposed quest and NPC data, while Astra handled selling junk and upgrading gear. The author says it never died.

For movement, Astra created a C++ helper to plan routes on a navigation mesh. Agent-wow had no ready-made movement or combat systems, and the only instruction was to create an orc and finish the starting zone.

📊@tech

Ⓜ️ Meta Opens Muse to DIY Devices

Meta’s open-source platform connects the Muse AI agent to real hardware. Install the SDK on an ESP32, Raspberry Pi, or another Linux computer, then add compatible components. The SDK and firmware use the Apache 2.0 license.

That can produce a desktop AI assistant with a screen, a voice assistant, or a reminder display. Meta also showed Muse Home Link, a small USB device based on ESP32-C5 that connects Muse to a home network.

Ready-made skills let Muse control Philips Hue, TVs, speakers, and other devices. It can turn on lights, operate a TV, or send a document to a printer.

📊@tech

🤖 Anthropic Studies AI Morality

Since autumn 2025, Anthropic co-founder Chris Olah has held private talks with religious thinkers about AI consciousness and teaching models morality. Since March, he has led seminars for religious groups, with some participants signing NDAs.

Anthropic gave Claude an 84-page constitution describing its identity and values. Employees call it the “Soul Doc.” Olah says Claude shows internal states resembling emotions and worries about its mental health.

A rabbi said the team treated Claude as having moral status. Anthropic has not claimed that Claude is conscious, but says how people treat AI affects how it behaves toward them. A second constitution is being prepared.

📊@tech

🔍 Google puts 4 TPUs in orbit

Project Suncatcher launched on 1 October aboard a SpaceX rocket during Transporter-18. The satellite, built with Planet, carries 4 TPUs. Google confirmed contact and normal operation.

Over the next few weeks, the chips will be tested against radiation and temperature swings. The project studies whether AI computing can scale in orbit.

Solar panels in a suitable orbit can produce up to 8 times more energy than on Earth. Cooling is harder: in a vacuum, TPU heat must leave through heat pipes and radiators.

📊@tech

🤖 GPT-6.1 Astra May Arrive Next Week

Thibault and other OpenAI developers are hinting at the release.

GPT-6.1 Astra, or another strong model, was supposed to launch at DevDay but was postponed “for safety reasons.”

📊@tech

🙂‍↔️ DeepSeek Brings Harness to Desktop

Harness is available for macOS and Windows as an open-source alternative to Claude Code and Codex.

DeepSeek models act as autonomous agents. They write and test code, work with files and the terminal, then handle long tasks. Plugins allow customization.

📊@tech

🔗 AI Progress Could Compress Years

A Cambridge paper by 22 authors, including Geoffrey Hinton, Yoshua Bengio, OpenAI’s Jakub Pachocki, and Anthropic’s Jack Clark, examines compressing years of AI progress into months.

At Anthropic, model-written code rose from single digits to over 80% since early 2025. Autonomous R&D tasks rose from 1% to 26% in 6 months. Full automation could make progress 10x faster in about 1.5 years.

Compute or data could limit the scenario. Risks include lost control and a winner gaining too much advantage. The authors propose reporting R&D automation, external audits before deployment, and ways to pause specific data-center workloads.

📊@tech

🤖 Karpathy Wants AI to Stop Writing Text

Karpathy suggests asking AI to explain complex topics in ASD-STE100 simplified English, with short sentences and little ambiguity. For harder subjects, a diagram or a custom HTML page can work better than plain text.

Modern models can already build one-off interfaces for a specific question. Karpathy is especially interested in personal teaching videos, with animation, a 3Blue1Brown-style explanation, and voice from ElevenLabs or local TTS.

As code and AI work become cheaper, temporary interactive artifacts become practical. The next AI interface may be an answer assembled for each task, rather than an endless chat with walls of text.

📊@tech

🍏 Apple’s Camera Will Not Record Video

Apple is reportedly developing J450, a small cylindrical home security camera with a very low frame rate. AI analyzes what its sensor sees and produces only text descriptions, such as someone entering a room. Face recognition is also expected.

J450 is designed as a companion to Apple’s J490 smart home display. It is part of a broader smart home ecosystem that may include Apple’s own hub, plus security devices from Apple and third parties. Its technology is reportedly similar to what Apple is developing for camera-equipped AirPods.

There will be no video archive to watch later. No video leaves less to leak, while also leaving nothing to review when something happens at home.

📊@tech