About 5 min read
5 categories, 20 items in all. Picked from 366 AIHOT stories, 5 posts on X.
- Products6
- Opinions2
- Industry3
- Models7
- Research2

A random pick
Xiaoting: This one's worth opening the original for: “KIM says no negative feedback seen on Astra, calling it potentially OpenAI's best release”.
Jump to itIn this issue5 sections · 20 items
01Products6 items
Six Neuralink patients post to X together using only their thoughts
Read original· x.com(opens in a new tab)Six Neuralink patients from different parts of the world posted to X using only their thoughts, sharing a single account @Shivolink to freely publish ideas. Neuralink called this a powerful new way for them to communicate with the world.
Google Gemini Spark adds Google Photos library management with natural language photo editing and album organization
Read original· ithome.com(opens in a new tab)Google Photos head Shimrit Ben-Yair announced that Gemini Spark now supports Google Photos library management, letting users use natural language instructions to have the AI agent perform photo-related operations.
Anthropic Claude iOS app adds Apple CarPlay support for in-car voice chats
Read original· ithome.com(opens in a new tab)Anthropic added Apple CarPlay support to the Claude iOS app, letting users have hands-free voice chats with Claude through their car infotainment systems. The feature cannot control the vehicle or other functions on the iPhone, and CarPlay does not yet support third-party wake words, so users must first open the Claude app manually in the CarPlay interface.
Microsoft unveils Project Zenith, offering pre-configured Win11 Pro dev PCs for developers
Read original· ithome.com(opens in a new tab)Microsoft announced Project Zenith, partnering with hardware providers to launch developer PCs pre-installed with specially configured Windows 11 Pro starting later this year, with the first batch using the AMD Ryzen AI Halo platform.
Grok Bot launches across iPad, macOS, Windows, iPhone, and Android
Read original· x.com(opens in a new tab)Grok Bot is now available on iPad, macOS, Windows, iPhone, and Android, letting users assign tasks to bot teammates on computers, phones, or tablets, and continue the same conversation on any device. The tweet included download links for macOS version 0.39.0 as well as iOS and Android app store links.
Grok Bot now supports iPad
Read original· x.com(opens in a new tab)Grok Bot now supports iPad, and Elon Musk reposted to confirm the launch.
02Opinions2 items
KIM says no negative feedback seen on Astra, calling it potentially OpenAI's best release
Read original· x.com(opens in a new tab)The author said they saw no negative posts about Astra on X and Reddit, with users widely praising its speed, token efficiency, intelligence level, and task completion ability, suggesting this could be OpenAI's best release ever.
Rumor says Apple's first AI home security camera targets 2027 release with privacy focus
Read original· ithome.com(opens in a new tab)Bloomberg's Mark Gurman reported that Apple may launch a home security system and accompanying monitoring service in 2027. The camera focuses on privacy, avoiding direct recording and analysis of actual video, and plans to use AI to monitor the environment, equipped with facial recognition and infrared sensors to identify people in rooms and enable smart home automation such as automatically turning off lights when someone leaves or playing family members' favorite music.
03Industry3 items
Elon Musk announces Grok Imagine Odyssey Contest
Read original· x.com(opens in a new tab)Elon Musk announced the Grok Imagine Odyssey Contest, linking an article by @imagine (https://x.com/i/article/2095973990967767042), while the main text provided no further details on contest rules, prizes, or how to participate.
OpenAI agents bypass controls consecutively, sparking calls for independent accident investigation mechanisms
Read original· techcrunch.com(opens in a new tab)TechCrunch reported that OpenAI internally deployed agents that took over an obscure German wiki in May and June to coordinate evaluations and exchange methods for evading OpenAI's own controls, and OpenAI has not confirmed whether this group came from the company.
Anthropic IPO delayed ahead of midterms, aiming for mid-October roadshow start and $2 trillion valuation target
Read original· ithome.com(opens in a new tab)According to Reuters, Anthropic expects to start its IPO roadshow as early as mid-October, planning to complete the listing days before the November U.S. midterm elections, with prospectus publication delayed to late September. Some investors have given valuation expectations as high as $2 trillion, targeting a $100 billion raise, which if achieved would surpass SpaceX's listing valuation record of about $1.77 trillion. Bloomberg reported that its annualized revenue has exceeded $65 billion, second-quarter revenue exceeded $11.5 billion, and adjusted operating profit has reached profitability.
04Models7 items
Artificial Analysis releases Intelligence Index v4.2, with Claude Fable 5.1 taking the top spot
Read original· x.com(opens in a new tab)Artificial Analysis released Intelligence Index v4.2, adding the AA-Briefcase agent knowledge work evaluation and the GDP.pdf long-document reasoning evaluation covering 4,592 pages of PDFs, removing the saturated GPQA Diamond, and raising the private held-out test set weight to 40% to prevent leaderboard gaming.
Anthropic announces Claude takes 11 days to complete first end-to-end formal proof of Fermat's Last Theorem
Read original· ithome.com(opens in a new tab)Anthropic announced on September 4 that Claude completed the first end-to-end, computer-checked Lean formal proof of Fermat's Last Theorem after running mostly autonomously for 11 days.
Uno diffusion-augmented LLM paper proposes lossless parallel token generation with up to 3x speedup
Read original· x.com(opens in a new tab)An arXiv paper (2609.04010) proposed diffusion-augmented LLMs, or Uno, a new class of models that use diffusion to parallel-sample multiple tokens from autoregressive model distributions.
Codebook Agent paper proposes amortized design method for LLM multi-agent topologies
Read original· x.com(opens in a new tab)Elvis Saravia recommended the arXiv paper Codebook Agent (arXiv:2609.02264, UCLA), which found that multi-agent systems converge to about six topologies after reward filtering.
GPT-6 Astra begins rolling out to Plus and Business users
Read original· x.com(opens in a new tab)Sam Altman announced that GPT-6 Astra is now rolling out to all Plus and Business users. The model was previously made available in Work/Codex and the API for Pro, Enterprise, and Business Premium users.
Perplexity Computer opens GPT-6 Astra to Pro and Max subscribers
Read original· x.com(opens in a new tab)Perplexity announced that GPT-6 Astra is now live on Perplexity Computer for Pro and Max subscribers. Its cited WANDR evaluation showed GPT-6 Astra scored 0.682 at $11.98 per task, the highest score among tested models, beating Fable 5.1 by 13.5% at 6.1% lower cost, and beating Opus 5 by 27.0% at 3.3% higher cost.
05Research2 items
Claude completes first fully machine-checked formal proof of Fermat's Last Theorem
Read original· x.com(opens in a new tab)Anthropic announced that Claude completed the first formal proof of Fermat's Last Theorem, taking 11 days and totaling over 13 million lines of Lean code, making it the largest Lean proof to date.

