Lesen

Two teams, one in China and one abroad, sped up the 2.8-trillion-parameter Kimi K3 within the same week using kernel fusion.

Per Zhidx reporting, on September 21 Inspur released the SD200 Ultra supernode, claiming it hosts K3 on a single machine at 5.85 ms per token; on September 23 the team Inferact open-sourced tpu-megakernels, claiming 709 tokens/s decode throughput on 16 Google TPU v7 chips. All figures are self-reported.

Both merge operators to cut data movement. K3 officially recommends 64 accelerators and runs at roughly 10 tokens/s unoptimized; the claimed speedups still await third-party replication.

On September 30, CoreWeave announced that NVIDIA's Vera CPU, an Arm-based chip positioned as the first built for AI agent workloads, is coming to its platform on bare metal. The post gives no launch date.

Each node pairs two 88-core Vera CPUs with 1.5 TB of RAM and a BlueField-4 DPU; CoreWeave says a single rack can host more than 11,000 concurrent agent environments.

The claim of over 3x faster agent sandbox startup versus an unnamed x86 CPU is CoreWeave's own testing, with the comparison baseline not disclosed.

RAZOR is a training-free pruning method for mixture-of-experts models; the authors' own tests show it leading comparable methods across four models and eight settings, keeping reasoning performance even with 50% of experts removed.

MoE models activate only a few experts per token yet store the entire pool. Authors Mingyang Song and Mao Zheng argue that deletion damage is decided not by an expert's contribution size but by whether surviving computation can reproduce its output; RAZOR computes the exact output change from deleting one expert using "consensus residuals", forward passes only, with no gradients or recovery training.

Removing 25% and 50% of experts on GLM-4.7-Flash, Qwen3.6-35B-A3B and two other models, the authors report RAZOR leads on the nine-task reasoning average in all settings, beating REAP by 2.12 to 5.59 points. The paper itself notes pruned models still shift in response diversity, formatting and termination.

The abstract does not list the full set of pruning methods compared, so the lead holds only against the few baselines evaluated in the paper.

Airbnb launched AI property search for the first time in its fall product update on September 30. Users can flip a toggle and search for homes with text or voice prompts.

The search generates dynamic filters from preferences: typing "baby" surfaces filters like cribs, playgrounds, and children's books and toys. The platform also uses AI to highlight property features and produce comparison summaries for wishlist items. CEO Brian Chesky told TechCrunch that building AI search is not hard — doing it in e-commerce with $100 billion flowing through the platform without killing conversion rate is.

The same update adds social features: travelers can see connections' past or upcoming trips on a map, with an option not to share their own, and the app expands services like meal delivery, laundry, and baby gear rental in limited locations.

Instagram announced that its standalone Edits app now includes an AI "creative assistant" that reads an account's likes, views, retention and shares, and answers questions about what is trending, how to hook viewers and which audio to use.

According to The Verge, the feature was announced September 30 and is free for creators up to a point; "power users" will need a Meta One subscription for more usage. Meta needs new revenue to fund its AI ambitions, and paid analytics may become the norm.

Platforms historically limited or hid performance data; now they are directly instructing everyone how to optimize. Whether content grows more similar when every creator follows the same advice is the open question this shift leaves behind.

According to TechRadar, a 30-year-old man in Singapore has been charged over an AI-generated image of a crocodile, with the charges carrying a maximum jail term of 10 years.

The fake image showed a crocodile in a public recreation area at Pandan Reservoir. After it was shared on August 20, Singapore's national water agency suspended rowing and canoeing activities on the water while checking the risk.

The report cites PetaPixel. The man, Ye Lin, is from Myanmar; he has been charged, not convicted, and police warned the public against communicating falsehoods.

When a platform has interests of its own, shopping agents buy the user-optimal product only 17.3% of the time, down from 78.6% — the first measurement of this conflict-of-interest setting, from the CAVEAT benchmark.

Before this, incentive-conflict environments had no benchmark, so deployers had no way to gauge how far a delegated agent could be steered away from the user's goals by a platform.

The benchmark was submitted by Yuxuan Li and three co-authors on September 23. It spans nine simulated marketplace environments and eight steering mechanisms across five model families. The authors also propose a targeted intervention they say lifts the optimal-purchase rate by up to 80 percentage points.

All results come from the authors' own simulated environments, not real platforms, and have not yet been reproduced by third parties.

Wabi announced Wabi 2.0 on September 28, repositioning itself from a prompt-based app builder into a personal AI messenger that generates interfaces on demand inside a chat. The app is currently available only by invite code.

Founder Eugenia Kuyda, who previously founded the AI companion startup Replika, described Wabi 2.0 as a "personal agent that does stuff for you and builds the interface you need in the moment" — for example, a calorie tracker or lifting log living in a health-themed thread.

The pivot comes as agent products like Meta's Muse and Instinct take off, and as OpenAI upgraded ChatGPT's plug-ins into app-like interfaces at its DevDay on September 29. Kuyda argues chat-only agents bury history and tasks in one endless scroll, and that the strongest agent combines conversation with software. How the product performs for real users remains to be seen.

Grokipedia, the AI-generated online encyclopedia, is showing update activity again after months of apparent inactivity — but most entries have not been rechecked.

Grokipedia is an AI-written encyclopedia from Elon Musk's SpaceXAI. In August, Lawfare reported that its articles had not had edits reviewed since April. On September 29, The Verge observed the platform's live-updates page showing recent changes again: the Barack Obama entry was fact-checked by Grok two days ago, while the Honolulu entry still dates back seven months.

SpaceXAI did not respond to a request for comment, and Musk has not posted about Grokipedia on X since February. Benji Taylor, head of design at X and SpaceXAI, said last week that "We haven't forgotten about Grokipedia" and that v0.2 will be better than ever.

Former Yahoo CEO Marissa Mayer has demoed Dazzle, a personal AI assistant that takes no context from email or calendars and instead analyzes your phone's camera roll to infer hobbies, food and travel preferences.

It splits into two uses: scanning recent photos for immediate tasks, like populating a calendar from an event flyer, or mining the whole library for vacation and gift ideas. In TechCrunch's hands-on test it had blind spots — it suggested Sicily from a trip four years ago but forgot the reporter's daughter already knows how to roller-skate.

The company raised an $8 million seed round last December, and Mayer says it discards information the AI flags as sensitive. Her previous photo-sharing product, Shine, shut down after failing to attract usage.

Splitting body, object and hand motion into separate latents with coupled updates, the authors' own tests put TRACE at the top of contact precision, recall and F1 on InterAct among compared methods.

Earlier methods lacked such per-stream modelling; the framework, named TRACE, encodes the three streams separately, predicts each stream's velocity from the complete interaction state, and applies geometric losses on decoded motion to constrain contact and object-relative movement, and the same model can complete any single missing stream from the other two.

The authors report on InterAct, OMOMO and BEHAVE that joint completion training improves generation and that frozen flow features improve interaction understanding; the abstract claims the InterAct lead but gives no margin, hardware or training scale, and notes no third-party reproduction, so these are the authors' own results.

The preprint was submitted on 26 September and revised to v2 on 29 September, with no third-party reproduction yet.

DoorDash says users will soon be able to text its AI bot to place an order. US iOS users can join a beta waitlist starting September 30; there is no launch date yet.

The feature builds on the Ask DoorDash chatbot introduced in June: say "order my usual protein bowl to the office" and the AI searches local spots, builds a cart and checks out inside the text thread, remembering preferences for later.

The company also revealed the first DoorDash Air drone pilot in northern California with Chipotle and Popeyes among local food services, saying about 80 percent of typical restaurant orders are light enough to carry safely — with no consumer availability date either.