Reading

Research firm Counterpoint forecasts that annual global shipments of smartphones priced under $200 will fall by more than 230 million units between 2025 and 2030, a drop of about 40%.

The report attributes the shift to rising memory and chipset prices, a higher minimum specification floor, and vendors pulling back from non-strategic markets. Counterpoint notes the lost volume will not simply migrate to higher price tiers: some consumers will delay upgrades, and some demand will disappear entirely.

The overall market is expected to recover to roughly 1.2 billion units by 2030, with premium phones growing about 7% a year. In low- and middle-income countries, phone price is a main barrier to internet adoption, so the entry-tier shrinkage will slow the move from feature phones to smartphones.

Microsoft's 2026 Surface PCs no longer carry the Copilot+ PC label, according to an IT Home report citing Windows Central's September 24 post.

Copilot+ PC is the Windows hardware category Microsoft created with Qualcomm in 2024, setting a baseline for AI-first devices with a dedicated Copilot key. After the core Recall feature was delayed over security concerns, the brand's reputation suffered, and over the past two years a growing number of OEMs have dropped the label from marketing and product names.

Microsoft Surface vice president Brett Ostrum confirmed the new devices still meet all the original hardware standards despite the dropped name; Qualcomm computing chief Kedar Kondap said the term may simply no longer be used.

At the Snapdragon Summit in Hawaii, Qualcomm executive Chris Lattner announced that Microsoft will partner with Modular to provide native Windows support for the Mojo programming language.

Mojo is a language for AI and high-performance computing, developed by Modular, with Python-like syntax aimed at systems-level performance; version 1.0 has been open-sourced. Qualcomm completed its all-stock acquisition of Modular on July 29 at a valuation of roughly $3.9–4.0 billion, aiming to let the same AI software run across GPUs, NPUs and data-center chips.

Qualcomm said the Modular software platform will enter its full Snapdragon portfolio, spanning phones, PCs, cars and wearables. These remain vendor statements from the keynote; no timeline for the adaptation work was given, and actual delivery needs to be tracked.

Google has added a Live Avatar feature to Gemini 3.8 Live voice conversations: while talking, users can watch a real-time animated persona that lip-syncs and changes facial expressions.

According to The Verge, the feature is currently available only to Gemini Enterprise customers. Google says the avatar can switch among the 97 languages it supports without degrading video fidelity, and organizations can build their own avatars. The demo videos are Google's own.

The launch comes about a week after the Gemini 3.8 Live model was revealed, and output carries an invisible SynthID watermark. There is no independent feedback yet on lip-sync or multilingual quality; enterprise adoption is the thing to watch.

Google has started testing a new feature called "Call for Me" that lets Gemini call businesses on a user's behalf. It is initially available to Pixel 11 owners in the U.S. who pay for a Gemini subscription, and requires the beta version of Google's Phone app for Android.

Unlike last year's "Ask for Me," which could only inquire about pricing or services, Gemini can now share personal information the user approves, letting it handle more complex tasks such as checking whether a product is in stock, making a restaurant reservation, rescheduling an appointment, or placing items on hold. Calls are dialed from the user's own number, and users can follow a live transcript and take over at any time.

Google says it is starting at a small scale because "real-world conversations are nuanced." Meta's Muse and similar agents already make calls on users' behalf, and how well this feature actually performs remains to be seen.

At Meta Connect 2026, Meta announced two AI game development tools: Horizon Create, a mobile app, and Horizon Studio, a browser app, which generate 2D or 3D mobile games with progression systems and multiplayer from prompts. Both are in early access with a waitlist.

The distribution is the point: Meta says these games can be recommended on Facebook and Instagram, where a user can tap a clip in the feed and enter a multiplayer session without downloading anything.

The comparison is Roblox, which has 123 million daily active users and already offers prompt-based game creation, while Horizon reported only about 300,000 monthly users in early 2022. Whether these tools bring real players remains unknown.

A Gallup survey of adults in 37 countries finds positive feelings about AI outweigh negative ones in 34 of them. In China, 93% of AI-aware respondents say AI will mostly help their country; in the United States, only 36% do.

The survey is part of a Gallup research program run with Microsoft that will eventually cover 140 countries. Across the 37 countries surveyed, a median 57% of adults have never used AI, while median awareness stands at 81% — optimism runs well ahead of actual use.

Optimism also does not track adoption: in Vietnam, 91% expect AI to help their country, but only 43% have ever used it. Sentiment is not adoption, and the full 140-country results will be worth comparing.

A researcher argues that J-lens, an interpretability tool that translates a model's internal states into words, is polluted by its default choice of the final layer as target.

The author, Kaley Brauer, checked 78 publicly released J-lenses and found 80% target the final layer. In her own tests on DeepSeek-V3, a final-layer target made about half of mid-depth readouts come out as mostly Chinese tokens; switching to the penultimate layer dropped that to 3%.

The cause, she says, is that DeepSeek-V3's last block strongly suppresses Chinese tokens on English text, and that amplified direction comes to dominate the lens. She recommends checking whether the final block injects a dominant direction when fitting or using a J-lens. The result is self-tested, covers only DeepSeek-V3, and awaits independent reproduction.

A research team has released SkillGym, a framework that turns human-written agent skills into verifiable training environments, and open-sourced 2,756 of them.

The environments span 12 task categories and come with 8,364 successful trajectories, usable for supervised fine-tuning and outcome-reward reinforcement learning. The paper's own tests report that fine-tuning lifts Qwen3.5-35B-A3B by 19.10 percentage points on Terminal-Bench 2.1.

The abstract offers no third-party reproduction, and the comparison scores for Claude Sonnet 4.6 and others are values cited by the paper; the environments' quality awaits external use.

Intermediate-layer representations of video and multimodal models can outperform final-layer and model-default outputs, according to the authors' own runs — LAYERSCOPE makes this comparable layer by layer without task labels, across seven architecturally diverse models.

Layerwise evaluation previously depended on task labels, which cost annotation effort and made systematic comparison of layers hard.

The team used local, global, distributional and correspondence-based geometric metrics across seven architecturally diverse models on video and multimodal classification, clustering and text-to-video retrieval tasks from MVEB/MVEB+, and report that no single geometric metric consistently predicts downstream performance; the results are the authors' own runs from a preprint, and whether they generalize to more models awaits third-party testing.

ProCredit lets every turn of agent training be scored by verified progress: the acceptance checks that decide task success are rerun on the intermediate state after each turn, rather than assigning a single outcome reward at the end, and the authors report a 4.1-percentage-point gain over the strongest outcome-reward baseline at 4B on AppWorld.

The known weakness of outcome rewards is that a group of attempts with no success yields no training signal, and failures cannot be told apart by how close they came.

The authors report that on the AppWorld benchmark, starting from Qwen3.5 base models at three scales, the method beats both outcome-reward and progress-based baselines at every scale; a second environment shows the same direction. Ablations indicate that adding final progress to the trajectory score alone does not help — the gain comes from crediting progress to the turn where it occurs. The numbers are the authors' own tests, the abstract gives no figures for the second environment, and independent replication is still pending.