Tech news picked on 2026-10-06
Pinterest AI Turns Beauty Pins Into Salon Action Plans
- Pinterest offers a feature called 'Beauty Guide' that transforms ideas from beauty-related Pins like hair and nails into professional terms and specific plans used by stylists.
- For example, it explains beauty industry terms such as 'balayage', 'root melt', and 'almond nails' in a way users can easily understand, while also providing information on treatment time, price ranges, and maintenance methods.
- This guidance can be accessed through the 'Get Guide' button below Pins within the app and currently applies only to Pins related to hairstyles, hair colors, nail shapes, and finishes.
- Users can get treatment details, estimated time, and price information based on their saved beauty Pins to refer to before showing them to others or visiting a salon.
Mistral Large 4 is a 49 billion parameter AI model that can handle text and images together
- Mistral Large 4 is a multimodal AI model with 49 billion active parameters and a total of 1.05 trillion parameters (source: official Mistral AI documentation).
- It includes a visual encoder consisting of 160 million parameters, enabling it to process not only text but also image inputs.
- The API supports various interactive functions including chat, document Q&A, function calls, and batch processing.
- Usage costs are charged based on the number of tokens, with approximately 0.68 dollars billed for processing 1 million input tokens (according to official pricing).
Pixel Watch 5 Eases Initial Price Burden with Amazon Discount
- Google's smartwatch Pixel Watch 5, released in August, is being sold on Amazon with a discount of 60 dollars.
- The 41mm model is available for 334.99 dollars, and the 45mm model for 364.99 dollars, which partially offsets the approximate 50 dollar increase compared to the initial launch price of the Pixel Watch 4.
- The Pixel Watch 5 features improved GPS performance, gesture controls, and an offline Gemini feature, but Google's AI Assistant proactive notification feature received mixed user reactions at launch.
- Google continues to expand functionality through software updates, with some health features still not yet activated.
Menlo Ventures Makes Large Investment in AI Coding Startup Factory
- Menlo Ventures recently made an investment in AI coding startup Factory, which was valued at $5 billion.
- Matan Grinberg, CEO of Factory, engaged in a public dispute with former advisor Chris Degnan who moved to competitor Cognition, raising concerns that Degnan may have leaked confidential information.
- Vinod Khosla criticized the state of Factory, but Menlo Ventures publicly supported Factory's founders, technology, and customer relationships.
- Matt Murphy, a partner at Menlo Ventures who previously invested successfully in Anthropic, views this investment as recognizing Factory as a promising next-generation company.
AI Chip Developer Etched Receives Investment Proposals Valuing the Company Over $40 billion
- Etched is a startup that develops AI hardware systems using its own designed AI chips. It previously received an investment of $700 million at a valuation of $21 billion in September, and is currently reviewing investment proposals ranging from $40 billion to $50 billion. Negotiations are at an early stage and terms may change.
- The company claims its chips process tokens faster and cheaper than Nvidia chips. Co-founder and COO Robert Wachen explained that to increase inference speed, they redesigned 2 core components from scratch.
- Quant trading firm Jane Street was the lead investor in the last $700 million funding round and an early customer of the system. Etched operates a 10 megawatt-scale data center in Silicon Valley and has established facilities in Taiwan near TSMC to coordinate production.
- According to the Wall Street Journal, about 15% out of approximately 400 employees are former Nvidia staff. A source said that raising funds similar to the previous round would secure enough capital to last up to 3.5 years.
- The 2026 Nobel Prize in Physics was awarded to Francis Halzen, who made significant contributions to the development of the IceCube Neutrino Observatory and the discovery of high-energy neutrinos originating from space.
- The IceCube Neutrino Observatory is a particle detection facility located in Antarctica, designed to detect the rare neutrinos arriving at Earth from space.
- The discovery of high-energy cosmic neutrinos provides a new observational method to directly study cosmic phenomena and represents a major advance in astrophysics and space research.
How to Pretrain Transformers Without Backpropagation Using Token-wise Activation Noise
- Dust is a new algorithm that applies random perturbations to the neural activation values of each token and estimates gradients by measuring and averaging how much these perturbations reduce loss on a per-token basis.
- Traditional evolution strategies required a separate forward pass for each weight, but Dust independently applies perturbations at the token level, evaluating thousands of perturbations in a single forward pass, greatly improving computational efficiency.
- Experiments by the Q Labs research team showed that Dust improved efficiency by 1000 to 10,000 times compared to the existing weight-space evolution strategy EGGROLL when training on over 1 million tokens; this efficiency gain became more pronounced as model size increased.
- They also confirmed that gradients estimated by evolution strategies matched backpropagation results well up to scales of 1 billion tokens; however, currently, the computational efficiency is not yet sufficient to fully replace backpropagation.
Route finder service for the flattest paths in San Francisco
- This web service analyzes about 160,000 road segments directly in the browser when users enter their start and end points within San Francisco, providing the route with the lowest slope.
- Elevation data is calculated using LiDAR data with 1 meter resolution provided by the US Geological Survey (USGS) and road information from Overture/OpenStreetMap.
- A slider is provided to adjust the climbing elevation relative to distance (cumulative ascent), allowing balanced selection between the shortest route and nearly flat paths.
- Pedestrian routes may include stairs, but bicycle routes exclude stairs, and routes are not provided if the slope exceeds the set threshold.
Remote-Controlled Electric Wagon is Useful on Slopes and Sandy Paths
- The DuneRover is a remote-controlled electric wagon that can stably move on inclined and sandy paths while carrying loads up to 90 kg.
- Each wheel is equipped with a brushless motor capable of 400 W output and 33 Nm torque, allowing automatic speed adjustments in sync with the user's walking pace.
- The battery, with a capacity of 88.8 Wh, operates for more than 8 km and features a USB-C port as well as solar charging capability. Although the prototype is not waterproof, the production model aims to meet IPX6 waterproof standards.
- The pre-produced prototype has LED lights always on by default, exhibits wheel resistance when the motor is not operating, and some motor shaking during certain runs was observed. According to manufacturer BougeRV, these issues are expected to be resolved in the mass-produced version.
How to Reactivate iChat Video Calling on the Old Mac OS X
- iChat AV was a video and audio calling program provided by Apple starting in 2003, used in conjunction with the iSight camera.
- iChat communicated with the SNATMAP server, operated by AT&T, to verify public IP and port for video call connections; however, the server was discontinued, causing call connections to fail.
- The author retraced iChat's network protocol using LLM analysis to recreate the SNATMAP protocol specification and developed a server program with the same functionality.
- By modifying the request route in the /etc/hosts file to point to a specific IP, video and audio calls via iChat on OS X Leopard or Snow Leopard become possible, and standard USB webcams can be used instead of iSight.
Classic Games Are Enjoyed Anew on the Web and in Virtual Reality
- Classic games are being moved from their original code and devices to various new environments such as web browsers and virtual reality devices.
- For example, you can play the original 'Halo' with multiplayer features on a web browser, or even skateboard in 'Call of Duty'.
- These unofficial projects exist in a legal gray area, and some companies like Nintendo view unauthorized use of their games on unapproved devices negatively.
- Many projects are developed using artificial intelligence (AI), which raises ethical and environmental controversies.
- Beam, developed by Reflection, is a sparse mixture-of-experts (MoE) model with 501 billion parameters, of which 23 billion are activated, designed specifically for coding, reasoning, and agent tasks.
- The model was pretrained on 23.8 trillion high-quality web and licensed data tokens and trained using over 100 million reinforcement learning rollouts by running 10,500 NVIDIA GB300 GPUs continuously for 4 weeks.
- Reflection stated that Beam shows competitive performance on coding and agent-related tasks compared to public models of the same or larger scale like GLM 5.2 and Qwen 3.8-Max, while achieving greatly improved efficiency by using 3 to 4 times less GPU compute during inference.
- The training employed a newly optimized asynchronous policy gradient algorithm for reinforcement learning and utilized about 1 million task environments. It also applied various techniques for balanced expert utilization and stable signal propagation, ensuring overall training stability.
OpenAI Introduces Detectable Watermark in ChatGPT Texts Within the EU
- OpenAI has applied an invisible watermark to texts generated by ChatGPT and Codex within the EU to enable detection of AI-generated content.
- The watermark introduces subtle changes in word choices that are not noticeable to humans but form patterns recognizable by dedicated detectors.
- The watermark is embedded within the text and remains intact even when copied; OpenAI states it does not include user information nor affect model performance.
- According to a technical paper released by OpenAI researchers in collaboration with the University of Pennsylvania and Yale University, this method has limitations: detection rates drop for short or translated texts and math answers, and also decrease when some words are replaced.
- Open source has been based on collaboration and communication among developers, beyond just sharing code, but the increase in AI-generated automated contributions is weakening these relationships.
- There are many incorrect or meaningless contributions generated by AI, causing maintainers to spend significant time reviewing them, leading to a burden on resources.
- Several open source projects and platforms have introduced policies restricting or banning external pull requests due to the rise in AI-generated contributions. For example, in 2026 October, multiple major projects limited AI-related contributions.
- As a result, open source participation is becoming more limited and centered around trustworthy small groups, reducing the previously open environment where anyone could easily contribute.
Google AI 'Gemini Call for Me' Preparing to Expand Simple Phone Message Automation
- Google is planning to expand its AI feature 'Gemini Call for Me' from business calls to simple personal call automation between family and friends.
- This feature works by having the AI deliver simple messages the user wants over the phone, for example, handling requests like "Tell my mother I'm 15 minutes late."
- During APK analysis, Android Authority discovered detailed app-specific permission settings for the Gemini Call feature, suggesting users may be able to choose whether to allow automatic call reception.
- However, it is not yet confirmed if Google will officially release this feature and permission settings. The feature does not support long conversation automation and is limited as a replacement for text messages.
Lucid Motors EV production falls to lowest level in 2 years
- Lucid Motors produced 2,954 electric vehicles in Q3 of 2026, down 54% from the same quarter last year. This was an intentional result of limiting production to match demand.
- The company's production has declined for consecutive Q3 recently, and production in this Q3 is the lowest since Q1 of 2025.
- New CEO Silvio Napoli is pursuing cost savings of about $1.4 billion through restructuring, workforce reduction, and the elimination of 2-shift operations at manufacturing plants, while delaying the launch of the low-priced EV 'Cosmos', priced below $50,000.
- Rival Rivian shipped about 20,000 vehicles based on the low-priced SUV R2, the first fully produced in Q3, posting record quarterly results. Lucid had targeted 90,000 deliveries for 2024 when it completed its SPAC merger in 2021, but is well behind.
Cases of ChatGPT Falsely Using New Yorker Cartoonists' Signatures to Create AI Cartoons
- ChatGPT is forging AI-generated cartoons by attaching the real signatures of New Yorker cartoonists, making them appear as authentic originals.
- Cartoonists feel this signature forgery infringes on their copyrights and moral rights, expressing strong objection.
- New Yorker cartoonists have maintained their cartoon copyrights and have not granted separate permission for AI training; the publisher Condé Nast also did not authorize agreements to allow AI training on these signatures.
- OpenAI has recognized the issue and begun relevant measures, but as of 2026 year October, some AI-generated cartoons still bear actual cartoonist signatures.
New Mac Productivity Apps Supporting Complex Tasks and Collaboration with AI Tools
- Gemini 4 Argon is an AI model developed by Google focusing on systematic reasoning and handling complex tasks in engineering, legal, and financial fields.
- ChatGPT Space is a collaborative platform where multiple users can work together within a shared space to create content and complete tasks using AI.
- MCP-related tools like CoreSpeed, Octri.dev, and Rival Workshop focus on infrastructure functions that enable AI agents to be utilized in actual work environments.
- Mac tools such as Smooth Recorder offer screen recording and automatic editing features, while Calnio supports two-way synchronization between Notion databases and Apple Calendar.
NVIDIA Shield TV Pro price increased by 100 dollars due to rising component costs
- The price of NVIDIA's streaming box, the Shield TV Pro, has risen by 100 dollars from 199.99 dollars to 299.99 dollars due to increased component prices.
- The initial model was released 7 years ago and features a Tegra X1+ processor, 3GB RAM, and 16GB of storage.
- NVIDIA has stopped producing the entry-level Shield TV (2GB RAM, 8GB storage) to focus solely on the Pro model.
- The price increase is mainly due to global memory shortages and increased demand for NVIDIA GPUs and AI chips.
OpenAI Introduces Visual Ads Next to ChatGPT Image Generation Results for US Users
- OpenAI is piloting a feature that displays visual ads next to image generation results in ChatGPT, initially in the United States.
- The ads consist of selected advertisers' products and services, are clearly marked, and do not influence ChatGPT's response content.
- This change aims to support free and low-cost subscription users, and the impact of ads on user experience is also expected.
- OpenAI is enhancing ad measurement tools and collaborating with multiple ad tech partners for click tracking and location-based experiments.
Borland Turbo Basic supports compiling BASIC programs without memory size limits on DOS
- Borland Turbo Basic is an interactive compiler for MS-DOS that took over BASIC/Z created by Robert Zale in 1987.
- Unlike the existing GWBASIC, which only executed .BAS interpreter code with a size limit of 64KB, Turbo Basic could create large programs without executable size limits within memory capacity.
- Along with the compiler, Turbo Basic supported an integrated development environment (IDE) where code writing, editing, compiling, and running could be done in one place, and it also generated .OBJ object files during compilation.
- In 1987, a separate 'Turbo Toolbox' set was released to assist with database, text editing, and communication programming, expanding the capabilities of the base language.