Sakhanda Wire
NVDA $150.96 -2.89% MSFT $508.29 +0.40% GOOGL $282.44 +0.80% META $616.38 -0.40% AMZN $267.74 +3.82%

Latest news

Older stories

MIT Tech Review

Schools test structured ways to use AI in the classroom

Generative AI has added pressure to teachers, who must adapt to tools that can answer homework questions and produce essays quickly. Although organizations including OpenAI and UNESCO support classroom use, many educators remain uncertain about appropriate boundaries and worry about errors, bias, privacy, and the quality of AI-generated materials.

MarkTechPost

Generalist AI Releases GEN-1.5 Robot Model for One-Shot Task Learning

Generalist AI has released GEN-1.5, a multimodal robot foundation model that can learn a physical manipulation task from one 3–12 second sensorimotor demonstration. The example is placed in a 30-second context window, after which the robot attempts the task without gradient updates, fine-tuning or task-specific programming. The model processes video, sensor, language and proprioceptive inputs and produces action trajectories at 100 Hz.

MarkTechPost

Google Research team presents mobility-aware POI embeddings

A team from Google Research and USC has introduced Mobility-Embedded POIs, or ME-POIs, a framework designed to add aggregate movement patterns to text-based representations of points of interest. While text can describe a venue’s category and location, the researchers say it does not capture how visitors use it, such as short commuter stops versus longer stays.

TechCrunch

Hugging Face reportedly receives acquisition offers valuing it at $13B

Hugging Face has reportedly been approached with acquisition offers valuing the company at $13 billion or more, according to Business Insider. No agreement has been reached, and the potential buyers have not been identified. The startup is said to be speaking with banks to assess the bids.

The Decoder

Thomson Reuters develops in-house legal AI model with $40 million investment

Thomson Reuters has developed Thomson, its first in-house language model for legal work, using Alibaba’s open Qwen as its foundation. The company says it spent about $40 million on staff and computing over more than two years. The amount excludes the value of content from Westlaw, Practical Law, Checkpoint and Reuters, as well as contributions from hundreds of domain experts. The current foundation is identified as Qwen3.5-397B.

The Decoder

Cerebras introduces CS-4 AI accelerator with doubled performance

Cerebras has introduced the CS-4, a rack-scale AI accelerator that CEO Andrew Feldman describes as the fastest system in the industry. The product is delivered as a complete server cabinet for data centers, including computing hardware, power systems and cooling equipment.

The Decoder

AI chatbots often direct pregnancy queries to anti-abortion groups

An AlgorithmWatch investigation found that ChatGPT, Gemini, Grok and Claude frequently referred people asking about unplanned pregnancies to anti-abortion organizations without identifying their ideological position. At least one in four queries contained a link to such a website. Researchers examined 270 responses produced for three fictional users in English, German and Italian, using ChatGPT-5, Gemini 3, Grok 4.3 and Claude Sonnet 4.8.

MIT Tech Review

Children Learn Language with Far Less Data Than AI Models

Human children acquire their native language from a small fraction of the material used to train today’s large language models. Toddlers generally begin producing grammatical sentences after hearing roughly 10 million words, with estimates reaching about 30 million at the high end. By contrast, modern models process vastly larger collections, including trillions of tokens during pretraining. This difference is known as the data efficiency gap.

The Decoder

Nvidia discusses investment in Perplexity at valuation above $30 billion

Nvidia is negotiating a possible investment in Perplexity at a valuation exceeding $30 billion, according to The Information. That would be more than 50 percent above the AI search company’s valuation in its previous funding round a year earlier. The size of the potential investment has not been disclosed.

MarkTechPost

Python Tutorial Recreates a LabPlot-Style Scientific Analysis Workflow

A tutorial presents a Python workflow modeled on LabPlot’s terminology and project structure. It recreates the aspect tree, analysis kernels, plotting system and project model, with reusable components for importing tabular data and organizing columns, spreadsheets and projects.

TechCrunch

Ox Alpha’s creator remains unidentified amid online speculation

A new AI model called Ox Alpha has sparked online debate over the identity of its developer. The free model appeared on OpenRouter on Thursday, described as a reasoning system for coding, extended agentic work and production workloads. The listing calls it a “stealth model” and says it was developed and operated by an anonymous third-party provider during the preview period.