GoKawiil
Tech News
Rosetta Stone, the 34-year-old language-learning company, is overhauling its core product with a mobile-first design, AI-powered conversation practice, and more customizable lessons. The update marks a shift from its earlier desktop-oriented approach toward smartphone-centered learning.
A programmer spent a summer at Recurse Center, a Brooklyn-based self-directed coding retreat, participating in peer-led study groups. Activities included an 'Agentic Adventures' group exploring LLM agents and sandboxed coding environments, a 'Practical Deep Learning' group working through Jeremy Howard's book, and a 'Math Monday' discussion group started with another attendee. Projects included building a remote sandbox for a coding agent, training a small text-prediction model on Romeo and Juliet, and experimenting with LoRA fine-tuning.
An opinion article contends that traditional education—built around specialization, one lifelong career, and four-year degrees—no longer fits how people actually work today. The author argues the notion of a single preordained career path, embodied by the question 'what do you want to be when you grow up?', is outdated and should be discarded.
Christian Kästner, who teaches Machine Learning in Production at Carnegie Mellon, says he has overhauled most assessments in his course because AI agents can now complete take-home assignments convincingly. Instead of grading homework, he now focuses evaluation on TA interactions, exams, and video demonstrations, while keeping the course's core learning goals on engineering tradeoffs and teamwork largely unchanged. Students are permitted to use AI tools freely except during written and oral exams.
Apple's AI research team has published LensVLM-9B, a 9-billion-parameter vision language model that processes long documents by first compressing text into image form and then selectively decompressing only the pages it judges relevant, using learned tools. The model, code and demo scripts have been released on GitHub and Hugging Face under Apple's Machine Learning Research Model License and Sample Code License, with configurable compression ratios of 5x, 10x, or 15x.
OpenAI now lists 27 open robotics positions on its careers page, more than double the 11 postings seen in May, spanning hardware, software, data and prototyping work. Base salaries range from $177,000 to $500,000 before equity and bonuses, with the top-paying role focused on building infrastructure to process robotics training data at scale.
A developer building the habit-tracker app Main Character reported reaching 1,000 consecutive days of daily math practice on Math Academy, progressing from Math Foundations through Mathematics for Machine Learning. They disclosed that relying on notes and answer-checking during difficult lessons masked underlying weaknesses, prompting them to reset their course progress and restart.
A survey by academic mentoring platform Nova Learning found that the share of middle and high school students naming Computer Science and AI as their top academic interest dropped from 42.3% in 2025 to 27.9% in 2026, while Engineering's share nearly doubled, rising from 11.9% to 23%, led largely by Mechanical and Aerospace Engineering. Separately, the National Student Clearinghouse Research Center reported an 8.1% drop in undergraduate Computer and Information Science enrollment at four-year institutions, to about 606,000 students, even as Engineering enrollment grew.
xAI released Grok 4.7, a new model built on a larger base and trained with a longer reinforcement learning run focused on lengthy, difficult tasks. The model improves self-verification and long-context handling, and is natively tuned for the Grok Bot harness to boost conversational and knowledge-work performance. It ships at the same pricing and speed as Grok 4.6.
A commentary argues that most enterprise AI deployments today only store and retrieve information rather than genuinely learn from experience over time. It draws a comparison to an employee who interacts with customers and attends meetings for a year but retains none of the skills or judgment gained, remaining as capable as on their first day. The piece calls for AI systems that actually improve and adapt with use, rather than simply accumulating data.
A hobbyist project called Mini-AGI has been shared on Hacker News as a byte-level language model that builds its own architecture and trains from scratch on a single consumer GPU with 8GB of VRAM. Rather than keeping all parameters in memory, it stores weights as disk files and loads them as needed, letting it expand or prune capacity while it continuously reads and learns from a text corpus without forgetting prior knowledge. The creator says it's currently a toy-scale demonstration, not a competitive model, and weights won't be released until the current training pass over its corpus finishes in a few weeks.
Researchers introduce RetroChimera, a retrosynthesis model that combines two neural architectures with different inductive biases through a learned ensembling method. It outperforms existing baseline models across varying data scales and generalizes well even with limited examples per reaction type. In blind evaluations, professional chemists preferred RetroChimera's suggested synthesis routes over both published reference reactions and predictions from other AI systems.