News from the AI industry and academia; AI-related updates such as new papers, new models, or new services from big tech companies including OpenAI, Google, DeepMind, Meta, and Amazon; news on big deals between AI companies (acquisitions, partnerships); news about large-scale foundation models; and news about major talent acquisitions by big tech companies.
An MIT experiment showed students using ChatGPT for essays had lower brain activity, creativity, and working memory than those writing unaided or using Google Search.
AI-generated essays converged on similar words and ideas, leading to uniform responses across different users and prompts.
A Cornell study found AI suggestions made American and Indian writers’ texts more alike and skewed toward Western norms.
+3 more insights
XSLT runs in browsers to convert XML data to HTML pages.
Using XML and XSLT avoids complex JavaScript frameworks and tools.
Browsers natively support XSLT, requiring no extra build setup.
+2 more insights
Two distinct periods of rapid molecular and microbial change occur around ages 44 and 60 in adults.
About 81% of the 135,000 + molecules and microbes studied exhibit non-linear fluctuations in abundance over time.
Mid-40s shifts affect molecules linked to alcohol, caffeine and lipid metabolism, cardiovascular health, skin and muscle.
+2 more insights
The proposal suggests adding a built-in declarative templating API to the web platform.
Templating offers better ergonomics, security against XSS, performance, static analysis, and server-side rendering.
Developers now rely on external libraries, adding download size and setup complexity.
+2 more insights
Matrix release v1.15 adds industry-standard OpenID Connect authentication, merging MSC3861 and related proposals into the spec.
New room summary endpoint (MSC3266) enables clients to fetch richer information about unjoined rooms for invites, knocks, and matrix.to links.
Rich topic support (MSC3765) allows bold text and lists in room topics via the m.topic content block, with fallbacks to avoid new room version requirements.
+2 more insights
Provides a conversational prompt builder to define business type, schema, and row count for dataset generation
Uses OpenAI to generate a detailed data specification, then Faker to create data locally
Offers real-time preview of 10-row sample with a small OpenAI cost (~$0.05)
+3 more insights
AI systems degrade empathy and social relations rather than enhance human qualities.
AI training data reflects and amplifies biases, leading to unfair outcomes.
Automation bias and deskilling reduce workers’ expertise and job security.
+2 more insights
Black Forest Labs released FLUX.1 Kontext [dev], a 12B parameter open-weight image editing model matching proprietary performance on consumer hardware.
FLUX.1 Kontext [dev] is available under a non-commercial license with support for ComfyUI, HuggingFace Diffusers, TensorRT, and optimized NVIDIA Blackwell TensorRT variants (BF16, FP8, FP4).
Human preference tests on KontextBench and independent evaluations show FLUX.1 Kontext [dev] outperforms existing open and closed image editing models.
+2 more insights
U.S. GDP fell 0.5% annualized in Q1, worse than prior estimate of ‑0.2%
First quarterly contraction in three years, reversing 2.4% growth in Q4 2024
Imports surged 37.9%, cutting GDP by 4.7 percentage points amid tariff-driven frontloading
+4 more insights
Google has fully released Gemma 3n, a mobile-first AI model optimized for on-device multimodal tasks.
Gemma 3n natively handles image, audio, video, and text inputs with text outputs in two sizes: E2B and E4B.
The model uses MatFormer architecture for nested model sizes and Per-Layer Embeddings to run large parameter counts in limited memory.
+3 more insights