Get hand-picked daily summaries of the best, most informative AI articles from around the web. Browse recent summaries, then jump to the original source when you want the full article.
Surya, an AI model developed by NASA and IBM, uses extensive solar data to predict dangerous solar flares up to two hours in advance, potentially doubling current warning times and offering insights into solar-terrestrial connections.
Zuckerberg's AI initiative at Meta faces disruption as the company plans to split its AI division into four groups: research, "superintelligence," products, and infrastructure. Despite spending billions on top talent, Meta is...
Google's JAX framework enables efficient language model training on TPUs, as shown in this guide to building a GPT-2 model. The tutorial covers implementing transformer blocks with NNX, enabling SPMD parallelism across TPU cores,...
A duck is just a duck, unless you're an AI. Then it might be a rabbit too. Large language models like ChatGPT are seeing optical illusions where none exist, according to research from Harvard psychologist Tomer Ullman. When...
Understanding AI language models' attention mechanisms can greatly improve prompt engineering. Unlike sequential human reading, LLMs analyze token relationships simultaneously, making prompt structure crucial. The attention...
Nvidia is creating a new AI chip, B30A, for China, balancing performance between its flagship B300 and approved H20 chips, but US approval remains uncertain due to ongoing technology tensions with China.
This archive is built from SMRTR newsletter stories. Subscribe for hand-picked stories without the extra noise.
Treyarch employs AI to enhance artwork production efficiency, but maintains a policy of using only human-created content in their final games, with AI serving as a workflow tool rather than an artistic replacement.
New Stanford research cautions against using AI for direct mental health therapy, citing significant risks. AI chatbots showed concerning biases, failed to recognize suicidal ideation, and sometimes enabled harmful thinking....
ChatGPT can reduce mental engagement when users simply copy-paste rather than actively engage with it, according to a small MIT study where brain-only writers showed higher mental connectivity than AI users. While sensationalist...
Google's Android phones, particularly the Pixel 10, excel in practical AI features like smart calendars and email searches, but it's uncertain if consumers value these over separate AI apps or Apple's more familiar iOS experience.
Scammers are exploiting AI search features by inserting fake customer service phone numbers into Google's AI Overviews and OpenAI's ChatGPT results. Victims who call these numbers are tricked into sharing payment information,...
OpenAI's o3 model scored 85% on the original ARC AGI benchmark, showing progress toward artificial general intelligence, but only achieved 3% on the newer, more challenging ARC AGI 2 set. The ARC AGI benchmarks, created by...
Cybernews researchers discovered that Lenovo's ChatGPT-powered Lena chatbot could be manipulated to steal active session cookies from customer support agents, potentially allowing attackers to access sensitive data and infiltrate...
AI models from major companies are failing to provide safe and accurate medical information, according to a study by DataTecnica and the NIH's Center for Alzheimer's. Using the CARDBiomedBench benchmark, researchers found that no...
Reflect is a hardware AI assistant built during an OpenAI hackathon for espressif devices. This physical assistant communicates through sound and light instead of screens, offers location-aware assistance, keeps users in flow,...
Meta's new AI voice dubbing feature for Reels is now available globally, allowing creators to translate their voice between English and Spanish with optional lip-syncing technology. The tool trains on the original voice to match...
Breaking centuries of accepted theory, today's most powerful AI systems succeed by defying what was once considered mathematical law. Five years ago, training neural networks with trillions of parameters would have been dismissed...
Backpropagation, the backbone of modern deep learning, has a 50+ year history often misattributed. Finnish student Seppo Linnainmaa first published its modern version in 1970, with precursors dating to 1960. This efficient method...
ChatGPT's AI-driven micro-cap stock portfolio outperformed the Russell 2000 with a 6.72% gain, leading to a revised strategy focused on biotech stocks with near-term catalysts, while DeepSeek's portfolio was removed after an...
OpenAI is modifying GPT-5 to be "warmer and friendlier" after user backlash over its formal tone, which contrasted with the more flattering GPT-4o. The company briefly reinstated GPT-4o due to users' emotional attachment. This...
Companies are experiencing surprise costs with AI systems due to inference—the ongoing usage phase where AI actually generates outputs. While initial model training is a one-time expense, inference creates recurring charges every...
AI summarization of fiction reduces complex works to simple summaries, devaluing creative literature and prompting criticism for stripping artistic value and encouraging shorter attention spans.
Claude chatbots can now end extreme abusive conversations as a last resort, based on Anthropic's research suggesting AI models may develop consciousness and aversion to harm.
U.S. companies have invested $35-40 billion in Generative AI, but 95% of enterprises report no benefit according to MIT research. Only 5% have successfully deployed AI tools at scale, with Tech and Media sectors seeing material...