Large language models such as OpenAI's GPT-3.5, GPT-4, and GPT-4o systematically prioritize Western moral values like care and equality while underestimating values such as purity that are important i…
Fine-tuning large language models offers 10-30% accuracy improvements for specific domains, 90% cheaper inference than GPT-4, and full data privacy. The guide covers when to fine-tune, common mistakes…
A developer investigated large language models with 1 billion parameters, finding that while they offer lower latency and reduced energy consumption compared to larger models, they face challenges in …
George Mack discusses the most important questions of our time, including whether an LLM like GPT-4 or Claude could win a literary prize, and explores concepts such as femmephobia, hypomnemata, and th…
WordPiece, Unigram, and SentencePiece are three subword tokenization algorithms that differ in how they split text, with WordPiece using a likelihood-maximizing merge criterion and a ## continuation m…
Gorilla, a large language model fine-tuned from LLaMA-7B on machine-generated instruction-API pairs, outperforms GPT-4 on the APIBench benchmark while reducing hallucinated API calls, according to Pat…
Wang et al. published Voyager, an open-ended embodied agent that uses GPT-4 to propose tasks, write executable code against a game API, verify it, and archive verified programs into a growing skill li…
AI researchers are increasingly studying how babies learn, believing that infant brain architecture could hold the blueprint for the next phase of artificial intelligence. Unlike current AI models lik…
A Harvard Business Review report found that workers toggle between applications 1,200 times daily, losing nearly 4 hours per week to reorientation, costing roughly $9,000 annually per knowledge worker…
A solo developer spent $47 and four weekends testing AI APIs from six providers to determine the best options for small operators versus enterprise teams. The key finding is that going direct to a pro…
OpenAI outlined a five-part framework for managing AI investments in the agentic era, urging enterprise leaders to focus on “useful work per dollar” rather than token price. The company’s latest model…
A developer created an AI-assisted Architecture Decision Record (ADR) template that uses large language models to generate documentation in 3-5 minutes instead of 30-40. The template covers ADR struct…
Cactus Needle, a 26-million-parameter open-weight language model developed by the Cactus project, is designed exclusively for function calling (tool use), outperforming much larger general-purpose mod…
A developer argues that as AI becomes capable of generating code, marketing strategies, and even legal advice, the value of human knowledge shifts from memorization to critical thinking, creativity, a…
TormentNexus' AI skill registry has surpassed 5,776 reusable modules, with top categories including code review, Terraform generation, and database migrations. The code review skill family alone compr…
A new study published in PNAS found that AI language models are far more susceptible to subtle nudges in choice architecture than humans, overreacting to default options and highlighted choices. Resea…
OpenAI launched GPT-Live-1 on July 8, a full-duplex voice model that listens and speaks simultaneously, using a two-layer architecture with a continuous interaction layer and a delegation layer that h…
A Mount Sinai research team led by Mahmud Omar and Eyal Klang found that OpenAI's GPT-5 amplifies sociodemographic biases in clinical decision-making, assigning higher urgency and less advanced testin…
A developer built a production AI scoring pipeline that processes over 10,000 job listings daily using GPT-4. The system uses function calling to enforce structured output, a pre-filtering stage to el…
Researchers have identified a 'Vestigial Heuristic' in large language models, a bias against token repetition that leaves a unique signature in AI-generated text. A new metric, Telescope Perplexity, l…