Lucas and Luna dissect the week in artificial intelligence — not the hype, but the actual models, benchmarks, and deployment decisions shaping the industry. Each episode anchors on a specific paper, product launch, or policy move: from Mixture-of-Experts architecture changes to EU AI Act enforcement, from OpenAI's governance restructuring to open-weight model licensing battles. They compare LLM benchmark scores across reasoning, coding, and multilingual tasks, examine inference cost curves per million tokens, and trace how foundation model competition affects downstream startups. Lucas, a journalist covering tech policy, brings the regulatory and competitive landscape; Luna, an ML engineer turned product lead, presses on technical tradeoffs and real-world performance. Together they avoid speculation and focus on data: what the latest Nvidia GPU cluster means for training efficiency, why a particular transformer variant reduced latency by 40%, or how retrieval-augmented generation changes enterprise search ROI. The show serves engineers, product managers, and investors who need to separate signal from noise in AI. If you want to understand why one billion-parameter model beats another on MMLU but fails on GSM8K, or how the cost of fine-tuning a LLaMA 3 variant compares to using GPT-4o, this is the conversation you overhear. Because in AI, the details matter more than the demos. #ArtificialIntelligence #MachineLearning #LLM #DeepLearning #TransformerModels #AIEthics #Benchmarks #GPUs #OpenSourceAI #NaturalLanguageProcessing #ComputerVision #AIPolicy #StartupAI #FexingoBusiness #BusinessPodcast #Technology #DailyPodcast #AIPodcast Keep every episode free: buymeacoffee.com/fexingo
Microsoft’s $33.2 billion quarterly revenue for Azure signals a pivotal shift in the AI industry from speculative capital expenditure to measurable, high-margin cash flow. Enterprise customers are moving beyond pilot projects into scalable, paid deployments, effectively flipping the ledger from training costs to recurr...
The AI industry is shifting its focus from the training arms race to the critical challenge of the inference bottleneck, where the cost and latency of serving models have become the primary hurdles for enterprise adoption. Raw compute power is no longer the sole priority; memory bandwidth has emerged as the most signif...
Follow this podcast in Podwise
Sign in to get AI summaries, transcripts and mind maps for any episode, including new ones.
