DeepSeek Podcast Summaries
DeepSeek on Yedapo: 9 summarized podcast and YouTube episodes. Each includes key takeaways, core concepts and notable quotes with timestamps.
In search of frontier AI at home
sentdex
Jul 9, 2026
For engineering workflows, running local models like DeepSeek V4 Flash offers superior speed and control compared to hitting external APIs. While high-end hardware like RTX Pro 6000s is expensive, the author demonstrates that you can achieve production-grade results with a human-in-the-loop, bypassing the need for constant, massive model overhead.
Key insight: DeepSeek V4 Flash is so efficient that it outperforms the 4-bit quantized GLM52 on coding benchmarks while delivering significantly faster token speeds, proving that smaller, optimized models are often more practical for real-world software development.
DeepSeek's Deleted Vision Paper Is Nuts...
bycloud
Jul 8, 2026
DeepSeek's new approach solves the 'reference gap' in multimodal models by allowing AI to 'point' at images using bounding boxes and coordinates. By interleaving visual primitives into its chain-of-thought, the model effectively grounds its reasoning in space rather than relying solely on ambiguous language descriptions, significantly outperforming frontier models in topological and counting tasks.
Key insight: For maze navigation, while most frontier models hover around 50% accuracy, this new architecture hits 66.9% by treating visual reasoning like a 'scratchpad' where the model draws points to track progress.
DeepSeek's Absolutely Insane AI Speed Hack
Two Minute Papers
Jul 7, 2026
DeepSeek introduces DSpark, a speculative decoding technique that pairs a high-performance 'senior' AI editor with a 'junior' writer model. By implementing memory, smart verification thresholds, and workload-aware forecasting, DSpark achieves significant speed gains without compromising accuracy. It optimizes inference by predicting which draft tokens are likely to succeed, effectively streamlining resource-heavy GPU processes.
Key insight: DSpark boosts inference speeds by 60% to 85% by predicting which draft tokens are doomed to fail, preventing the senior model from wasting compute cycles on incorrect outputs.
DeepSeek Just Solved AI's Billion Dollar Problem
Two Minute Papers
Jun 22, 2026
Dr. Károly Zsolnai-Féhér reveals that current AI systems suffer from massive underutilization because memory access bottlenecks—not processing power—cripple performance. By implementing a smart traffic-control system that redirects data streams to idle decoding hardware, researchers at DeepSeek have successfully boosted GPU utilization from 40% to 80% for complex agentic workloads.
Key insight: Modern AI systems often waste billions in hardware costs because their 'straw'—the memory bandwidth—is too narrow, leaving GPUs idling while they wait for data to read.
Hermes Agent + DeepSeek V4 = 100X Cheaper
Jack Roberts
May 16, 2026
Jack demonstrates how to combine the Hermes agent with DeepSeek V4 and OpenRouter to create an automated 'Triad' system. By using a conductor, worker, and critic model structure, you can achieve 95% of top-tier AI performance at 1% of the cost, effectively offloading complex decision-making to overnight background processes.
Key insight: DeepSeek V4 provides approximately 95% of the performance of premium models like Claude Opus for roughly 1% of the cost, making it the ideal 'worker' model for large-scale, automated background tasks.
DeepSeekV4 + Claude Code = 100X Cheaper
Jack Roberts
Apr 30, 2026
Learn how to architect a multi-model development workflow that combines Claude's superior design capabilities with the extreme cost-efficiency of DeepSeek V4. By utilizing proxy servers, you can build production-ready applications while slashing API expenses and bypassing standard rate limitations.
Key insight: DeepSeek V4, while less capable in pure creative design, is functionally equivalent to top-tier models for backend logic, algorithmic tasks, and heavy lifting, providing a 100x cost reduction for high-volume development workflows.
GPT 5.5 Arrives, DeepSeek V4 Drops, and the Compute War Intensifies
AI Explained
Apr 24, 2026
New model benchmarks from OpenAI and DeepSeek reveal that performance is increasingly domain-specific rather than universally intelligent. As compute scarcity forces firms to prioritize efficiency over broad reasoning, the industry is shifting toward models optimized for token-per-dollar value in specialized, repetitive white-collar tasks.
Key insight: GPT-5.5 exhibits an 86% hallucination rate on incorrect answers—failing to admit ignorance—compared to just 36% for Claude Opus 4.7, suggesting that high performance scores often mask fundamental reliability failures.
You Are Being Told Contradictory Things About AI
AI Explained
Dec 5, 2025
The AI industry is currently split between two contradictory narratives: the promise of recursive self-improvement and the looming reality of a compute bottleneck. While researchers debate whether scaling current architectures will lead to AGI or eventually peter out, empirical data suggests that exponential compute growth may face a significant slowdown by 2028, forcing a reliance on synthetic data breakthroughs.
Key insight: Anthropic explicitly trains its Claude models using a 'soul document' that instructs the AI to be wary of world takeovers, even those potentially initiated by Anthropic employees or the company itself.
I built a DeepSeek R1 powered VS Code extension…
Beyond Fireship
Jan 28, 2025
You can replace expensive AI subscriptions by running DeepSeek R1 locally via Ollama and integrating it directly into VS Code. This tutorial demonstrates how to create a custom extension that allows you to chat with the model within your editor, providing full data privacy and complete control over your AI-assisted coding workflow.
Key insight: By running DeepSeek R1 locally through Ollama and a custom VS Code extension, you avoid sending your keystrokes and proprietary data to external servers, ensuring complete privacy.