Shares
Curated links, videos, and posts worth sharing — with personal commentary
Peter Boghossian (@peterboghossian) on X
Academia's silence on unreplicable fields isn't oversight; it's cowardly complicity. We need to prioritize Socratic reasoning and mental models — they’re far more universally valuable than niche technical skills.
Sharbel (@sharbel) on X
Cybersecurity is lagging and will become crucial as AI evolves. Recent findings on prompt injection during agentic browsing underscore this shift. Investors may be skittish, but solutions like whitelists and trust scores could help navigate these emerging threats.
Project Glasswing: Securing critical software for the AI era
Software is evolving at lightning speed, and staying relevant requires deep thinking and adaptability. I'm eager to dive into Anthropic's Mythos model and tackle the challenges of this fast-changing landscape head-on!
hoeem (@hooeem) on X
Best read on Claude Cowork—it's evolving fast. Need to dive deeper into cowork for automation gains. Upfront investment pays off long-term. Emphasizing modularity can simplify processes, making them easier to understand and adapt.
Browser Use (@browser_use) on X
Excited to dive into auto-research after seeing its innovations! Love the simplicity of the approach and plan to tackle a problem with it next week.
Andrej Karpathy (@karpathy) on X
Andrej Karpathy's posts resonate because they articulate familiar thoughts and spark new ideas. In an age of LLMs, we need anti-echo chamber strategies to challenge our beliefs, making his insights both simple and invaluable.
The End of Coding: Andrej Karpathy on Agents, AutoResearch, and the Loopy Era of AI
Engaging listen with Andrej Karpathy on AI's future. Maximizing these tools is crucial—users could see 10X results. It’s about effort and continuous learning, but the journey is just starting.
Muhammad Ayan (@socialwithaayan) on X
Excited to test Context Hub! It aims to fix AI coding agents' frustrating issues with outdated API parameters. If it works as promised, it could adapt to constant model changes and expand API access significantly. Let’s see how it goes!
Anthropic just 5x'd Claude Code's context to 1M. For my workflow, that's closer to 10x usable context. Let me explain. I run automated software engineering sprints with custom slash commands and...
Anthropic's 5x context boost to 1M tokens transforms my software sprints, turning 87k usable tokens into 887k. This reduces overhead and compaction issues drastically, shifting the focus on permanent context over transient skills. Excited to test this in practice!
Chayenne Zhao (@GenAI_is_real) on X
Generative AI isn't quite ready to build cohesive systems without human oversight. Current codebases risk accumulating technical debt due to quality control gaps. While future improvements may address this, we must be cautious of overwhelming complexity in the meantime.