Hacker News: sebastianperezr

AI Agents Show Long Context Gains with LCM, Specialized Applications Emerge

sebastianperezr — Mon, 11 May 2026 13:02:39 +0000

Article URL: https://presciente.com/edition/75

Comments URL: https://news.ycombinator.com/item?id=48094438

Points: 2

# Comments: 0

LLM Inference Throughput Rises 4.5x with Parallel Verification

sebastianperezr — Sun, 10 May 2026 13:00:15 +0000

Article URL: https://presciente.com/edition/74

Comments URL: https://news.ycombinator.com/item?id=48083638

Points: 2

# Comments: 0

Anthropic NLAs translate LLM activations to human-readable text for safety

sebastianperezr — Sat, 09 May 2026 14:54:06 +0000

Article URL: https://presciente.com/edition/78

Comments URL: https://news.ycombinator.com/item?id=48075461

Points: 1

# Comments: 0

Canvas Data Breach; DeepSeek V4 Flash Boosts LLM Inference 4.3x

sebastianperezr — Sat, 09 May 2026 14:08:43 +0000

Article URL: https://presciente.com/edition/77

Comments URL: https://news.ycombinator.com/item?id=48075126

Points: 2

# Comments: 0

New comment by sebastianperezr in "Sabotaging projects by overthinking, scope creep, and structural diffing"

sebastianperezr — Fri, 24 Apr 2026 18:55:07 +0000

I am a solo builder and one thing that helped me a lot was this: most of what looks like a "necessary abstraction" is actually scope creep with a different name. I was adding a flag for every new feature and I noticed the pattern in my own code, so I made one rule: a feature can not land until its flag-off behavior has a test. That changed how I see flags. The flag is part of the product, not an escape hatch. Three features in my backlog died by themselves when I started thinking like this.