# second day (blogs) — RSS Amplifier

Recent posts from the 3 feeds in the RSS Amplifier directory that cover second day.

Page: <https://rssamplifier.com/topics/second-day/blogs>  
Feed: <https://rssamplifier.com/topics/second-day/blogs.md>

---

## [A paper reproducibility agent: PDF in, report out](https://blog.jasony.me/posts/2026-08-16-paper-reproducibility-agent.html)

_2026-08-16 · Juncheng Yang_

We turned the MLSys'26 AE agent's core into a standalone paper reproducibility agent. Give it a paper PDF; it finds the repo, installs and runs it, and adjudicates the paper's claims.

## [Same open weights, prompt-cache lifetimes from seconds to a day](https://blog.jasony.me/posts/2026-08-12-prefix-cache-retention-open-weight.html)

_2026-08-12 · Juncheng Yang_

Across 23 providers serving two open-weight models, prompt-cache retention ranged from under 30 seconds to more than a day. Even for the same model, the serving stack determined the cache.

## [Midnight on the Dragon Tower](https://thousandtales.substack.com/p/midnight-on-the-dragon-tower-3d7)

_2026-08-10 · Alex Evans · A Thousand Tales_

Ceremonial Third Day

## [Deploying an artifact-evaluation agent at MLSys'26](https://blog.jasony.me/posts/2026-08-09-deploying-an-ae-review-agent.html)

_2026-08-09 · Juncheng Yang_

We deployed a tool-using agent on the MLSys'26 AE HotCRP as a real reviewer. Across its completed reviews, it usually scored within one point of its human co-reviewers.

## [Measure first, design second](https://blog.jasony.me/posts/2026-08-08-measurement-first-systems-research.html)

_2026-08-08 · Juncheng Yang_

AI is making solutions cheaper, but it is not making important problems easier to find. Measurement is how systems research discovers what matters.

## [The Dungeon Crawl Which Has Been Running Every Summer Since 1990](https://thousandtales.substack.com/p/the-dungeon-crawl-which-has-been)

_2026-08-04 · Alex Evans · A Thousand Tales_

Fort Boyard

## [Proprietary prompt caches expire in minutes, not hours — cheap per token doesn't mean a cheap bill](https://blog.jasony.me/posts/2026-08-04-prefix-cache-retention-proprietary.html)

_2026-08-04 · Juncheng Yang_

OpenAI, Gemini, Anthropic, and xAI all cache your prompt prefix — but the probe that caught open-weight backends holding prefixes for hours catches the big proprietary APIs letting go in 5 to 30 minutes.

## [Midnight on the Dragon Tower](https://thousandtales.substack.com/p/midnight-on-the-dragon-tower-53e)

_2026-07-25 · Alex Evans · A Thousand Tales_

Ceremonial Second Day

## [Your LRU cache is working too hard](https://blog.jasony.me/posts/2026-07-23-lazy-promotion-cache-eviction.html)

_2026-07-23 · Juncheng Yang_

Across 6,357 production traces, most LRU promotions did little useful work. Delaying the decision made caches faster without making them less effective.

## [The Cursed Object Aisle](https://thousandtales.substack.com/p/the-cursed-object-aisle)

_2026-07-19 · Alex Evans · A Thousand Tales_

A Trip Through Vide-Greniers

## [100,000 Bonus Points Await You (Sponsored)](https://crawlproof.com/a/SFpzd1Kv6Rs4)

_2026-07-19 · **Sponsored**_

Earn 100,000 bonus points after spending $5,000 in the first 3 months.

## [Midnight on the Dragon Tower](https://thousandtales.substack.com/p/midnight-on-the-dragon-tower-b2f)

_2026-07-10 · Alex Evans · A Thousand Tales_

Ceremonial First Day

## [Authorship inflation at OSDI: counting papers two ways](https://blog.jasony.me/posts/2026-07-10-osdi-authorship-inflation.html)

_2026-07-10 · Juncheng Yang_

OSDI author lists have more than doubled since 1994. Whether the field's most prolific authors are actually publishing more depends on whether a paper counts as one, or as one divided by its author count.

## [The Real Dragon's Hoard](https://thousandtales.substack.com/p/the-real-dragons-hoard)

_2026-07-03 · Alex Evans · A Thousand Tales_

The Dynastic Jewels Exhibition

[Listen](https://api.substack.com/feed/podcast/204972035/069f5dfd48ec9196ecb289e064858d4c.mp3)

## [Midnight on the Dragon Tower II](https://thousandtales.substack.com/p/midnight-on-the-dragon-tower-ii-fdf)

_2026-06-25 · Alex Evans · A Thousand Tales_

The Windworker

## [Why I'm Skeptical about writing with AI, episode 2](https://thousandtales.substack.com/p/why-im-skeptical-about-writing-with-b24)

_2026-06-18 · Alex Evans · A Thousand Tales_

Comparative Style Analysis

## [Why I'm Skeptical About Writing With AI](https://thousandtales.substack.com/p/why-im-skeptical-about-writing-with)

_2026-06-03 · Alex Evans · A Thousand Tales_

No, not for the usual reasons

## [Make model precision elastic](https://blog.jasony.me/posts/2026-05-24-make-model-precision-elastic.html)

_2026-05-24 · Juncheng Yang_

MorphServe temporarily converts selected model layers into KV-cache headroom when traffic spikes, then restores full precision when the pressure passes.

## [Why FIFO is (almost) all you need for cache eviction](https://blog.jasony.me/posts/2026-05-20-fifo-is-all-you-need.html)

_2026-05-20 · Juncheng Yang_

LRU tries to rank everything. Production traces showed that caches need something simpler: reject one-hit objects quickly and delay promotion until it matters.

## [What I look for in systems research students](https://blog.jasony.me/posts/2026-01-02-advice-for-prospective-students.html)

_2026-01-02 · Juncheng Yang_

I am looking for curious, persistent students who want to build real systems. Here is what the work requires and how to write an email I will read carefully.

