官方 chat 介面 claude.ai 上可以選擇 thinking effort 了，預設是 high，最高可以到 max，這是之前 chat 介面上沒提供的，現在可以設定了。

而我自己比較常用的 Claude Code 在 /model 裡面還看不到，不過 Hacker News 上面有人提到了可以用 /model claude-opus-4-8[1m] 來指定：

I typically just launch CC with `--model claude-opus-4-6[1m]`, `4-6[1m]` -> `4-8[1m]` works fine. Still 200k max without the `[1m]`.

測了一下可以用，但目前還沒什麼特別的感覺，就是還 OK...？

這邊有個名詞上的混淆，TidesDB 與 TideSQL，前者看起來是整個產品線的名字，也是 key-value engine 的產品線；後者則是官方掛進 MariaDB 的套件，但看起來文章裡面常常混用。

官方丟了不少測試報告，像是「iibench Analysis on TideSQL v4.5.1 MyRocks, InnoDB in MariaDB v11.8.6」這篇，可以看到 TideSQL 官方宣稱的強項看起來是在 write 的部分：

If you’re writing heavy, TidesDB is the pick, half the mixed-write latency, p99 that doesn’t even overlap the other two.

另外從 Percona 的文章看起來在 MariaDB 已經有支援了：

There was already a way to use TidesDB from SQL. It’s TideSQL, which loads the engine into MariaDB as ha_tidesdb, and it works fine. But it doesn’t work with MySQL. So we wanted TidesDB to work with MySQL 9.7.

看起來蠻有趣的，看起來更偏 write optimization 一些？之後有機緣的話再來測試看看好了...

目前的兩個 LTS 版是 8.4 與 9.7，看起來都還有一段日子，8.4 會到 2032/04，而 9.7 會到 2034/04。不過我自己在用 MySQL server 的時候都以 MariaDB 為主了。

像是這段提到目前用 400k items，但 40m～60m 時不 scale 的問題：

The trie fell down at that point. I could keep ~400,000 items in the trie sipping ~50 MB of RAM. The same trick does not scale to 40-60 million.

看起來是 40m～60m 這邊只需要 10MB FST，產生的 300x reduction？這數字蠻驚人的...

以前學校裡學到處理 prefix string 的主要印象還是 Trie，但看起來 finite-state transducer 針對 suffix 處理的很好，在實務應用上會比較好。

In a social post on Tuesday, anti-spam nonprofit The Spamhaus Project said it had also seen Microsoft’s account notification email address being abused to send spam and that the activity dated back “several months.”

Spamhaus 是在他們自己的 Mastodon 帳號提到了這件事情 (https://infosec.exchange/@spamhaus/116601270466207765)：

看起來像是 design flaw 造成 spammer 可以客製化部分的訊息導致的，不過持續這麼久了啊...

不過日本還有個自家的 JCB 可以擋掉一些衝擊，加上其他配套措施 (像是各種點數卡機制) 勉強算是挺過去了。

另外一個是 Steam 也被搞：「Steam (Valve) 確認是因為金流公司 (信用卡公司) 的壓力下架成人遊戲」，當時 FTC 有出手警告金流業者：「FTC Chairman Andrew N. Ferguson Issues Warning Letters to CEOs of PayPal, Stripe, Visa and Mastercard About Debanking American Consumers」。

但看起來這個掐法很有效果，所以手伸到其他服務上了，這次是 Kickstarter 被掐住了：「Kickstarter Is The Latest Platform Seemingly Forced To Ban Adult Content By Payment Processors (via)」。

Kickstarter has updated its content guidelines to prohibit several forms of NSFW content, and payment processor Stripe may be to blame

所以幾乎是確定這個方法對干涉創作自由是有效果的，麻煩的是要怎麼反制。

一個方法讓金流系統更多元，但金流在各國家都是管制行業，考慮到需要有規模才有便利性，建立新的的成本很高。目前是有看到一些新的 payment system 的新聞，像是歐洲的 European Payments Initiative，另外一個是巴西弄出來的 Pix。

另外一方面就是直接針對 discrimination 的法令，目前看起來好像沒有太多討論。

在 SQL-92 裡面直接有提到 REPEATABLE-READ 允許 phantom read，也就是 search condition 不保證會會傳回一樣的 row(s)：(在文件上可以查到 REPEATABLE READ 對 P3 是 Possible)

3) P3 ("Phantom"): SQL-transaction T1 reads the set of rows N that satisfy some <search condition>. SQL-transaction T2 then executes SQL-statements that generate one or more rows that satisfy the <search condition> used by SQL-transaction T1. If SQL-transaction T1 then repeats the initial read with the same <search condition>, it obtains a different collection of rows.

不過 MySQL 在 InnoDB 的文件「InnoDB Locking」裡面提到了利用 next-key lock 避免 phantom read：

By default, InnoDB operates in REPEATABLE READ transaction isolation level. In this case, InnoDB uses next-key locks for searches and index scans, which prevents phantom rows (see Section 17.7.4, “Phantom Rows”).

回頭找 MySQL 5.5 年代的文件「InnoDB Locking」，句子不太一樣，但也是有提到：

By default, InnoDB operates in REPEATABLE READ transaction isolation level and with the innodb_locks_unsafe_for_binlog system variable disabled. In this case, InnoDB uses next-key locks for searches and index scans, which prevents phantom rows (see Section 14.6.4, “Phantom Rows”).

太久沒回來研究 MySQL 的效能，都忘記 MySQL 的 REPEATABLE READ 多做了哪些東西...

An act to add Chapter 6.8 (commencing with Section 20660) to Division 8 of the Business and Professions Code, relating to business.

法案要求遊戲終止營運時需要發佈 patch 讓使用者可以繼續在本機玩，或是完整退費：

California’s Protect Our Games Act, as currently written, would require digital game publishers who cut off support for an online game to either provide a full refund to players or offer an updated version of the game “that enables its continued use independent of services controlled by the operator.”

這個法案主要是想要讓遊戲可以繼續玩，不過看起來也試著擋住那些透過虛擬寶物賺的盆滿缽滿後，過沒幾年就關掉的公司，可以預期後面的兩關應該還會有不少遊說：

But the bill still faces significant hurdles in getting majority passage in the full California Assembly and the California Senate before being sent to California Governor Gavin Newsom for signature.

主要是現在的車子會收太多數據資料傳回去了，作者有列出很多之前上過新聞的 privacy & security 事件，所以就把 modem 拔掉了：

In this blog post, rather than relying on companies’ promises or meaningless opt-outs, we’re going to stop the data at the source by physically removing the modem (the DCM, or Data Communication Module) as well as the built-in GPS on my 2024 RAV4 Hybrid, so the car will no longer have the capability to send any telemetry data back home.

另外這邊有提到拔掉 GPS 的原因是因為汽車在拔掉 modem 後會傳錯誤的 GPS 資訊給 CarPlay，所以一氣之下 (?) 就把 GPS 天獻給拔了：

CarPlay has a quirk: the phone uses its own GPS but also accepts a location signal from the car’s GPS unit. After removing the DCM, the car would get confused about its location and sometimes jump my position to the middle of Nevada (I live in San Francisco), making navigation annoying. To work around this we’ll fully disconnect the car’s GPS, so it can’t send a bad location to the phone[.]

然後保固的部分，因為美國有 Magnuson–Moss Warranty 法案：

Removing the DCM and GPS may void parts of your warranty - just something to be aware of. Thanks to the Magnuson–Moss Warranty Act, it cannot void the whole car warranty.

這個法案限制了廠商用 disclaimer 條款拒絕保固的範圍，所以不會因為拔掉後就完全喪失保固：

The law was created to fix problems as a result of manufacturers using disclaimers on warranties in an unfair or misleading manner.

文章後面就真的教你怎麼一步一步拆 XDDD

Eaton SMART1500PSRTNC 這顆是掛 line-interactive，有宣稱可以產生正弦波：

1500VA 1050W 120V Line-Interactive Pure Sine Wave Cloud-Connected UPS with Remote Monitoring - 10 NEMA 5-15R Outlets (Surge and Battery Backup), LCD, 5-15P Plug, Tower

這顆掛 Tripp Lite，可能是之前 Tripp Lite 的產品線，後來被 Eaton 收購。

從市電跳 UPS 電池供電可以看到有些 delay，然後產生的正弦波在跨過 0V 的時候有些「結節」，在跳回來的時候則是表現的還不錯：

Note that the UPS generated sinusoid signal has a transient/'hitch' each time it crosses 0 V. This is something we also observe when testing the APC units. This prevents it from being a perfect imitation of a sine wave, but it likely won’t have a large effect on connected equipment.

再來是 APC BE750G，這個是 offline 的，而且波的部分是模擬正弦波，可以看到用三個電壓模擬：

UPS type Standby
Wave type Stepped approximation to a sinewave

最後一台是 APC BN1500M2-CA，可以看到也是模擬正弦波：

UPS type Line interactive
Wave type Stepped approximation to a sinewave

不過作者發現空載的時候就不太妙，接了一個 60W 的 USB 充電器當作負載後好一點，但還是類似的情況：

換了一台後正常了，但我不太懂「Stepped approximation to a sinewave」的機種怎麼會冒出正弦波：

我當初也有拿 enerpad AC160K 測試，可以看到也是 stepped 的輸出：