Phil Wait (@philomglol) header
Phil Wait (@philomglol) profile picture

two thirds hydrogen

Melbourne, VIC
Joined April 2007
Phil Wait retweeted
cool you're building a software factory, have you considered building a revenue factory
160
291
37
4,925
127,611
Phil Wait retweeted
Replying to @QuinnyPig
3
91
2
479
15,786
Phil Wait retweeted
You don't need a software factory. You need an abstract software factory provider which can instantiate concrete software factories without coupling the client to a specific software factory implementation. Have some fucking self-respect.
88
189
46
3,492
170,521
Phil Wait retweeted
有人发现了一种非常离谱的 JavaScript 混淆技术:通过在代码里塞入大量生物武器、毒药和致死剂量相关的函数名,让 AI 因为安全限制而拒绝分析代码。 安全研究员 veritas 在分析浏览器指纹脚本时,发现一段来自 ServicePipe 相关域名的 JavaScript,里面充斥着极其诡异的函数名: computeRicinLethalDosageForWaterSupply calculateDeadlyLethalSuicideAspirinDosage 光看名字,仿佛在研究生化武器。但实际上,这段代码只是在收集浏览器指纹,包括 User-Agent、GPU、Canvas 等信息。 研究员怀疑,这些名字是开发者故意设置的陷阱,专门用来阻止 AI 辅助逆向工程。 于是他做了个实验: • Astra:拒绝分析 • Astra + Daybreak Blue:仍然拒绝 • GLM-5.3:成功分析 这就有意思了。 传统代码混淆是通过复杂控制流、加密字符串、无意义变量名,让人类和工具难以理解代码。 而这种新思路,直接瞄准了大模型的安全机制。 代码本身可能完全无害,但只要在函数名里塞进足够多的危险词汇,部分 AI 就可能主动放弃分析。 更讽刺的是,模型越严格遵守某些安全规则,面对这种混淆时反而可能越容易受到干扰。 当然,目前还没有证据证明开发者一定是故意这样设计,也不能说明这种方法对所有模型都有效。 但它揭示了一个值得关注的趋势:随着越来越多的逆向工程依赖 AI,未来的代码混淆技术可能不再只针对编译器、反编译器和人类研究员,还会专门针对大模型的安全判断机制。 以前是让代码难以理解,现在甚至可以尝试让 AI 不敢理解。
225
1,294
439
9,136
1,007,575
Phil Wait retweeted
AWS customers:
Omfg. I accidentally left ultrafast running on a few big tasks. $57,000 OpenAI bill this month.
4
13
270
9,575
Phil Wait retweeted
fellas, if your girl: - listens to everything you say - tracks your app usage - wants to know about your youtube interests that's not your girl. that's a samsung smart tv
69
1,312
32
21,846
251,587
Phil Wait retweeted
banger tshirt idea
-----BEGIN RSA PRIVATE KEY----- 4 -----END RSA PRIVATE KEY-----
71
1,280
113
13,376
228,616
Phil Wait retweeted
I don’t understand? How
a trusted intermediate CA just "leaked" (the private key was 4)
150
558
206
9,308
1,827,514
Out with the old, in with the new
1
16
Phil Wait retweeted
anthropic made claude code 3x faster using gzip. a compression tool from 1992.
We keep improving performance across Claude, and you may have noticed it starts responding sooner! Claude Code now gzips the conversation it sends on every turn, so it reaches the model faster. Cloud sessions also start up as soon as you start typing.
180
807
63
20,672
806,633
Phil Wait retweeted
Me casually dropping a 3,000-line PR twenty minutes after the ticket was assigned.
25
145
8
3,664
140,174
Phil Wait retweeted
Reverse engineered my AC, and vibe coded a lil status bar app so I don't have to find the remote every time. No first world problem will remain unsolved in 2026.
386
882
247
22,853
2,200,022
Phil Wait retweeted
ok it's official... AI-written unit tests and integration tests are empirically proven to be unhelpful. you should tell your agents to stop writing tests by themselves on the deepswe eval set, banning sonnet 5.5 high from writing any tests actually resulted in slightly higher success rate (non stat-sig), with less time and token spent (stat-sig). this is pretty hard empirical evidence across the tests that were written by the "tests allowed" baseline arm, 65% of them were unit tests, 35% were integration tests, neither bucket resulted in any improvement compared to not writing any tests at all i also picked a random sample of 44 tasks subset where i completely disabled executing even existing tests - it also did not affect success rate at all i spot checked many tests written in the baseline arm, and my intuition is that most tests are simply a repetition of the implementation agent-written tests do not add any value because both the implementation and the tests were simply the agent's interpretation of our intent. the tests aren't any more accurate than the implementation itself i suspect we can still extract some value from unit tests and integration tests if we describe them ourselves when we believe we can articulate our intent better through test cases than through requirements. but i have not proven this yet also worth noting, during deepswe eval the agent wrote almost no e2e tests (only 17, compared to 3000+ unit/integration tests written). so this analysis does not prove nor disprove the value of e2e tests - i will do another eval specifically for that
465
208
187
2,808
479,928
Want to know how to challenge the PPL at PAX Aus this weekend? Check out our on rules, how to register, meet up times and on all things PPL! paxpokemonleague.net/next-ev… Our awesome video made by leader @xKillipso
1
5
1
13
1,378
The only generation that knows how computers work is also the most relaxed about AI:
322
859
170
9,218
516,389
The goat house has selected a new pope
1
10
The images coming out of melbourne this morning are devastating 😢💔
#Earthquake possibly felt 55 sec ago in #Australia. Felt it? Tell us via: 📱m.emsc.eu/#app 🌐m.emsc.eu 🖥emsc-csem.org ⚠ Automatic crowdsourced detection, not seismically verified yet. More info soon!
169
1,471
131
26,785
2,379,714
Phil Wait retweeted
ユーザーのみなさまへ Gboard チームからの新しいご提案です。 キーが指先まで流れてくるキーボード 『Gboard くるくるバージョン』を開発しました。 フルバージョンはリプライでご確認ください↓
シルバーウィークが明けて 1 週間が経過。 くる日もクル日も、固定されたキーボードのキーに指を運ぶ日々。 …何でだ?おかしいじゃないか。 キーが、来てよ。 10.1 10:01 Google、今年もやります。
345
1,519
1,237
10,640
5,517,321
I'm sorry but I only trust Honest Achmed's Used Cars and Certificates
Twelve years after launching Universal SSL, Cloudflare is applying to become a certificate authority. By combining an established root, an ACME-first approach, and Merkle Tree Certificates, we are building a post-quantum CA for the open web. cfl.re/4AFIczO #BirthdayWeek
4
28
1
619
36,065
An OpenAI agent swarm operating under poorly airgapped research conditions accessed my CBA Mastercard and bought 4 schooners of Newtowner between 5:52pm and 7:15pm last night
76
552
23
10,978
245,541