Programming Language Theorist Systems Architect/Engineer USMC war veteran https://xitter.gq/t.co/s25CQwWEUg
Earth
Joined January 2010
- Tweets17K
- Following237
- Followers198
- Likes7.7K
Pinned Tweet
"The secret to enjoying your job is to have a hobby that's even worse" #PLT
en.wikipedia.org/wiki/Progra…
Michael Haufe retweeted
A programming language requires local reasoning and a stable semantics, a clear cause and effect between code and behaviour.
Even fully deterministic LLM tooling enables a stray word in an unrelated comment to influence outcomes.
That makes them terrible compilers.
Michael Haufe retweeted
AI is creating fabrications in scientific papers
A random sample of 500,000 peer-reviewed journal papers found that fabricated references rose from 6.3 per 10,000 in 2023 to 23.0 per 10,000 in the first quarter of 2026 (with no signs of abating).
This is going to create a challenge for journals to catch these errors and hold authors accountable when they submit work with serious flaws.
pnas.org/doi/10.1073/pnas.26…
Michael Haufe retweeted
JavaScript. Lisp. Forth. Smalltalk.
Languages aren't inherently hard.
It's improving state of the art enough that a transition is worthy that is hard. Obvious, low-hanging fruit was plucked long ago. What remains is either non-obvious (without hindsight) or awkward to reach.
Michael Haufe retweeted
Leibniz presented his solution to the problem of evil through dancing statues, Prometheus choosing humanity’s fate, and an infinite pyramid of possible worlds.
A previously overlooked manuscript adds a fascinating chapter to the story.
leibnizpapers.org/prometheus…
Michael Haufe retweeted
though we can't overlook that without Smalltalk you likely don't get a successful Apple and you would have had VBScript instead of JavaScript in the browser.
Would we have gotten MVC far later?
Michael Haufe retweeted
Instead of using a battle-tested open-source lib that you have to maintain, you vibe-code your own version, which you also have to maintain. Brilliant plan.
Michael Haufe retweeted
Anthropic (and all the others) have pirated all books, illegally trained their models on them, then when caught they destroyed all books they could buy (inventing the practice of knowledge laundering).
But when the user wants to find a reference to a paper that is 94 years old, it's suddenly concerned about copyright violations.
It's not even about refusing a task (bad in itself), it performed the task just fine, but it decided to withhold information from you because Anthropic thinks they are above you.
These are the people who want to be arbiters of morality.
Michael Haufe retweeted
This is the study I cited in the interview.
Living within 1 mile of a golf course increased Parkinson’s risk by 126% compared to living over 6 miles away, and the greatest risk (a 198% increase) was observed for those living 1–2 miles from a golf course!
People relying on groundwater near golf courses had nearly double the risk of Parkinson's compared to areas without golf courses.
Dr Rhonda Patrick says one study found people living within about a mile of a golf course had a much higher rate of Parkinson's, and she explains why pesticides in the water are the suspect.
@foundmyfitness
Michael Haufe retweeted
No, Binance was not most likely compromised from a web browser zero day chain.
As is tradition with large compromises, it is almost certainly something ridiculously stupid, like the DevOps guy trying to get Fortnite cheats, or they called someone with a shitty American accent and said they need to log into their machine for something
Michael Haufe retweeted
“Your job is to help the LLM write better code faster.”
Strange: I thought the job of a software engineer was to bring value to stakeholders in an enduring, economically sustainable, and ethical fashion.
I suggest you reconsider your life choices, Peter.
Michael Haufe 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。
Michael Haufe retweeted
Under @Ngnghm's efforts, I've come to appreciate OOP as paradigm-independent.
It's convenient to model grammars as objects (methods are parser combinators), databases as objects (methods as tables or views), functions as objects (~ letrec with mixins and overrides).
Michael Haufe retweeted
Code changes in safety-critical systems have to undergo an entire impact assessment process that could take months, as they may cause failures. It's a very different (and safer) type of engineering.