gmonaco@🦋 Engineer. he/him. Opinions my own.

Joined February 2011
React CHUNKS? 😱 Learn about some of the surprising history of Server Components in this finale of Bite Sized Server Components. youtu.be/vWHwBXP-sQk #reactjs #nextjs #javascript
6
11
3,370
It's David v. Goliath and everyone on Twitter is throwing rocks at David.
1
98
Apparently a lot of people on this website staunchly believe that the Earth's atmosphere is but a warm fuzzy coat protecting us from the frigid winter temperatures of outer space. I just cannot even anymore.
41
I guess if it were a choice between not citing prior work at all, or citing them even without reading the work, the latter would be preferable...
1
47
The Supreme Court’s legitimacy depends on providing “Equal Justice Under Law,” as the courthouse says. When the Supreme Court rules one way when it hurts Democrats and another way when it helps Republicans, it’s not doing law — it’s doing raw power politics. Americans are rightly questioning why it deserves such power.
Supreme Court today: We'll allow Alabama Republicans to change the congressional map in the primary election even though absentee voting has literally already started Supreme Court last December: November's too late to strike down the GOP's map in TX's March primary election!!
262
1,017
29
3,258
219,394
Your air-gapped servers are covered by a faraday cage, you think you're safe from key exfiltration? You fool. Low-frequency magnetic fields pass right through Faraday cages. Researchers were able to extract data off an air-gapped shielded computer by spawning fake work loads spiking CPU power and generating magnetic signals.
53
The RF world is insane. Researchers recovered AES-128 keys from a Bluetooth chip by listening to its own antenna from 10 meters away. Crypto-engine switching noise couples into the RF chain, rides the 2.4 GHz carrier, and leaks out as radio.
1
69
Good in theory. But having more room to talk about how good a game is, while the top two numbers are exponentially better/harder is precisely where score inflation occurs, which is what makes 7 start to feel like barely passing.
Too many people think this is like grade school where a 69 and below is a failing grade. 5 is average, 6 is fine/okay, 7 is good, 8 is great. Glad our scale is like IGNs because if a 7 was considered a BARELY passing score (a 70 or a D in school) and 0 to 6.5 meant a failing grade, that's way too much real estate on the scale to grade a bad game. I don't need that many levels to let you know how badly a game is failing. I prefer to have more space on the scale to describe a game I genuinely recommend or enjoy.
1
73
You should bucket into as few of buckets as you can, where each bucket is categorically different. E.g. 5: Must Play 4: Recommend playing 3: Consider playing 2: It's complicated 1: Don't recommend 0: Can't Play
1
34
Still more positive space than negative. Decimals can be used for more nuance while disarming inflation. Broad categories means scores stay grounded over time.
26
Seeing a lot of "they're using older or non-agentic models." That's... the point? Solving the problem with a compiler or interpreter in the loop isn't reasoning, it's bruteforcing. So many people acting like this research is a personal attack on them. Don't we want better AI?
🚨 Shocking: Frontier LLMs score 85-95% on standard coding benchmarks. We gave them equivalent problems in languages they couldn't have memorized. They collapsed to 0-11%. Presenting EsoLang-Bench. Accepted to the Logical Reasoning and ICBINB workshops at ICLR 2026 🧵
1
2
101
"It's not fair because humans need feedback too" — Engineers used to be judged by their ability to perfectly solve a problem at a whiteboard without a compiler. If the goal is to surpass human _reasoning_, this seems like a reasonable benchmark, no?
1
29
Last week: "AI is just going to generate machine code directly" This week: "You can't expect AI to write Brainfuck!" Brainfuck's abstract model is almost literally just a turing machine. Translating any program to it, as with any turing machine, is a mechanical process.
More proof LLMs aren't conscious and aren't generalizing any information, and therefore aren't going to become generally intelligent, but are in fact (still extremely useful) trained statistical responders.
1
1
269
The vast majority of engineers I've worked with in my career, while unwilling to complete this task, would be capable of completing it after a tweet-sized explanation of brainfuck. That's what generalization means.
1
24
Claude has also seemingly reinvented llvm's GetElementPointer, which isn't really something that a standalone C compiler should need. Possibly more indicators of the training set being reproduced.
New Engineering blog: We tasked Opus 4.6 using agent teams to build a C compiler. Then we (mostly) walked away. Two weeks later, it worked on the Linux kernel. Here's what it taught us about the future of autonomous software development. Read more: anthropic.com/engineering/bu…
126
It's worth pointing out that this "clean room, Rust std only" project somehow includes a copy of rustc's Apache-2.0 licensed fx_hash, which Claude has apparently relicensed under CC0.
New Engineering blog: We tasked Opus 4.6 using agent teams to build a C compiler. Then we (mostly) walked away. Two weeks later, it worked on the Linux kernel. Here's what it taught us about the future of autonomous software development. Read more: anthropic.com/engineering/bu…
1
8
1,084
Also: a recount wouldn't discover this, since all ballots are "valid". Instead, you have to audit the ballots themselves, ensuring the voters 1.) were actually alive at the time, and 2.) voted for who the ballot claimed.
This will be my only tweet about this, but: if I wanted to steal an election, I would first spend four years accusing the other side of doing X, and then do X. Where X is: get loyalists delivering ballots in my districts to include an extra few boxes for me en route to counter.
161
This will be my only tweet about this, but: if I wanted to steal an election, I would first spend four years accusing the other side of doing X, and then do X. Where X is: get loyalists delivering ballots in my districts to include an extra few boxes for me en route to counter.
1
1
303
By accusing them first, I've primed them to be skeptical of cheating accusations. By cheating in my own districts only, I can't be discovered by partisan audits, and I get more bang for my buck from gerrymandering. It's the Office Space approach to fraud.
79
Gerald Monaco retweeted
The Chrome Aurora team is excited to share that Nuxt Scripts has launched! Nuxt Scripts, developed by @nuxt_js in collaboration with @googlechrome, improves performance of third-party scripts and offers prebuilt solutions for many popular third-parties. scripts.nuxt.com/
1
3
6
1,268
This is a suggestion to level up your Scadutree Blessing
369
3,492
682
32,051
2,391,439