Eric Ng

Lead OSS dev @OpenZeppelin | Rust and Codex enjoyer

[Eric Ng @ericng39]

Codex needs an ultracode-style orchestrator asap.

Saying that as a Codex defender and early Symphony user. This feature is awesome.

[Eric Ng @ericng39]

The other half is either OpenAI, Anthropic, or Elon btw

[Eric Ng @ericng39]

Half my timeline in the last two weeks is people tweeting either about leaving a company or joining a new one.

I understand this happens often but why is X showing them all to me?

[Eric Ng @ericng39]

@banteg Are you actively using Atlas? would you recommend migrating to it from chromium based alternatives or is it still too early?

[Eric Ng @ericng39]

@gusgonzalezs This ofc has some rather obvious exceptions, but is certainly a good metric to follow.

[Eric Ng @ericng39]

@gusgonzalezs I saw an interesting YC talk today about building a self-improving company where the hottest take was that engineers who token-max, even today before the harness is perfect, are the only ones worth keeping around as a competitive company

[Eric Ng @ericng39]

I honestly don't get how Opus models always look better than GPT on every chart I see, while my impression over the last 5 months is that Codex is way smarter than Claude.

What is it? Context handling? I usually use GPT in Extra High and Opus in Max, so maybe faster versions?

[banteg @banteg]

first impression of claude 4.8 is it's extremely convincing but still a slopus. tried it to criticize a new project and it identified it fell into a local minima and invented a new parser for when we could've used ast.

almost convinced me, glad i checked myself that ast is not emitted in older versions of the compiler we are targeting. codex chose a gnarly but ultimately justified approach. claude didn't bother to verify any of its claims and has used absolutist language like "delete https://t.co/zuys0EhoHP", which is basically 80% of the codebase.

when presented with evidence:

> That contradicts my earlier byte-count check, and it matters enormously

> My earlier "v0.2.9" was a double false-positive (a git log -S hit on an internal symbol, plus a verification grep that mis-read a VersionException as success). Corrected in the review with a note owning the error

the biggest bullshitter model in the world! if you rely on claude for anything, god help you.

[Eric Ng @ericng39]

@mert So if I unfollow you go back to 999k?

[OpenZeppelin @OpenZeppelin]

Sui developers, put your Move security skills to the test ⛳️

Move-over is an open source, browser-based CTF. Read the vulnerable contract, write your attack logic and return the flag.

Like Ethernaut, but for @SuiNetwork 💧

[Eric Ng @ericng39]

People will really start caring about onchain privacy the moment their local IRS starts sending tax letters based on wallet activity.

[Brother Odin 🥷🏽 @odin_free]

everyone wants more tps. nobody asks if the chain is even private.

congrats, you built the fastest way to broadcast your whole financial life to strangers. respect.

[Eric Ng @ericng39]

@segun_os_ There is vibe coding and then proper vibe coding. You can properly vibe code in C++ no doubt

[Eric Ng @ericng39]

@sri9s You should probably be using both of them

[Eric Ng @ericng39]

@did0f It takes effect next prompt. I confirmed it with Codex.

[Eric Ng @ericng39]

Ok, help me out here.

I use Codex daily through Chat GPT Pro for coding, researching, debugging, writing... usually in Extra High with multiple agents

And somehow, I still haven't hit any usage limits.

Am I underusing it?

[Gustavo Gonzalez @gusgonzalezs]

How to go broke in just a few prompts

[Eric Ng @ericng39]

@gusgonzalezs @taikoxyz @InferenceRoom Congrats @gusgonzalezs! Bullish on the things to come.

[Eric Ng @ericng39]

@avihu28 This is fire! and probably worth requesting early access to Claude Mythos 😅

[Eric Ng @ericng39]

Hey, adding some context from OpenZeppelin here.

U512 is not a general-purpose integer. It exists purely as an intermediate type to safely hold the result of a u256 × u256 multiplication before reducing back to u256. That's how it's documented and how it's intended to be used.

On the division side, there are two completely separate paths:

Fast path: if the numerator fits in 256 bits, it does a single native u256 division. No loop, no overhead. This is what you hit in the vast majority of real usage.

Slow path: the loop only runs when the value genuinely exceeds 256 bits, meaning two large enough u256s were multiplied that the result truly needs 512 bits. Even then, it starts from msb() with early exits.

So the "256x for every user" claim doesn't hold: fast path is effectively 1x, and the slow path only triggers for true 512-bit values. In practice, Sui's ecosystem overwhelmingly uses u64 types, so near-zero users ever touch that code path.

That said, we're aware the slow path can be optimized further, and it's on the roadmap.

[Eric Ng @ericng39]

The first example that comes to mind is incident response.

If you have many contracts that are effectively instances of the same system, like pools in a DEX, accounts in a wallet provider, markets in a prediction protocol, etc. and a critical bug is found, upgrading one-by-one gives bad actors potential enough time to exploit.

[Eric Ng @ericng39]

What's the current state of the art for upgrading hundreds of contracts to the same implementation on Starknet?

[Eric Ng @ericng39]

Let's work on a standard for this together 👇

https://t.co/PDz7mugxno