asker the gauche, glycojohn destroyer of carbs

Random idea for web standards people:

An interesting test of a written standard is an implementation. Since we now seem to be living in a world where prototypes are made quickly (AI) one could request such an implementation and view confused interpretations as an opportunity to clarify.

It also leans in the direction of automated reasoning

> By enforcing invariants … we let agents ship fast without undermining the foundation.

We’ve been yelling about this at work: trust wants structured output and checks via ground truth written in logic with interpretable failures

A lot of it reads like guidelines for how to work in large teams

> In practice, this meant working depth-first: breaking down larger goals into smaller building blocks (design, code, review, test, etc)

Which makes me wonder if the people are getting better at communicating amongst themselves too?

How the GNU C Compiler became the Clippy of cryptography FOSDEM 2026: Security devs forced to hide Boolean logic from overeager optimizer

Ancient Aliens meme:

*verified compilers*

Yah totally, we can make nondeterministic systems correct/ robust. In the case of LLMs that’s a per request/per system proposition but there are other such very complex (inscrutable) translation systems with universal correctness properties (like compilers) which makes this analogy somewhat weak 🤷

I think we probably agree. On point 2 one can verify generated programs of course (do the verification too please!) Quite separately I can say once and for all what it means for a compiler to be correct in translating between languages but I can’t do that for LLMs which go from English to code.

When one needs to understand the build system to answer questions about outcomes that’s significantly easier for deterministic pipelines.

Similarly, I see this same argument made wrt compilers but it is possible to say what a compiler should do to be correct, that’s not true for an LLM.

Huge. Congratulations friend!

There’s a substantial portion of the US population that’s having a full mental breakdown right now.

This is almost entirely unrelated to the content of the article.

Whenever UB is discussed I rarely ever hear about the reason why it exists and why it’s good: it’s how the compiler tells you about assumptions it needs for optimizations you almost certainly like a whole bunch.

Best modern example is stacked/tree borrows in Rust unsafe regions.

Buried way way down in that article:

> Waymo has been fairly upfront about its human operators. In a May 2024 blog post, the company compared it to a “phone-a-friend.”

There’s also clearly an interesting opportunity to hook up CSmith in a tight loop with the output here to see how well it responds to feedback about output discrepancies

Separately, imagine project/code specific compiler optimizations backed with general purpose translation validation to ensure right output even when the pass is generated by an LLM.

Building a C compiler with a team of parallel Claudes Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.

www.anthropic.com/engineering/...

Now imagine that we could convey what a correct compiler is in a useful way for this process to leverage. The statement is complex if you care about optimization but it does exist as math.

new CSmith approach just dropped

this has some real “our food product is uranium free” energy, ie no credit awarded

<<WARNING: explosions may occur>>

I am going to assume for fun that he just forgot to press the ‘q’ in which case … yes you are still correct 😅

A poster with a smiling man wearing a helmet on it. It says we ride in unity sat 1/31. 1 pm meet 130 pm roll. Dupont circle Washington dc. Ride together stand together

A ride for Alex Pretti is being organized in DC on Saturday, 1/31. Meetup at Dupont Circle Fountain at 1pm, roll at 130. Dress warmly and wear layers. #BikeDC
www.instagram.com/p/DUERoBsj6e...

“Lock up your premises friends, linear logic is out to devour them.”

choosing to interpret “software should work” to mean that this is an AI conference

I’m worried your heat sink is too small.

many many handspans on the map for some

okay I was hoping this was about guaranteed termination

One can also delete the account from the web as a matter of data privacy in some states (I did this in CA)

www.hilton.com/en/hilton-ho...

slot pulling outdoors is an under explored business opportunity

It’s a very weird sign of the times but even a cursory understanding of Fed history makes clear that we are well off the rails if the chair is making public addresses about the pressure campaign from the Whitehouse.

you do have a strong moral compass @steveklabnik.com

feel like this deserves a dropped ‘g’