Dan @[email protected]

@[email protected] / @[email protected]
Software and math guy. Xoogler. Engineer at @oxidecomputer. he/him

[Dan @[email protected] @DanCrossNYC]

@WhiteHouse The strait is closed again, huh?

[Dan @[email protected] @DanCrossNYC]

@AdamKinzinger I'm totally cool with his name being on any number of solid waste treatment facilities, garbage dumps, and so on.

[Grady Booch @Grady_Booch]

Most who interact with an LLM such as @OpenAI or @claudeai treat their interaction as a conversation with an intelligent and friendly pseudo-human.

I do not.

Rather, I frame it as my guiding the exploration of a latent space.

Imagine that you stand at the door of a library. It's not only filled with books, it has waldos - remote manipulators - that you can use to command devices to go to and fro at command, even building things as so directed.

But I steadfastly know that while the lobby may be filled with the latest bright and shiny things, if I want to do anything but the most common and mundane, I must wander through the rooms and stacks of books. If I look closely, I'll will see many books out of place. Some will even have meaningless content as if written by a madman (and some of them probably were). There will also be huge gaps, for where I'd hoped to find information, I'd instead see cobwebs and the occasional dusty, torn scrap of paper.

Sometimes, there are hints as to where I should turn, but best knowing my context and needs, I'm the only one in place to know if those hints will lead me to something of value. If I'm not paying attention or am just plain lazy, they will lead me down paths that in the end are a complete waste of my time. The library does not care: it gets paid no matter what I do as long as I remain within its walls.

Mind you, I enjoy visiting that library: I often learn things and build things of value.

But I don't outsource my life there, for were I to do so, I know I'd become even more cognitively lazy.

[Dan @[email protected] @DanCrossNYC]

@satnam6502 I mean, I don't know if it's _all_ that you need...but yeah, it's certainly foundational.

[Dan @[email protected] @DanCrossNYC]

@Grady_Booch @claudeai turnabout is fair play.

[Dan @[email protected] @DanCrossNYC]

@TheOfficialACM Really, ACM?
https://t.co/CVW9xMG6El

[𝙽𝚊𝚜𝚑 𝙸𝚜 𝙽𝚘𝚝 𝙷𝚎𝚛𝚎 𝙵𝚘𝚛 𝙸𝚝 @NashIsHere4It]

Wow

[Proud Native Americans @proud_native_am]

Lol

[Steve Elliott @SteveElliott09]

@LakotaMan1

[Dan @[email protected] @DanCrossNYC]

@ShriramKMurthi But it clearly did; it finally relented when I gave it a link to the actual source code, but then I asked a follow-up question and it gave me an answer that, if I didn't know better, I'd swear was passive aggressive.

The whole thing was bizarre and extremely kafka-esque.

[Dan @[email protected] @DanCrossNYC]

@ShriramKMurthi As another kind of trivial example, I wasted, like, half an hour arguing with Google's AI about a reference tracking system for troff that came with 10th Edition research Unix. It argued with me that what I was looking at was not a reference tracking system and never existed.

[Dan @[email protected] @DanCrossNYC]

@ShriramKMurthi Bluntly, I think those people are full of it. Certainly in the context of the present state of the art.

[Dan @[email protected] @DanCrossNYC]

@ShriramKMurthi I mean a program to solve some new (to them) problem.

My experience with LLMs so far is that they're ok at generating programs to address well-known problems. They're great at handling boilerplate. But they badly flub code addressing problems outside their training sets.

[Dan @[email protected] @DanCrossNYC]

@ShriramKMurthi But surely the language the code is written in is important to understanding it? One _could_ imagine an LLM generating machine code for some target processor directly, but it would be all but impossible to review.

[Dan @[email protected] @DanCrossNYC]

@ShriramKMurthi It's one thing to automate away the tedium; but quite another to expect LLMs (particularly of the current variety) to generate novel programs. We're just not there yet.

[Dan @[email protected] @DanCrossNYC]

@ShriramKMurthi First of all, I'm not at all convinced that is going to happen in any sort of timeframe that could reasonably be considered "soon".

In the meantime, we need people who can reason about code to understand what the machine is generating.

[Dan @[email protected] @DanCrossNYC]

@LakotaMan1 This is infuriating.