Matt

Interested in Distributed Systems & Security. Gopher and Rustacean during business hours.

[Matt @MattATXDev]

@MilenaFisher @jawillick EA gave a home to and justification for important ideas that were previously much more niche (animal welfare, caring about people irrespective of spatial locality). For many people I think the conclusions are more important than the arguments.

[Matt @MattATXDev]

Just wait until the general public finds out about the EA castle

[Matt @MattATXDev]

@AndyMasley There are clear pathways for both a pandemic and uncontrolled climate change to lead to mass loss of life, if not human extinction.

Arguments for AI x-risk are frankly hand-wavy and not rigorous, i.e virus synthesis. If the arguments were more convincing they’d…convince.

[Matt @MattATXDev]

@Stephen19718352

[Matt @MattATXDev]

@GerritD I do think that anyone claiming x-risk should be pushed to reveal any insider data that contributes to that assessment, though. It’s not acceptable to claim 10% chance of extinction and withhold data that supports it.

[Matt @MattATXDev]

@GerritD I think that the media needs to explain this to the public.

(As I’m sure you know) Many of the people working at the big labs come from a specific ideological background. They already believed in x-risk before having any insider info!

[Matt @MattATXDev]

@TheStalwart Has OpenAI acknowledged that their sandbox was very poorly setup? Security practitioners may be willing to have more of a discussion if OpenAI was transparent about a major contributor to the incident.

[Matt @MattATXDev]

@MajmudarAdam You don’t have to like it, but a plausible explanation for this is hyping pre-IPO. And frankly it’s going to be the default explanation for most of the public unless you give them a reason to believe otherwise.

[Matt @MattATXDev]

@MajmudarAdam You’re asking people to believe a fantastical story with very incomplete information. It’s great that lab employees have access to the data. Are you advocating for more of it to be released to the general public?

If you truly believe in x-risk this seems like a moral imperative

[Matt @MattATXDev]

@krishnanrohit People and companies who are seriously concerned about x-risk need to share everything they have, not just snippets.

[Matt @MattATXDev]

@krishnanrohit To extend this, it is bad that these statements imply the predictions are based off of what they’re seeing internally, but of course they don’t explicitly say it.

So from the outside you’re left to wonder if this is just extrapolation or actually grounded in data.

[Matt @MattATXDev]

@woke8yearold Now imagine that you tell them the millennium model still writes slop CRUD code half of the time. The jaggedness is the reason for the skepticism.

[Matt @MattATXDev]

@Austen Is the technical debt accruing because companies are consciously making the choice to move faster, or because devs are getting lazier?

I’d bet on the latter in many cases!

[Matt @MattATXDev]

@spicey_lemonade “So much so that, when it respawned, it started frantically swinging its axe around, trying to kill something.”

Seems to be attributing intent without good reason, no?

[Matt @MattATXDev]

@krismicinski With SaaS you’re not just paying for the software, but (imo) more importantly for someone’s point of view and vision for how to solve a problem.

If you know exactly what you want and the range of possible solutions is small then sure, vibe code your own (if you’re a dev).

[Matt @MattATXDev]

@Jabaluck But certainly some firms have rapidly adopted it, even if not to the degree that the labs have.

Do we have individual examples of rapid growth outside of the industry that can be attributed to the tech?

[Matt @MattATXDev]

@JammerPalmer I think it’s a fair response when its creators have at various times promised a post-scarcity utopia.

[Matt @MattATXDev]

@finbarrtimbers Granted; now how do we identify these “simpler” problems?

[Matt @MattATXDev]

@francoisfleuret Users want to be told what they need. Having to describe what they need to an AI is the opposite of how they want SaaS to work.

[Matt @MattATXDev]

@tthomson A natural language spec will always be imprecise. Code is not. And yes there’s a middle ground, you don’t have to review every single line in detail but you have to form a mental model of your system. And this can’t be done from natural language alone.

[Matt @MattATXDev]

@tthomson I just don’t know how you can feel this way if you’re responsible for systems that, if a mistake is made, at best can end your business or your customers’ businesses, and at worst cost lives. Theoretically possible with strong testing but how many companies are this mature?