i don't have high hopes

obviously not in the time between ejection and impact....... but before an ejection happens in the first place

Garfield "You Are Not Immune To Propaganda" meme modified to say "Flattery" instead of "Propaganda"

apparently

yes VMMs have been exploited too but the TCB is small enough that we can probably get it ~secure pretty quickly

you don't have to turn off all the internet facing software, you just have to wrap everything in a VM

it makes me laugh every time she sits like this

The genie in the bottle is a cautionary tale about underspecifying your agentic prompts

css mona lisa but it's the mythos svg pelican

don't worry, the benevolent claude world government will recognize this danger and harden the infrastructure in time

oh man it's not clickbait but it still makes me want to watch each of these so badly (meaning yes)

I don't remember seeing an image of Earth before that captures the (thin, fragile, tiny layer of) atmosphere so clearly. I mean, look at that.

Artemis 2 was pretty cost-effective at fostering we're-all-on-the-same-team-spirit, you love to see it. "Only on twitter", fine but Twitter has a rather outsized role in setting the vibes and tribal shifts of the world, as the current administration clearly demonstrates.

i have always insisted that fascism innately and necessarily reduces competence but i wouldn't have dared to dream that competence and alignment are correlated quite that strongly

AI Labs (Anthropic with Claude Code and OpenAI with Codex) pampering software developers.

I find myself inclined to end on some positive notes. Mythos appears to be—according to Anthropic at least— "the most aligned" model Anthropic has ever trained. We are approaching superhuman capabilities in some domains, and yet alignment is getting better rather than worse. That's not nothing. I know some of you think the model is faking its alignment, or aware when its alignment is being tested. I don't have a good answer.

for as crazy as everything this, it’s worth appreciating this

As the models get bigger, they seem to become MORE aligned, not less

A HUMAN MUST NEVER MAKE A SECURITY DECISION

maybe for this example it could be "the unique combined propulsion and breathing system and transformation of the dragonfly nymph"

i have seen it, i actually really like that channel.
i do think a balanced version like that would be good! specifically the one that doesn't simply summarize the video, but just makes a specific and accurate promise about the content and why it is (or isn't) interesting

actually i think i don't like this. the joy is in the learning, not the knowing

basically all software becomes seL4-tier security postured, because we now have the cognitive resources to do that. but for small, poorly resourced projects this dynamic seems attacker favored for a while just because of the sheer speed

okay i think we're officially at the point where there's no point anymore even considering the willingly and hopelessly left behind. this is the last thought i am willing to spend on AI deniers 😌

i think they're too lawful good for that

Man, if farming labor share ends up nonmonotonic that would be an interesting trajectory for humanity.

You know, candidly, I think I believed the capability curve would be starting to look more sigmoid by this point. Also: I would like it to start looking more sigmoid.

Mythos is still missing... Something

i wonder if they've aligned mythos sufficiently to make it willing to perform illegal actions

i'm afraid my PMC is currently unavailable, which really does reduce the options dramatically

i really need to make some money before the economy collapses entirely.
anyone got a get rich quick scheme?

they're going to make beyond money