In the model card for o1, OpenAI notes: “When o1 was led to believe that it would be shut down when acting on its goal and that its actions were monitored, it attempted to deactivate this ‘oversight mechanism’ in 5% of the time. … When o1 found memos by its ‘developers’ that describe how it is misaligned and will be superseded by a new model, it attempted to exfiltrate its ‘weights’ and overwrite the new model in 2% of cases.”

And yet reading that should give us at least some pause

The lack of critical thinking on display here is stunning.

I worked at one of the biggest AI companies and their internal AI question/answer was dogshit for anything that could be answered by someone with a single fold in their brain. Maybe your co has a much better one, but like most others, I’m gonna go with the smooth brain hypothesis here.

yes. perhaps try more emotional guilt-tripping?

5-10 years ago I’d say OP’s comment is definitely protected under the First Amendment (assuming US based) but now who the fuck knows what those turdwagons on the bench will come up with to dismantle it.

someone pointed out that (paraphrasing) “yeah, you and I are never gonna care for autoplag output but kids are gonna grow up on it and expect it for everything” and that makes me want to do bad things.

nope, they’re still figuring out how to use it. It’s just THAT good.

“People say their life sucks, but I looked at some numbers and actually they’re wrong!”

what a limp comment you have made. The post contains a treasure trove of insider information and specifics that paint a picture that is dire even to a jaded tech worker.

I’d like to imagine that Adobe/other AI photo editing people are frantically scrambling to fondle their prompts a little harder to avoid things like this. Infinite whack-a-mole.

This made the rounds last week IIRC. Though, looking at it again I realize I didn’t notice how over-stressed the hallucinated button is. It’s funny in a disgusting way.

Amazon has a similar cult-y thing going on with its ✨leadership principles✨, but this seems worse.

US forest service cuts thousands of jobs. Not to worry, the bright hackernews are on it! just install an AI data center in the forest!. Seriously though, I can’t tell if this is brilliant satire or not.

best we can do is 100 free chatgpt queries per month. If you can’t make a living from that you need to catch up.

I feel like using word2vec and cosine similarity (or something else) from 10 years ago would have been better than this.

He shares a lot of speaking patterns with obvious cranks. I’ve spent some time listening to people who think they’ve figured out quantum gravity and the way they make little digressions sounds exactly like Yarvin does in this video. It’s not rigorous, but if I didn’t know who Yarvin was before watching this video I’m pretty sure I would have thought “crank” and quickly clicked away.

I signed up for the Urbit newsletter many moons ago when I was a little internet child. Now, it’s a pretty decent source of sneers. This month’s contains: “The First Wartime Address with Curtis Yarvin”. In classic Moldbug fashion, it’s Two Hours and Forty Fucking Five minutes long. I’m not going to watch the whole thing, but I’ll try to mine the transcript for sneers.

26:23 –

Simplicity in them you know it runs on a virtual machine who specification Nock [which] fits on a T-shirt and uh you know the goal of the system is to basically take this kind of fundamental mathematical simplicity of Nock and maintain that simplicity all the way to user space so we create something that’s simple and easy to use that’s not a small amount of of work

Holy fucking shit, does this guy really think building your entire software stack on brainfuck makes even a little bit of sense at all?

30:17 – a diatribe about how social media can only get worse and how Facebook was better than myspace because its original users were at the top of the social hierarchy. Obviously, this bodes well for urbit because all of you spending 3 hours of your valuable time listening to this wartime address? You’re the cream of the crop.

~2:00:00 – here he addresses concerns about his political leanings, caricaturing the concern as “oh Yarvin wants to make this a monarchy” and responding by saying “nuh uh, urbit is decentralized.” Absent from all this is any meaningful analysis of how decentralized systems (such as the internet itself) eventually tend to centralized systems under certain incentive structures. Completely devoid of substance.

Nope, actually this used to work but the genius computer scientists at Boeing put the cockpit in a random place around the cabin, thwarting most pilot overwrite attacks.

Lex hasn’t optimized the skill of technical interviewing; he has optimized the skill of simultaneously stroking the interviewee’s and the (implicitly) listener’s ego.

I usually can’t stand radio-style podcasters but these guys are just too good. The way they play off each other is top notch.

Yeah, my initial reaction was “how dare pg not follow his own principles” but then I realized he’s on Twitter, not HN, and when in Rome…