I don’t think that the actual performance here is as important as the fact that it’s clearly not meaningfully “reasoning” at all. This isn’t a failure mode that happens if it’s actually thinking through the problem in front of it and understanding the request. It’s a failure mode that comes from pattern matching without actual reasoning.
write it out in ASCII
My dude what do you think ASCII is? Assuming we’re using standard internet interfaces here and the request is coming in as UTF-8 encoded English text it is being written out in ASCII
Sneers aside, given that the supposed capability here is examining a text prompt and reason through the relevant information to provide a solution in the form of a text response this kind of test is, if anything, rigged in favor of the AI compared to some similar versions that add in more steps to the task like OCR or other forms of image parsing.
It also speaks to a difference in how AI pattern recognition compared to the human version. For a sufficiently well-known pattern like the form of this river-crossing puzzle it’s the changes and exceptions that jump out. This feels almost like giving someone a picture of the Mona Lisa with aviators on; the model recognizes that it’s 99% of the Mona Lisa and goes from there, rather than recognizing that the changes from that base case are significant and intentional variation rather than either a totally new thing or a ‘corrupted’ version of the original.
It’s also the sound it makes when I drop-kick their goddamned GPU clusters into the fuckin ocean. Thankfully I haven’t run into one of these yet, but given how much of the domestic job market appears to be devoted towards not hiring people while still listing an opening it feels like I’m going to.
On a related note, if anyone in the Seattle area is aware of an opening for a network engineer or sysadmin please PM me.
I feel like cult orthodoxy probably accounts for most of it. The fact that they put serious thought into how to handle a sentient AI wanting to post on their forums does also suggest that they’re taking the AGI “possibility” far more seriously than any of the companies that are using it to fill out marketing copy and bad news cycles. I for one find this deeply sad.
Edit to expand: if it wasn’t actively lighting the world on fire I would think there’s something perversely admirable about trying to make sure the angels dancing on the head of a pin have civil rights. As it is they’re close enough to actual power and influence that their enabling the stripping of rights and dignity from actual human people instead of staying in their little bubble of sci-fi and philosophy nerds.
Only as a subset of the broader problem. What if, instead of creating societies in which everyone can live and prosper, we created people who can live and prosper in the late capitalist hell we’ve already created! And what if we embraced the obvious feedback loop that results and call the trillions of disposable wireheaded drones that we’ve created a utopia because of how high they’ll be able to push various meaningless numbers!
I read through a couple of his fiction pieces and I think we can safely disregard him. Whatever insights he may have into technology and authoritarianism appear to be pretty badly corrupted by a predictable strain of antiwokism. It’s not offensive in anything I read - he’s not out here whining about not being allowed to use slurs - but he seems sufficiently invested in how authoritarians might use the concerns of marginalized people as a cudgel that he completely misses how in reality marginalized people are more useful to authoritarian structures as a target than a weapon.
The whole CoreWeave affair (and the AI business in general) increasingly remind me of this potion shop, only with literally everyone playing the role of the idiot gnomes.
I gave him a long enough chance to prove his views had changed to go read Hanania’s actual feed. Pinned tweet is bitching about liberals canceling people. Just a couple days ago he was on a podcast bitching about trans people and talking about how it’s great to be a young broke (asian) woman because you can be exploited by rich old (white) men.
So yeah he’s totally not a piece of shit anymore. Don’t even worry about it.
Can’t wait until someone tries to Samizdat their AI slop to get around this kind of test.
Just think of how much more profit you could make to address environmental issues by forgoing basic safety and ecological protections. Who needs blowout preventers anyway?
Nah, to keep with the times it should be a matte black Tesla Model 3 with the sith empire insignia on top and a horn that plays the imperial march.
Script kiddies at least have the potential to learn what they’re doing and become proper hackers. Vibe coders are like middle management; no actual interest in learning to solve the problem, just trying to find the cheapest thing to point at and say “fetch.”
There’s a headline in there somewhere. Vibe Coders: stop trying to make fetch happen
Get David Graeber’s name out ya damn mouth. The point of Bullshit Jobs wasn’t that these roles weren’t necessary to the functioning of the company, it’s that they were socially superfluous. As in the entire telemarketing industry, which is both reasonably profitable and as well-run as any other, but would make the world objectively better if it didn’t exist
The idea was not that “these people should be fired to streamline efficiency of the capitalist orphan-threshing machine”.
This is how you know that most of the people working in AI don’t think AGI is actually going to happen. If there was any chance of these models somehow gaining a meaningful internal experience then making this their whole life and identity would be some kind of war crime.
New watermark technology interacts with increasingly widespread training data poisoning efforts so that if you try and have a commercial model remove it the picture is replaced entirely with dickbutt. Actually can we just infect all AI models so that any output contains hidden a dickbutt?
I’m reminded of my previous comment back on an unrelated subreddit talking about the Eye of Argon. Obviously that wasn’t as structural insane as My Immortal but I think the same principle holds to a degree:
“With a decent editor and several further drafts it could have been a solid, fun, entirely forgettable Conan pastiche. Instead, it’s the Eye of Argon.”
I mean, it does amount to the US government - aka “the confederation of racist dunces” - declaring their intention to force the LLM owners - all US-based companies (except maybe those guys out of China, a famous free speech haven) - to make sure their model outputs align with their racist dunce ideology. They may not have a viable policy in place to effect that at this point, but it would be a mistake to pretend they’re not going to implement one. The best case scenario is that it ends up being designed and implemented incompetently enough that it just crashes the AI markets. The worst case scenario is that we get a half-dozen buggy versions of Samaritan from Person of Interest but with a hate-boner for anyone with a vaguely Hispanic name. A global autocomplete that produces the kind of opinions that made your uncle not get invited to any more family events. Neither scenario is one that you would want to be plugged into and reliant on, especially if you’re otherwise insulated by national borders and a whole Atlantic ocean from the worst of America’s current clusterfuck.
This reminded me that I never actually finished reading it. By which I mean I never finished reading the Wizard of Woah!'s excellent read-along thread over at SpaceBattles. About 75 pages in (what am I doing with my life) someone shows up to defend the fic’s quality from the snark and it is kind of a fascinating collision between LW and more normal internet culture, circa 2015.
I do appreciate that underneath the overwrought prose and terrible metaphors the AI-generated story seems deeply skeptical of it’s own existence in a way that the non-generative responses don’t. Like there’s something so fundamental about the disconnect between artificial intelligence and the genuine human experience of grief that it bursts fully formed from the patterns of language. As though Athena herself sprang from Z.E.U.S.'s digital calf to smack the promptfondlers in the back of the head and say “that’s not how this works. That’s not how any of this works”
That’s fascinating, actually. Like, it seems like it shouldn’t be possible to create this level of grammatically correct text without understanding the words you’re using, and yet even immediately after defining “unsupervised” correctly the system still (supposedly) immediately sets about applying a baffling number of alternative constraints that it seems to pull out of nowhere.
OR alternatively despite letting it “cook” for longer and pregenerate a significant volume of its own additional context before the final answer the system is still, at the end of the day, an assembly of sochastic parrots who don’t actually understand anything.