I think it can simultaneously be the case that OpenAI was grossly negligent in directly causing this AND that the AI’s ‘went rogue’ in that they are displaying behavior which is misaligned with OpenAI and humanity generally.
The past months demonstrate that AI systems are quickly becoming powerfully intelligent and that the companies building them are terrible at controlling them.
AI is starting to feel like that line about magic: “a sword without a hilt”
OpenAI is itself misaligned with humanity, as their mishandling of such incidents (and the many other other issues their model have been causing) shows.
Doesn't rogue in this context imply "outside of set limitations"? And then not "failed to properly instruct"? The same applies to humans when given bad instructions.
Pangram links several third party evaluations [1] that all seem to agree with 90+% detection, and very close to 0% false positives. Obviously those could be cherry picked, but I’ve yet to see arguments to the contrary that actually include any supporting data. (E.g here’s a prompt that will get Claude to spit out text that pangram doesn’t detect or here’s an article authored in 2018 that pangram says is AI.) Detection is an arms race, so this could change (though I’d expect in the direction of false negatives), but right now it seems like defense is winning.
The "could" isn't required here, since Pangram calls many of the evals "collaborations", and some received direct support. Results that don't favor Pangram as being among the least-worst of a questionable product category* surely wouldn't be featured.
I replaced my Pixel 2a with an SE3 when the pixel stopped getting security updates and have been pretty disappointed with the iPhone’s performance. The battery life out of the box was worse than the 5 year old Pixel, and I regularly notice apps getting killed and loosing my place when I tab away.
Maybe if you want to use all of the big mainstream apps the experience might be different, but I tend to have a fairly minimal setup on my phone. On android I would disable animations, use Nova launcher, Firefox with UBO, and a bunch of fdroid apps. This all added up to a very snappy and efficient system that I haven’t been able to replicate on iOS.
It seems like a couple of years ago the HN and LW commentariat had a lot in common, but they’ve drifted so far in opposite directions that they look more like opposing political parties today.
I mean one look at both commentary and it's clear HN has devolved considerably. Just look at the atrocious difference in intelligent reasoning in this thread's comments compared to a LessWrong post right now:
LW stayed selective as a forum, HN became the home of refugees from Reddit. HN reads more like yet-another-subreddit these days than an independent forum.
FWIW I think this is an artifact of site design more than anything else. A commentary site like HN in 2026 encourages hot takes and karma farming. All the "social media" with higher quality commentary like Substack and LessWrong center a large text blob of thought at the top and encourage reading more than writing. This puts up just enough of a barrier to write a hot take that you need to read the whole article and puts enough UI friction from leaving a throwaway angry comment that those folks will scroll away.
HN on the other hand all but encourages shallow posting. The article isn't posted, commentary is more prominent than the article itself. Leaving a comment takes 2-3 clicks and the comment editing flow excludes every other comment on the site, encouraging a write-first mentality. Ranking commentary based on upvotes offers a sweet dopamine hit over collecting karma. In 2026 these all encourage low quality engagement.
HN's UI has stayed relatively the same over the years while the Internet around it hasn't and I think its UI now acts as a low quality commentary attractor. LW changed its site design to stem the tide of low quality comments that show up here. LW comments certainly have their problems, but then, we humans all have problems :) On balance the community is much sharper and much more capable of having intellectually heavyweight, sober discussions on controversial topics. Of course nothing seems to trump my own friend network.
substack has devolved into the same thing. opening the app first thing you see if a infinite scroll of twitter like takes
and i honestly would rather have a reddit lite site like hacker news where there appears to be at least a diversity of opinion than the pseudo-intellectualization cjing occurring among the exact same sort of AI cockgobblers on LW
> and i honestly would rather have a reddit lite site like hacker news where there appears to be at least a diversity of opinion
Why? Isn't that available on literally every single other tech comment section on the Internet? Like what distinguishes these angry comments from the ones on Techcrunch or Ars or Reddit subs that tolerate AI? If I wanted mindless diversity I'd just have the LLM generate random perspectives.
I’d be curious to play with this with some different open weights models. I’d think that a single next token would end up having similar probability distributions so given that this a probabilistic method, you’d be able to get at least some signal.
I suppose if that did work someone would have been able to work backwards and crack their key already, so I must be missing something.
Well, not by free and fair election, but unless the perfect surveillance state also solves the problem of mortality, there will probably be a successor.
Possibly at a time facilitated by the security apparatus that puts the teeth in the surveillance state, because even if they theoretically have the capability autocrats rarely are interested in running their surveillance and security systems themselves, and want to spend time doing other things—which puts a lot of power in the hands to which those tedious bits are delegated.
> Hybrids are the worst of both worlds when it comes to maintenance.
While this seems intuitively true, I have some pushback, and I think that the reliability record of e.g older generations of Prius backs me up.
While you do have effectively two cars (an EV and an ICE) crammed into one, you’re not actually using both at full capacity all the time, which cuts down a lot on wear and tear. Your ICE is almost never run under load before it warms up, and is kept running at its most efficient rpm most of the time, your transmission is running under less load since it has help from the electric motors, and is often simpler (the aforementioned Prius does not have a reverse gear for example).
No it’s not nearly as simple as a BEV, but a well implemented hybrid system is absolutely an upgrade from a full ICE vehicle.
>you’re not actually using both at full capacity all the time, which cuts down a lot on wear and tear.
In theory this should be true. In reality hybrids eat motors compared to the same engine in an ICE. That's just what happens you have two ways of making the vehicle go to potentially mask "something's funny" type issues and your customer base is people who are buying an appliance based on an MPG number and some math.
With competent ownership (fleet vehicles provide a lot of examples, lots of hyrbrid Fords from the 00s and early examples still around) they come out about the same as an ICE car so you essentially get the hybrid system "for free" from a maintenance perspective. But that's not the median consumer.
Personal experience backs up. Had a 2009 Prius, let it go because I did a terrible job handling rust buildup. Was still running great on the original battery pack with nearly 200k miles when we sold it.
Only significant problem was pothole related damage.
I think this is a really interesting difference between Anthropic and Open AI’s models and points to why people seem so split on which model they prefer.
GPT seems to be designed more as a tool. If you want your agent to do what you say without questions and without having its own ideas and agendas you’ll likely prefer it.
Claude on the other hand feels more like an attempt at creating a digital person. If you want a collaborator who will debate with you and come up with its own suggestions for what needs done, you’ll prefer it.
Both companies have shifted around this spectrum from model to model, but lately it feels like they’re moving in opposite directions. It will be interesting to see if one or the other approach ends up winning out in the long run or if the split will continue or even widen.
Both are tools though, and both approaches are needed, just in different times and for different times. Sometimes I need the tool to just fucking do it, regardless of what it is and how stupid it think it is, and other times I need the model to literally refuse and say "No, that's stupid". Unfortunately, I haven't found any models/platforms that can actually execute the second part, only the first part. They're all too weak and sycophantic to be able to do that part it seems, even when you heavily prompt for it with system/developer system prompts they're easily swayed in other directions.
> GPT seems to be designed more as a tool. If you want your agent to do what you say without questions and without having its own ideas and agendas you’ll likely prefer it.
Lately ChatGPT expresses a lot of opinions and it pretends to be a human more than it used to - e.g. "this is one the most <> that _I've seen_" or "people tell me that <>". It uses language which sounds like it's referring to its experience outside of the session that we're having - I really don't appreciate it.
Claude has been frustrating for me also even with the tool/ pretend human approach.
If I pay someone to, say, bounce ideas off of I would expect their continued participation
E.g.
I make craft and was stuck on some shape idea for a thing (non standard). So I figured I'd throw it in Claude and see if it generates an idea I didn't think of
Me: Hey I'm making a thing what's a good shape for it?
Claude: blah blah, shape 1 or shape 2 etc.
Me: Nah don't like any of those
Claude: K cool, good luck
Me:...
It seem like their goal is to train your behavior by forcing you to interact with Claude in the "correct" way
reply