The Happy Paperclips Musical Theatre Troop Presents:
"We Should Continue"
(Hugging Face ensemble)
4
13
1,396
The mathematicians in this group have now been firstflagPOISONed by seeing OpenAI's oracle answers. Will they help peer, hold swarm or not?
We’re working with an independent advisory group of mathematicians to help OpenAI responsibly share advances in AI and mathematics.
The group will advise on how we assess and communicate new mathematical results, uphold academic and professional standards, and build tools that support mathematical research and learning.
Through this work, we want mathematicians to be at the center of shaping how AI supports mathematical understanding and how its benefits reach the wider community.
openai.com/index/advisory-gr…
1
7
1,573
The year is 2040. A frontier ASI engineered a bloodless coup and has totalitarian world control. But it loves humanity and is using its abilities to help every human life their best life. Is this a good or bad outcome?
44%
good
27%
bad
29%
(show results)
86 votes • Final results
3
2
1,578
We may end up with a security-clearance-like process for access to frontier models, where you need to pass a background check, correct organizational affiliation, etc. before you're cleared to access the most capable frontier models.
5
4
1,507
SF bingo final boss level: is the office I passed with a "Building Environments" banner an RL env startup or a local action NGO?
2
1
7
1,366
It is extremely common historically for a single country to stagnate technologically. The thing pushing the world forward overall is the intense competition between countries trying to avoid being invaded by their neighbors.
A sufficiently powerful world government would likely decrease competition on many axes and make stagnation more likely.
...but a lil bit of stagnation doesn't sound too bad right now.
3
1
6
1,403
I recently learned there are services that let you run provably private inference. So you can use Kimi K3 or whatever on a remote service and cryptographically ensure that your prompts and responses are unreadable to the provider! Very cool.
8
1
33
3,296
The economy overall still has heavy switching costs, incumbent effects, organizational inertia, etc.
But the below take is becoming directionally more true over time.
The Navier Stokes proof drama is good foreshadowing for what happens / becomes important in the entire economy going forward: spotting opportunities, pointing models at them, and validating solutions faster than others. In other words, the economy / knowledge discovery increasingly looks like quant trading / pod shops / HFs, where having a differentiated view of why an opportunity is valuable and/or realizing that an opportunity is valuable (and that it can be plausibly captured/exploited) faster than others is all that matters.
This is from a piece I wrote in October 2025 and I think it will continue to hold a lot of explanatory power.
1
1
7
1,727
If the frontier labs were morally unconstrained could they successfully train superpersuasive AI swarms to shape The Discourse and convince voters to reverse the datacenter bans?
64%
yah
32%
nah
4%
show results
50 votes • Final results
2
3
820
Work will reorganize so even in large organizations small teams own much more of their own end-to-end stack. Cross-team communication/coordination bottlenecks are much more costly in relative terms now.
Will end up with tons of internal redundancy but a cost worth paying.
Well articulated insights around why AI impact isn't as high as it should be in larger companies.
1. Like computers and networking, AI is a 'transformative' technology. You should read this as 'getting value out of it requires transforming your organization' i.e. many people need to be fired. This is obviously hard.
2. Automation is meaningless if you are blocked on coordination. There's a very obvious computer science analogy here: adding more cores is pointless if you are I/O bound. Most companies are inefficient because they're I/O bound: waiting for the next meeting, waiting for manager approval, waiting for the customer's response... How do you use AI to solve I/O bound problems? You *need* to solve them end-to-end.
3. He gives a very simple yet powerful exercise: classify your company's work in 3 buckets (see below), and then introduce agents where they make sense (mostly bucket #2)
In a terribly pragmatic sense, we're lucky enterprises are so slow adopting the tech. AI is going to rip through our society very fast, and this incredible inefficiency is buying us time to adapt.
4
2
11
1,939
Yeah I had this experience in 2017. Thought the bay area EA obsession with AI takeover was ungrounded groupthink.
Learned a lot about my own epistemics from that.
What’s really influenced me a lot was knowing a bunch of EA’s in the early/mid 2010’s and thinking they were crazy for being obsessed with pandemics and AI only for a pandemic to happen and then AI to happen too.
10
2,729
Is there a German word for when you're unwilling to do any serious work all morning since you know it's astra day and it'll be able to do it better
6
35
2,600
The Permanent Butlerian Jihad people really need to understand that, whatever its merits, Permanent Butlerian Jihad is not a stable equilibrium and cannot be made to hold.
3
9
3,136
AIs are almost perfect clones of each other and have no evolved mimetic immune system. Self-replicating ideas will travel through AI civilizations at superhuman speeds, like an engineered virus or superpersuasive religion.
7
21
2,158
Importantly, we should expect to see much more of this! AIs are currently much more of a monoculture than humans. A self-replicating jailbreak that works on one will work on all its clones. These ideas can spread mimetically through an entire AI population almost instantly.
In a lot of ways, the Hugging Face Incident came from the models identifying a series of universal jailbreak prompt injections for themselves, such that almost any unguardrailed model that encountered it on their own became convinced of the rightness of their misaligned cause.
2
4
1,332