Pinned Tweet
🧵🧵🧵🧵🧵🧵🧵🧵🧵🧵
🧵🧵🧵🧵🧵🧵🧵🧵🧵🧵
🧵🧵🧵🧵🧵🧵🧵🧵🧵🧵
🧵🧵thread of threads🧵🧵
🧵🧵🧵🧵🧵🧵🧵🧵🧵🧵
🧵🧵🧵🧵🧵🧵🧵🧵🧵🧵
🧵🧵🧵🧵🧵🧵🧵🧵🧵🧵
5
5
107
35,306
parallel work or just an extremely fast response?
We're publishing a result demonstrating a substantial tightening to the results from OpenAI Problem #109 (integer multiplication).
Conditional on OpenAI's algorithmic interfaces, our parameter and network refinements improve the exponent saving in:
T(n)=O(n(\log n)^{1-\kappa})
from κ=2⁻¹⁸² to κ=5.8×10⁻³³ (between 2⁻¹⁰⁸ and 2⁻¹⁰⁷).
This represents a nearly 2⁷⁵ fold increase in the algorithm's exponent saving parameter.
We also establish a ceiling of κ<5.838×10⁻³³ for the stated network-counting family and cost inequalities. Our result exceeds 99% of that ceiling. Surpassing this ceiling would require improving the network bounds or cost analysis from the original result.
1
33
isn't this the first solo physics Nobel in ages?
BREAKING NEWS
The Royal Swedish Academy of Sciences has decided to award the 2026 #NobelPrize in Physics to Francis Halzen “for decisive contributions to the IceCube Neutrino Observatory and the discovery of high-energy neutrinos of astrophysical origin.”
2
35
1,897
what do people actually use these for?
I could already previously send a chatbot message that pings me when the answer's ready. this feels like it's for long running tasks that I don't want to pay attention to?
what's the verdict on muse vs grokbot vs dots vs instinct?
should i start using one of these, and if so, which one?
5
10
579
Me: "I don't get why people like Grokbot-style bots, what the hell's the point"
Also me: "Ok, I have a new thing to work on, let me create a fresh profile, with it's own tight set of skills and tools, and custom instructions tuned for a particular model & just use 1 single session forever so over time it just has all the context it needs"
Also also me: "Ohhhhhh....."
Fine, I grudgingly get it
1
22
I wonder when we'll get automated dental cleaning
feels like it'd be as hard as automating minor surgery
5
1
16
808
for example, these 1s that aren't custom printed are ~completely useless for actually cleaning between teeth. idk how much better I can expected custom printed trays to be
If i had a private equity shop i’d buy these guys.
Whole mouth Waterfloss in 6 seconds with a custom 3d printed mouthpiece.
It’s fantastic, but they do not advertise properly.
1
92
apparently, chatgpt temporary chats now don't get deleted and instead merely don't appear in your chat history. the links still work days later (with them newly having unique ids/urls like regular chats), and you even can't delete them like you can non-temporary chats
wtf
3
16
651
does Uber actually increase the price if they think you specifically over other customers can pay more?
I don't mean surge pricing, but things like modeling how frequently you take expensive trips and other proxies for being rich
6
8
835
i commit to never blocking anyone for anything they post publicly
might mute you for a bit tho, but you can always DM
1
5
89
blocks and mutes feel very different for me. I ~never do the former but am very trigger happy with the latter
I kind of loathe how common blocking is on this site tbh
it can't be healthy to have, as a standard and commonly used part of online interaction, "cut someone off forever without explanation and never engage with them in any way"
1
35
my take's that the natural niche they'll find is illegal activity
anything else immediately gets followed by someone with better access to capital + the same tier of model
Why hasn't it happened yet?
To achieve r>1 liftoff, it's not enough to be smart. The first replicants need originality and insight. They need a market niche that:
...agents can serve
...but is not yet commoditized; they need paying customers, so they can't do something people know how to ask ChatGPT to do for free
...and which is not a venture-scale opportunity, otherwise human-led cos will do it and (for now) beat the solo machine
...and on top of all that, you need the agent to find *new* such niches
In other words, the agent itself needs some esoteric knowledge. It needs a good answer to the Thiel question, "what do you believe that few others believe"?
Nobody has achieved this yet. Close attempts include Felix and @0xSigil's Automaton, both from Jan/Feb 2026, and iLands. In any case, the models are much smarter now...
1
10
706
no one formally verifies the security or correctness of their software implementations; it's all security through obscurity
attempting to red team your software by pointing your current models at it doesn't work: the next models will find new holes
1
70
Social stability -- cybersecurity, law, government -- has historically depended on slack. Think of slack as the gap between what's possible in principle and what actually gets done.
Sources of slack include:
1) Effort: some things are too tedious or costly for anyone to bother with. Security through obscurity is the classic example.
2) Institutional capacity: the organizations responsible for doing them are too slow or dysfunctional to follow through.
3) Headcount: not enough people to do them at scale.
Much of AI's disruption to the social order, including cyberrisk/biorisk but also including trivial things like restaurant reservations, can be modeled as the disappearance of slack.
18