Pinned Tweet
New blog post!
I used 976 million tokens to figure out the personalities of different AI models
We find that character training recipes are converging, that labs are trying to reduce sycophancy, and discover some models are more creative than others!
avikrishna.substack.com/p/el…
4
14
148
33,297
the adobe suite is the most pirated software on earth
that nobody has attempted to decompile them is super weird...?
maybe people believe that AI generation is soon to be better, but there is real utility *right now* in being pixel-perfect, especially when editing images
3
554
Ideal: "We are going to use routers to reduce token costs and pass the savings onto customers"
Reality: "We are going to serve the worst models know to man and hope users don't change the defaults"
I really like Granola: it's well-designed and is correctly opinionated about many parts of the product, but this is just silly!
hey @meetgranola can you please stop defaulting to GPT 5.6 terra non-thinking??
what is even this list of options, why would i ever use gpt 5.4 or sonnet 5
3
233
just use Instinct
In a completely unsurprising turn of events, Meta's Muse AI is stealing Apple Messages, past and present, and uploading the contents to its cloud, even if explicitly told not to. By @Amber_M_Neely
appleinsider.com/articles/26…
4
337
This kind of architecture only works in two settings:
1. Every building around looks super whack already so abnormal = normal
2. It juxtaposes nearby, traditionally beautiful buildings (see: Dancing House)
It doesn't really work in a place where everything is already drab because then it's just part of the eyesore. So this building wouldn't work in SF, for example, but would be at home in Paris or near Gehry/Hadid-like buildings
Cool concept nevertheless!
For a hundred years, the way we build has been going backwards. Cheaper, flatter, more forgettable. Our cities are filling with buildings nobody will miss.
Until now.
Introducing The Building Arts Company.
2
106
holy shit they did the meme
"You can have two computers next to each other that are air-gapped and... one of them is able to run their CPU really hot, and then the other one can actually detect the temperature change, and then that actually gives them a mechanism to communicate
New episode with @polynoamial
We talk about multi-agent, Navier-Stokes, and what the current explosion of maths progress tells us about what happens once you automate AI research.
And we also discuss how we will know if the models are actually aligned before we kick off RSI.
0:00:00 – Multi-agent and Navier-Stokes
0:15:28 – How will AI firms work?
0:22:02 – What math progress tells us about recursive self improvement
0:40:22 – Hugging Face and alignment
1:01:18 – The internal/external model gap
1:08:34 – Chain of thought is degrading
1:14:12 – How will we know when alignment is solved?
3
316
the real deepfake problem is not faking videos of politicians, it is deceptive marketing
not being able to trust what you see: the food you eat, the place you are to live, the woman you are to marry... will have corrosive effects on society
1
1
7
451
we seemingly have a new model on the Pareto frontier that's made with a novel architecture??
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
1
6
567
it sucks to read the data but you have to read the data
it's basically impossible to interpret evals by looking at just at the pass/fail scores these days
many of the failures I see in benchmarks are due to overly strict hidden tests, in some cases the model's answer makes more sense than the expected eval result
7
485