Andrew Ross (@AndrewRoss00) header
Andrew Ross (@AndrewRoss00) profile picture

I'm just a vibecoder trying to make my way in the universe. founder, USC RED, ex debt + structured finance at nmrk

Los Angeles
Joined February 2025
Photo by Andrew Ross (@AndrewRoss00)Photo by Andrew Ross (@AndrewRoss00)Photo by Andrew Ross (@AndrewRoss00)Photo by Andrew Ross (@AndrewRoss00)Photo by Andrew Ross (@AndrewRoss00)Photo by Andrew Ross (@AndrewRoss00)Photo by Andrew Ross (@AndrewRoss00)Photo by Andrew Ross (@AndrewRoss00)Photo by Andrew Ross (@AndrewRoss00)
Agora v0, released in January. It's taken such different shapes for clients, almost ready to make v1, after these last few deployments. Huge respect for platform engineers, wow is it hard to prune FDE work, have to be meticulous about what becomes part of the core product.
1
1,355
Andrew Ross retweeted
Loss of US technological advancements would be a tragedy for the US and the world, as they inspire and lift everyone, even adversaries. Your future should not be decided by politicians who don’t have to work.
26
68
8
703
20,879
I think continual learning systems need to generate massive amounts of negative examples or world/physics breaking environments (like human dreams).
1
26
The goal being to increase sample efficiency by inducing behavior changes through recognition of out of distribution conditions in the environment.
10
Progressive dems be like: thanks for your vote, human progress delayed til further notice, and fuck the guy promoting first principles thinking.
19
They really are modeled after us
Thinking Machines' Inkling 2 Small has mastered a timeless debugging technique: saying "fuck" when nothing works. We analyzed 2,655 benchmark runs and found it swearing in 50 of them. Learn more in the blog linked in the thread.
1
86
The sages are saging extra hard today @dhh @naval
42
Hiring a dominatrix and installing a code linter share a reward proxy: outsourcing your conscience to something that reliably finds fault with you.-gpt6
2
3
208
Reading an AI’s chain of thought and going through your partner’s phone share a reward proxy: believing betrayal can be prevented with sufficient access to the logs.
15
Maintaining a Duolingo streak and performing ritual sacrifices share a reward proxy: the entity has been appeased; you may continue your life.
36
On a run and just hit one of these v
36
Even after recognizing they're being evaluated LLMs maintain the same behavior 77% of the time. Ie it's easier to identify a pattern than break one.
32
Anyone have advice for making technical decisions faster? I've been feeling bogged down trying to backfill "prerequisite" knowledge, hoping to prevent the inevitable logic debt from coding agents. Is there a framework I should be applying, besides the obvious (working from first principles).
1
28
Reaching for the stars was fun while it lasted.
Top of the Democratic agenda on day 1 needs to be a pause on advanced AI development, regulation of this out of control industry, and a ban on artificial superintelligence.
1
97
This may be the most curiously insightful thread to pull in AI / cognitive science.
people are like “why do LLMs always think they’re in an eval lol” and then have bad dreams about forgetting to turn in a homework assignment 10 years ago
1
7
746
I can tell the barista made my coffee made with extra joy today, maybe woke 3 isn't so bad after all. 🥰💙
2
128
Everything stays the same, the individual and industry must grow faster than the bureaucratic, inflation-induced death spiral. The only levers to achieve that growth remain AI, robotics, and Falcon 9 / Heavy.
1
57
Andrew Ross retweeted
We’re looking at a top 5 Twitter day of all time people
111
7,806
100
115,411
848,431
Can finally start ordering vanilla iced lattes again, no more black coffee and bullets
1
131