My goal is to create knowledge that helps keep civilization afloat and growing 馃嚚馃嚘馃嚚馃嚘馃嚚馃嚘馃嚚馃嚘馃嚚馃嚘馃嚚馃嚘馃嚚馃嚘馃嚚馃嚘馃嚚馃嚘馃嚚馃嚘
O'Neill Cylinder 7b
Joined November 2022
- Tweets26K
- Following223
- Followers9.8K
- Likes11K
New research from @hlyons_bd
The model that underlies this report is incredible. We can discover and understand things about the nuclear fuel industry that nobody else can.
TIL that the Soviets used fucking RTGs to power a network of Arctic lighthouses and there was a big international effort to secure the radioactive materials after the fall of the USSR
Fucking lmao
ChatGPT deskslop app is completely unusable.
Can't go more than a few mins without it crashing and stopping all of my agents. Kinda incredible that they shipped this.
NIMBY's are one thing, but you know who I really hate?
People who think they can tell you what you can or cannot put in your OWN backyard.
A quirk I have found with gpt-6 is that it's quite enthusiastic about duck hunting
I ask it questions in a neutral tone and all of a sudden it's like "Nice! Wood ducks are my favorite ducks! Does it have a nice layer of yellow fat beneath the skin? Here are some recipes!"
I am a humble believer in the fact that you can just explain things. And this is a massive, glorious whitepill.
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
They don't want you to know this but the hermes agents are free, you can just take them. I have 200 hermes agents at home.
Labs will talk about "alignment" and then build products that intentionally obfuscate to the user how they work.
You ever try to talk to a Dot about how it works? It's prompted to hide it from you.
How you gonna align AI to humanity when you're not even close to aligned with me
