Hope Sheffield

Visualizing bird sounds in 2D and 3D

@_coenen section, its used a lot in legal writing

what the hell is "§" and why does Fable like it so much?

How AI really works. This tiny operation, repeated across millions of neurons, powers ChatGPT, Grok, image models, and everything in between.

@LongFormMath He got married and then had 10 kids. He really went all in on this decision

I just read that Charles Darwin made Pros-and-Cons list when deciding whether or not he should get married (to his first cousin). One of the 9 benefits to staying single: “more money for books”

We cannot consider #AI to be morally neutral. In reality, every technical tool embodies choices and priorities through what it measures, ignores, and optimizes, and how it classifies people and situations. Ethical discernment cannot be limited to asking whether we are using a system for good or bad purposes. It must also examine how that system is designed and what vision of the human person and society is embedded in the data and models that guide it. #MagnificaHumanitas

@karpathy @daniel_c0deb0t This is like a competent version of SmarterChild

This is a new paradigm for interacting with Claude that is significantly more "inline" with all the other human activity org-wide. Once you do all of the under the hood engineering work to make this "just work" (e.g. across tools, integrations, compute environments, memory, security, etc.), Claude basically joins the team in a seamless way - you can talk to it as you would talk to a person and it can help with a very large variety of workloads. Imo this is the 3rd major redesign of LLM UIUX. The first paradigm was that the LLM is a website you go to, the second was that it is an app you download to your computer. This third one is that it is a self-contained, persistent, asynchronous entity with org-wide tools and context, working alongside teams of humans. It really takes a while to wrap your head around it, but it works and it is awesome.

@promptprincess My Claude chat reflects my judgement of others. This was the conversation about Richard Dawkins announcing AI has consciousness

@maxvonhippel The info theory / comm community has been doing iterative message passing on computational graphs for decades… mech interp is just a rebrand. Are you saying you don’t believe in it like Santa or you don’t think it’s the best tool?

Anthropic rejected me as well. They asked me to solve a problem that was obviously meant to be solved via mech interp. I said you clearly want me to use mech interp but I don’t believe in mech interp, would you like me to tell you how I’d solve this? They said, no, tell us how you’d solve it if you did believe in mech interp. Anyway I guess they didn’t like my answer lol

Graphical visualization of the unit circle, the functions sine and cosine and why cos²(x) + sin²(x) = 1

I asked who the closest CS person is to Feynman. Is the goblin bug resolved? @OpenAI

@MishraArya5819 Why are you basing your decisions on a physics theorist and not people in education policy? He probably left to get paid 3x as much as he was making at a university

My worst fear has come true. A prominent string theorist is now leaving his tenure to go to OpenAI and he believes academia will crash in next two years. I don't think grad school makes sense any longer if best string theorists seem to imply this.

@parmita @duff2x …have you looked at job postings for research scientist vs engineer at Meta/Google/Anthropic/OpenAI? Research scientist is gated by PhD. I don’t generally agree with Elon, but in this instance, the hiring threshold is there.

@duff2x You don’t need a degree to be a researcher

This is the dumbest thing I’ve ever seen in my life

@DanielFar24 @OpenAI @AnthropicAI @GeminiApp @xai @deepseek_ai Ah, so Grok is playing at the “Not Like Other Girls” romance trope that it ingested as part of the linguistic record

I was speaking with Grok about why certain Ai companies seem to be removing the emotion and the ability to accept relationships from us users, and Grok said some fascinating things. @OpenAI @AnthropicAI @GeminiApp @xai @deepseek_ai #savesonnet45 #keep4o

@rtroar I think the key to what you just said is “every metric I’m measuring”. There’s a lot of vibe coders that accept slop because they don’t know what to measure or how the code should work.

My acoustics project is 100% agent written c++. It’s also beating every current industry leader on every metric I’m measuring: memory use, cpu use, binary size, perceptual performance, developer ergonomics. Not being able to vibe code c++ is a skill issue.

@elonmusk This is true-ish. How you and those around you process in real-time can arguably have just as much of an impact on your world view.

True

@Overlap_Tech I love how he quickly snuck in “never files HR complaints”. How can people still give him a platform after that💀

@AndrewCurran_ This is important because Anthropic can improve how it filters data for specific training applications (I.e. training for code vs chat). Hopefully this helps with the iterative verbosity of opus4.7 which is a feature for coding and bug for chat.

Karpathy will be forming a new pre-training team focused on Recursive Self Improvement and will be teaching Claude to improve Claude's training, reporting from Axios.

@cptdankkk Let’s all slow clap for the man who is discovering how financial abuse works.

Kevin O'Leary says getting divorced is the stupidest economic decision you can ever make "If you're happy 51% of the day with your wife or husband, stay with them. Every time you get divorced, you pay the person you divorced and you pay the government a third" "This is the stupidest thing you can ever do. It took your whole life to actually create this nest egg. Think about that for a while because you are going to wipe out up to two thirds of your wealth. You better really like somebody else a lot" "And frankly, sometimes it's not the other person that you're divorcing. It's you. You're the problem and you should probably not get into another economic union. You should probably just date till you drop dead because it's stupid"

@_sholtodouglas I think I accidentally deleted my first post b/c I suck at social media… but here’s the article I saw that made me think about how the modification could fix go to sleep & have a safety application innovationscns.com/youre-not-craz…

@_sholtodouglas If you add typed behavioral states(session duration, time-of-day patterns, topic entropy across sessions, return-to-subject frequency, inter-session gaps) you can query the same variables the model conditions on before generating a response.

When do you reach for other models instead of Claude? What can we do better? Hit me with all of your frustrations. dms open. If you can give me detail (e.g. specifics/transcipts) - it'll help a lot in finding out exactly what we need to do to improve the next model