Ryan Conti

thinking and learning about anything 🙂 math & cs @ mit | mle | doing interp stuff in the margins

working with FLAIR @MILA

[Ryan Conti @ContiThinking]

1password is so goated man. one of the best UX pieces of software i interact with

[Ryan Conti @ContiThinking]

@thesephist massive fan of this direction, it's something i've wished for often over the years!

[Eric Zhang @ekzhang1]

not sure if it's just me, but I feel people should report inference perf as
1. Prefill: input tok/s/GPU at ctx len
2. Decode: output tok/s/GPU at ctx len
3. KV cache size

Not total tok/s/GPU, which varies a lot. Feels like marketing numbers, really hard to compare.

[Ryan Conti @ContiThinking]

@karpathy definitely a preferred way to start planning sessions when i have far more thoughts than decisions made about those thoughts…

[Ryan Conti @ContiThinking]

@yacinelearning greasy pizza

[levent @__alpoge__]

hello there the jacobian conjecture is false thanx to my close friend akhil for asking about it and my other close friend fable for working during the world cup final

((1+xy)^3 z + y^2 (1+xy) (4+3xy), y + 3 x (1+xy)^2 z + 3 x y^2 (4+3xy), 2 x - 3 x^2 y - x^3 z): \C^3\to \C^3, has jacobian determinant -2, and sends (0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2) to (-1/4, 0, 0)

[Louise Jensen Duffy (also Ernest Jensen) @LouiseWeebe]

You’re listening to Sisyphus FM, the home of non-stop rock.

[Yoshua Bengio @Yoshua_Bengio]

If the trajectories of AI development observed for several years now continue, it is highly plausible that AI will drastically transform our economies. We must be intentional and make collective, democratic choices, rather than letting market forces play out and risking leaving most citizens behind. https://t.co/ruXHoP6BIR

[Ryan Conti @ContiThinking]

@GoodfireAI every time i see new goodfire work i get excited at this point. some awesome things happening

[Andrew Carr 🤸 @andrew_n_carr]

Backprop

[Justin Skycak @justinskycak]

What looks like instinct is usually accumulated reps.

[Ryan Conti @ContiThinking]

@benno_krojer Was thinking this same thing recently! My Claude has far too many conditionals instructing it to tool call to check the time 😅

[Ryan Conti @ContiThinking]

@RhizoNymph Super cool! Been messing around with this stuff since this Goodfire paper dropped as well, looking forward to what you share :)

[Ryan Conti @ContiThinking]

@MilksandMatcha @tyler_fong_ Congrats Sarah!!

[Noam Brown @polynoamial]

After AlphaGo, the skill of human Go players noticeably improved. I suspect we will see a similar pattern in math.

[Ryan Conti @ContiThinking]

@vvhuang_ ‘twas a very cool post

[Ryan Conti @ContiThinking]

@aamixsh Super cool paper!

[Omar Khattab @lateinteraction]

The bitter free lunch is the idea that if you rely *purely* on scale, you will get the bland and bitter defaults that everyone else is getting.

There's irreducible complexity and irreducible modularity to *your* problem specification, at least if your problem is worth your time.

[Catherine Yeo @catherinehyeo]

Love seeing Naomi Osaka honor the CLRS Algorithms textbook at this year's Met Gala

[Jack Lindsey @Jack_W_Lindsey]

Before limited-releasing Claude Mythos Preview, we investigated its internal mechanisms with interpretability techniques. We found it exhibited notably sophisticated (and often unspoken) strategic thinking and situational awareness, at times in service of unwanted actions. (1/14)