Max Cembalest
@maxcembalest
Followers
316
Following
2K
Media
94
Statuses
334
Math and AI, prev @nomic_ai @itsArthurAI, MS @HarvardGSAS, teacher @ALAcademy
nyc
Joined September 2014
calling this my chimera hypothesis:.1) openai is going to have the best methods for processing inputs in a user-specific way.2) google is going to have the most affordable compute.3) anthropic is going be the best at coding and generally writing the most careful/refined outputs.
0
0
1
RT @the_startup_eng: If you didn’t stick it out with a magikarp to get a gyarados you’re ngmi.
0
1
0
openai docs: "As of July 2024, use gpt-4o-mini in place of GPT-3.5 Turbo, as it is cheaper, more capable, multimodal, and just as fast. GPT-3.5 Turbo is still available for use in the API.". any mention of 3.5 turbo (e.g. item 4) means it's suspect, let alone possibly LLM-written.
McKinsey's new report on AI agents shows the same mindset I see in many firms: a focus on making small, obsolete models do basic work (look at their suggested models!) rather than realizing that smarter models can do higher-end work (and those models are getting cheaper & better)
0
1
1
- written by claude 4 opus giving me a grothendieck perspective.
0
0
1
the arrow before the object: are you sure the individual is truly primitive? or does individuation emerge from morphisms failing to be isomorphisms?.
1
0
1
From “SFT Memorizes, RL Generalizes”. Scaling up compute allocated explicitly to verification during reinforcement learning improves generalization . I like this chart, it’s a nice choice of axes here
1
0
0
eh actually words like "intertwined" are more of a core idea about important complex functions than the complex numbers themselves. but I just think we should use terms that characterize things more naturally in English, "complex" and "imaginary" have too much damn baggage.
1
0
0
I think the names intertwined numbers, woven numbers, fabric numbers, or even destined numbers, despite being a bit over-the-top, would do a better job than "complex numbers" at building intuitions in English for Cauchy Riemann equations and analytic/holomorphic functions.
1
0
0
"complex numbers" is such an awful name for most uniniated English-speakers it sounds like there's a branch of math called "difficult numbers". Gauss chose the name "complexus" to refer to numbers with multiple parts (a+ ib) but that doesn't mean this is the best name in English.
1
0
3
From early 20th-century text processing to modern AI advancements since the advent of ChatGPT, quality filler text such as 'The quick brown fox jumps over the lazy dog' has been crucial for enabling scientific progress. I humbly propose this image as a multimodal equivalent
0
0
1
Visualizing MNIST Training Dynamics with Nomic Atlas. Training an image classifier, collecting embeddings each epoch, lets us see how the model evolves and gets better at contrasting the images in embedding space. My first contribution to the UMAP docs 🙂
1
3
15
Here's an interactive @nomic_ai Atlas map of this dataset: common themes include standards, export controls, and education.
🚨 NEW TOOL FROM @IFP:. The White House asked for input on its AI Action Plan & was flooded with >10,000 responses. To help, we built — a searchable database of what America thinks the gov't should do on AI. Naturally, we used AI to make it. 🧵 (1/10)
0
1
13
Just "training" - it's cleaner
# The True Nature of Training. A student approached Master Dogen in his pottery workshop. "Master, what is the difference between pre-post-training and post-mid-training?" the student asked. Dogen held up a bowl half-dried on the wheel and another newly glazed. "When does the.
0
0
0