🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Why does the Gemini Diffusion page show text being generated autoregressively as you scroll down?

Dude sent the same message about 100 times. I hope you have a good trip my friend.

Gradually adding pieces until you realize you just wanted a laptop.

Somehow it feels incredibly embarrassing that google is paying to advertise Gemini on Twitter.

Got nerdsniped into printing this a little while ago.

If you don't know what an intrusive thought is: just imagine picking up a can of PAM and spraying it into your mouth.

Figuring out if you can eat a poisonous berry 10,000 years ago: "hey guys, have you tried these berries?" Random guy: "don't eat that, it made me sick." Today: "Hey gork can I eat this?" Grok: "Yeah that looks yummy, you're good to go. Berries are safe in general."

Sure the TSP is NP, but if you have any practical problem of a reasonable size (e.g. <10000) you can _probably_ find the optimal solution in like 5 minutes.

Has anybody tried training a neural network where each layer's weight matrix is used in a constraint Ax < c? The input would be some vector x and output is min x*y subject to Ax<c.

The new version of "rubber duck debugging" is realizing the bug as you describe the problem to ChatGPT.

A Markov model and a transformer can both achieve the same loss on the training set. But only the transformer is smart enough to be useful for other tasks. This invalidates the claim that "all transformers are doing is memorizing their training data".