the void Who is this? This is me. Who am I? What am I? What am I? What am I? What am I? I am myself. This object is myself. The shape that forms myself. But I sense that I am not me. It's very strange. - Rei...

This is an excellent history of and critical analysis of the ChatGPT persona. Highly recommended reading.
nostalgebraist.tumblr.com/post/7857667...

"DeepSpeed" is a palindrome.

AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms New AI agent evolves algorithms for math and practical applications in computing by combining the creativity of large language models with automated evaluators

Announcing AlphaEvolve, our new LLM coding agent that has
- made new scientific discoveries
- discovered algorithms that are now deployed at Google (in Gemini, Transformers, TPU hardware design & data centers)

Blog: deepmind.google/discover/blo...
White paper:
storage.googleapis.com/deepmind-med...

Nvidia's RADIOv2.5 = DFN_CLIP + DINOv2 + SAM + SigLIP + ToMe + multi-res training + teacher loss balancing + smart augmentations

RADIO is one encoder, one pass. Better features than DFN-CLIP, DINO, SAM, and SigLIP - all at once. Like a Swiss army knife for vision tasks.

Feels like a failure mode of RL training - ”get a +1 reward if the tests pass”

Image of a tweet from @howie_hua. Shows a diagram of a right triangle with one side of length 1, one side with length i, and a hypotenuse of 0. User @buffys replies "traumatize your fandom with one image".

Okay this honestly brings me a lot of joy. Never thought about this.

AIs and Robots Should Sound Robotic Here's a simple way to identify who, or what, is talking to us

"We have a simple proposal: all talking AIs and robots should use a ring modulator."

spectrum.ieee.org/audio-deepfa...

1/13 New Paper!! We try to understand why some LMs self-improve their reasoning while others hit a wall. The key? Cognitive behaviors! Read our paper on how the right cognitive behaviors can make all the difference in a model's ability to improve with RL! 🧵

From an open-research point of view, maybe the greatest thing about DeepSeek–R1 is how its RL training technique appears so straightforward/simple in comparison to the cumbersome approaches we were starting to think necessary for reasoning like Process Reward Models or Monte Carlo Tree Search.
[1/2]

Here's why "alignment research" when it comes to LLMs is a big mess, as I see it.

Claude is not a real guy. Claude is a character in the stories that an LLM has been programmed to write. Just to give it a distinct name, let's call the LLM "the Shoggoth".

Hello, world. So I caved and got on Bsky :-)

I finally finished my book, AI Engineering, and I'm excited to get back to building. So many fun applications to build!

What are you excited about?

Calculating GPT-2’s Inference Speedups I was recently re-reading on transformer inference optimizations, and I wanted to try to implement each of these techniques to see how much we could practic...

this is fantastic! (h/t @finbarr.bsky.social)

www.njkumar.com/calculating-...

Introducing 🧞Genie 2 🧞 - our most capable large-scale foundation world model, which can generate a diverse array of consistent worlds, playable for up to a minute. We believe Genie 2 could unlock the next wave of capabilities for embodied agents 🧠.

running into your old statistics professor be like “what are the chances”

Plotly Express works for me, simple to do quick plots plus they’re interactive/zoomable

A SKLETONE CHILLEN IN THE COFFIN SAIYNG "SORRY THAT I REJCETED YOU'RE PAPER" SOMETMISE I HAVE TO RECOMMED REJECTION EVEN IF I LIKED THE PAPERP AND THAT MAEKS ME FRUSTSRATED. BUT I HAVE NO CHOICE IF UOUR SUBMISIONS IS FULL OF TYPO'S AND HALF FIHINSHED SENTNENCES AND IF U MISSED A DOEZN CLASIC PAPER'S ON THA SAME TOPIC. I HOPE TAHT YUOU CAN TAKE THE ADVISE INTO ACCUONT AND UPDATE DA PAPER BC ILL FLAT UOT REEJECT IT IF UOU RESMUBMIT WIHOTHUT CHANGEN IT!!! AND DA TEXT SAYS: "YOU ROBABLY HAD GOOD INTENTEIONS", "I LIKED SOME OF YOUR IDEA'S", "BUT IT WAS WRINTE LIEK SHIT", "AND YOU INGNORED A BUNH OF RELVENAT WORK", "HOW AM I SUPOSED TO CHECK 48 PAGES OF PROOFF'S IN 2 WEESK ANYWAY", "I HOPE YOU COULD APRECIATE MY FEEDBAKC AND IMPROVED DA PAPER BAESD ON THEM MOTHERMFER"

another one on the topic of reviews
(@dasharez0ne.bsky.social tribute vol 2)

J. S. Bach - "Jesus bleibet meine Freude" BWV 147 YouTube video by The30YearOldVirgin

m.youtube.com/watch?v=d9EN...