Tanya Grinberg

Robots and life.

[Tanya Grinberg @eoly23]

So, AI houses have taken to hiring philosophers. And yet, we live in the society where kindness isn’t really a subject of study. Ethics? Sure. Empathy? Of course. Benevolence? Maybe. But kindness is not really something scientists or philosophers look at squarely. Isn’t that interesting?

[Rohan Paul @rohanpaul_ai]

Terence Tao says the math behind today’s LLMs is actually simple. Training and running them mostly uses linear algebra, matrix multiplication, and a bit of calculus, material an undergraduate can handle. We understand how to build and operate these models.

The real mystery is why they work so well on some tasks and fail on others, and why we cannot predict that in advance. We lack good rules for forecasting performance across tasks, so progress is largely empirical.

A key reason is the nature of real-world data. Pure noise is well understood, perfectly structured data is well understood, but natural text sits in between, partly structured and partly random. Mathematics for that middle regime is thin, similar to how physics struggles at meso-scales between atoms and continua.

Because of this gap, we can describe the mechanisms but cannot yet explain capability jumps or give reliable task-level predictions. That mismatch, simple machinery versus hard-to-predict behavior, is the core puzzle.

----

Video from 'Dr Brian Keating' YT Channel (Link in comment)

[Tanya Grinberg @eoly23]

Probably the most annoying thing that happened in science curtesy Sir Newton is the entrenchment of practice to avoid sharing of phenomenological insight that led to a discovery. The few lineages where this was not the norm had outlier productivity-eg the “Martians”

[Tanya Grinberg @eoly23]

@MindaMyers Ad astra, @MindaMyers ! Was great to see you shine :)

[Minda Myers @MindaMyers]

Exciting news: @eoly23 and I won the first round of the #AIConferenceHackDay HackrChallenge! The true reward? Realizing I'm capable of so much more than I thought. 🎉

[Joscha Bach @Plinz]

We are organizing the 2023 #AAAI Spring Symposium on Evaluation and Design of Generalist Systems (EDGeS), Mar 27-29, 2023 at Hyatt Regency San Francisco Airport in Burlingame, California.

Registration: https://t.co/YAycPQ9g1K

Symposium website: https://t.co/D8ry7Ha1KA

[Tanya Grinberg @eoly23]

The first thing I tried with GPT-4: render some formulas in LaTeX. Heaven!

[Tanya Grinberg @eoly23]

Registrations now open for #AAAI Symposium on Evaluation and Design of Generalist Systems. Event details: https://t.co/6Hp90WiVIs

[Tanya Grinberg @eoly23]

PSA: Extended abstract deadline for the #AAAI #EDGeS symposium is this coming Monday Jan 23!
See more details (and apply to join our new Discord server) here: https://t.co/vgqo96nvvq

[Tanya Grinberg @eoly23]

Recording will be available here shortly: https://t.co/krdRPa3qDR

[Tanya Grinberg @eoly23]

Starting in 1 hour! Join us for conversation on the future of Generalist #AI with @Plinz, @drmichaellevin and Prof Christoph von der Malsburg. https://t.co/47kXLpLvHJ

[Tanya Grinberg @eoly23]

Planarians regenerate when cut in half, are immune to cancer and can live forever. They’re also one of the most genetically messed up creatures we know of. Something is clearly off about our ideas on what leads to longevity.

[Joscha Bach @Plinz]

Generalist AI beyond Deep Learning: a discussion with Christoph von der Malsburg, @DrMichaelLevin and myself, Monday Jan 09, 9am PST https://t.co/eNdEvnFnil

[Quinoah 🔍⏸️ @NoahTopper]

Okay, y'all ready for round two? Where I explain why this whole argument was bunk? Thanks to everyone for giving me a lot of interesting things to look into. The crux, it seems to me, is that many people claimed these systems (planets, bubbles, etc.) *do* perform computation.

[Quinoah 🔍⏸️ @NoahTopper]

Okay, sorry if this totally confused, but… Planets orbit stars following certain differential equations, without having to internally compute those equations. Bubbles take the shape of minimum surface area, without having to internally minimize an integral.

[Tanya Grinberg @eoly23]

Are you doing research on evaluation or design of generalist AI systems, or ones that can reason on the same epistemological priors as a human does? Join us at the #aaai #EDGeS symposium! Abstract submissions due Jan 16. https://t.co/PKGJhQuMIQ

[Joscha Bach @Plinz]

We'll be organizing an AAAI Symposium on designing and evaluating generalist AI in March. Deadline for submissions will be January 16th, 2023 https://t.co/dVMGeR4yJr

[Tanya Grinberg @eoly23]

Art is about giving objectness to things that would otherwise diffuse into the background.

[Tanya Grinberg @eoly23]

Midjourney is pretty excellent for sci-fi illustration. I finally get to indulge in Zelazny’s God of Light with visuals.