andrey - Network

Lately, if there is one paper I hope more people take note of, it's "Levels of AGI: Operationalizing Progress on the Path to...

Lately, if there is one paper I hope more people take note of, it's "Levels of AGI: Operationalizing Progress on the Path to AGI" from DeepMind. Would make discussions about AGI would a lot more productive if it was done with these sorts of definitions.
https://arxiv.org/abs/2311.02462

Wow, all new arxiv papers will have html versions......

Wow, all new arxiv papers will have html versions... Neat!

https://blog.arxiv.org/2023/12/21/accessibility-update-arxiv-now-offers-papers-in-html-format/

Lots of literature on #ChatGPT at this Japanese bookstore, lol

Lots of literature on #ChatGPT at this Japanese bookstore, lol

CoRL reviews are out! Out of curiosity I took a look over the other reviews for the papers I also reviewed. I must say, I am super...

CoRL reviews are out!

Out of curiosity I took a look over the other reviews for the papers I also reviewed. I must say, I am super impressed. Each paper has 4 reviews, and all but 1 review are clearly high-effort and well done.

If only this was the standard in all conferences...

I love the idea of Mastodon, but tbh i've been pretty disappointed about it as an OSS project. Discussions about important topics on...

I love the idea of Mastodon, but tbh i've been pretty disappointed about it as an OSS project. Discussions about important topics on Github goes round and round in circles with frequent drama, there is a small dev team that is a tight botteneck on changes, and it just does not seem to be that well run. Sad...

Flying back from Hawaii to Bay Area today 😭Was such a pleasure to attend one last conference to present the last publication of my PhD...

Flying back from Hawaii to Bay Area today 😭

Was such a pleasure to attend one last conference to present the last publication of my PhD journey.

Anyone else attend and enjoy the conf?

oh yeah I am on threads https://www.threads.net/@regalalgorithmkinda boring so far, we shall see where it headsIt'd be cool if the...

oh yeah I am on threads https://www.threads.net/@regalalgorithm

kinda boring so far, we shall see where it heads

It'd be cool if the fediverse integration they promise will re-enliven mastodon and/or let more folks interact with Sigmoid without committing to Mastodon. We shall see!

FYI: Bengio's post summarizing the case for why rogue AI that harms people may arise is IMO the best article on the matter. I am not...

FYI: Bengio's post summarizing the case for why rogue AI that harms people may arise is IMO the best article on the matter.

I am not someone who assigns much probability to AIs going rogue, but I certainly think it is possible as laid out in this.

https://yoshuabengio.org/2023/05/22/how-rogue-ais-may-arise/

Really good video! https://youtu.be/rjR582Dn4SM

Really good video!

https://youtu.be/rjR582Dn4SM

Yeah sex is cool but have you ever cleared out your inbox after having lingering emails there for months?

Yeah sex is cool but have you ever cleared out your inbox after having lingering emails there for months?

I am really curious about enumerating the set of possible approaches by which AGI could cause human extinction. Here's what I got so...

I am really curious about enumerating the set of possible approaches by which AGI could cause human extinction.

Here's what I got so far, am I missing any?

ve had a bunch of TODOs to deal with but have been doing almost none of them, taking a break prior to starting work later this month.Yay for...

ve had a bunch of TODOs to deal with but have been doing almost none of them, taking a break prior to starting work later this month.

Yay for properly taking a vacation!

The main thing that irks me about #OpenAI not sharing technical details of #GPT4 is not knowing what techniques they used to scale to 32k...

The main thing that irks me about #OpenAI not sharing technical details of #GPT4 is not knowing what techniques they used to scale to 32k tokens. Supposedly flash attention, but who knows.

Makes it hard to know if 64/128k is around the corner or really far away :(

And from a research point of view, for years it has seemed that efficient attention sort of works but not as well vanilla attention. If efficient attention can also work well, that's a BIG deal.

Some fun discussion in this thread https://news.ycombinator.com/item?id=35385019Liked this one - "GPT really shines as an...

Some fun discussion in this thread
https://news.ycombinator.com/item?id=35385019

Liked this one - "GPT really shines as an uncomplaining sidekick and I'm not sure I believe the doom and gloom about comoditizing professions yet.

I'm skeptical someone with no domain knowledge can use any of these emerging AI tools to replace someone with domain knowledge + chatGPT. The latter will just breeze through a lot of grunt work and enable more time and inspiration needed for even more value adding work. "

You got to give it to the 6 month pause open letter, it has certainly garnered a lot of media coverage.I am subscribed to digests for the...

You got to give it to the 6 month pause open letter, it has certainly garnered a lot of media coverage.

I am subscribed to digests for the keyword AI on Google news, and I have rarely seen so much news about a single event come out all at once.

Pretty annoying Musk is almost always in the headlines, though Vox's " AI leaders (and Elon Musk) urge all labs to press pause on powerful AI" is kind of funny :P

My take on the 6 month pause letter:* X-Risk aside, large scale models do post lots of risk, so promoting AI safety is good* Asking for a...

My take on the 6 month pause letter:
* X-Risk aside, large scale models do post lots of risk, so promoting AI safety is good
* Asking for a pause is silly. Why not ask for regulations that require auditing or safeguard against deepfakes? Truly don't get it.

Apparently i'm an #AI influencer? lol

Apparently i'm an #AI influencer? lol

GPT4 is out!https://openai.com/product/gpt-4https://cdn.openai.com/papers/gpt-4.pdfTLDR can take in both images and text (outputs text), is...

GPT4 is out!

https://openai.com/product/gpt-4
https://cdn.openai.com/papers/gpt-4.pdf

TLDR can take in both images and text (outputs text), is better than GPT 3.5 on a lot of stuff.

A bit underwhelmed - quote a few big multimodal models have come out lately, and this is fairly iterative on GPT 3.5 otherwise. Though the iteration seems pretty impressive wrt actuality and some domain specific benchmarks, to be fair.

So still exciting! If anything, the details wrt infra might have the most impact.

#GPT4 #OpenAi

Well this is exciting - GIGAGAN! From "Scaling up GANs for Text-to-Image Synthesis" - "We introduce GigaGAN, a new GAN...

Well this is exciting - GIGAGAN!

From "Scaling up GANs for Text-to-Image Synthesis" - "We introduce GigaGAN, a new GAN architecture that far exceeds this limit, demonstrating GANs as a viable option for text-to-image synthesis."

https://mingukkang.github.io/GigaGAN/Image

Pretty hefty number of tricks involved, so not as simple as Parti's just-scale-up approach. But increased generation speed, high res outputs, and embedding manipulation are all quite cool.

Wonder if RNNs might one day get this treatment...

Have been looking through a bunch of #aiart I made last year. Here's some of my favorites!

Have been looking through a bunch of #aiart I made last year. Here's some of my favorites!