Soursos Anastasios

Systems Architect

[Roan @RohOnChain]

This 2 hour Stanford lecture on AI careers will teach you more about winning in the AI race than every piece of AI content you have scrolled past this year.

Bookmark this & give it 2 hours, no matter what. It'll be the most productive thing you could do this weekend.

[Andrew Yang🧢⬆️🇺🇸 @AndrewYang]

Please, put your phone down! @joinnoblemobile

[EuroLeague @EuroLeague]

Start your day the right way...

4 minutes of Dimitris Diamantidis magic 💎

[The Big Apo @aps13bsk]

Όχι 1, όχι 2 αλλά 3 buzzer beater από τον @nunnbetter_

29 Points with 6/8 2FG, 5/9 3FG and 2/2 FT.

In a league of his own!!!

#paobc #euroleague

[Hyung Won Chung @hwchung27]

I gave a lecture at @Stanford CS 25.

Lecture video: https://t.co/Js7SThVO6S

AI is moving so fast that it's hard to keep up. Instead of spending all our energy catching up with the latest development, we should study the change itself.

First step is to identify and understand the dominant driving force behind the change. For AI, a single driving force stands out; exponentially cheaper compute and scaling of progressively more end-to-end models to leverage that compute.

However this doesn’t mean we should blindly adopt the most end-to-end approach because such an approach is simply infeasible. Instead we should find an “optimal” structure to add given the current level of 1) compute, 2) data, 3) learning objectives, 4) architectures. In other words, what is the most end-to-end structure that just started to show signs of life? These are more scalable and eventually outperform those with more structures when scaled up.

Later on, when one or more of those 4 factors improve (e.g. we got more compute or found a more scalable architecture), then we should revisit the structures we added and remove those that hinder further scaling. Repeat this over and over.

As a community we love adding structures but a lot less for removing them. We need to do more cleanup.

In this lecture, I use the early history of Transformer architecture as a running example of what structures made sense to be added in the past, and why they are less relevant now.

I find comparing encoder-decoder and decoder-only architectures highly informative. For example, encoder-decoder has a structure where input and output are handled by separate parameters whereas decoder-only uses the shared parameters for both. Having separate parameters was natural when Transformer was first introduced with translation as the main evaluation task; input is in one language and output is in another.

Modern language models used in multiturn chat interfaces make this assumption awkward. Output in the current turn becomes the input of the next turn. Why treat them separately?

Going through examples like this, my hope is that you will be able to view seemingly overwhelming AI advances in a unified perspective, and from that be able to see where the field is heading. If more of us develop such a unified perspective, we can better leverage the incredible exponential driving force!

Slides: https://t.co/RccKFOqK7e

[EuroLeague @EuroLeague]

It's the moment you've waited for @paobcgr fans...

Lift the Trophy HIGH

#F4GLORY

[EuroLeague @EuroLeague]

A movie 🍿

[Panathinaikos BC @Paobcgr]

#ΝέαΦωτογραφίαΠροφίλ

[Massimo @Rainmaker1973]

That’s actually a real person doing that.

Ray Castoldi has been the stadium organist at Madison Square Garden since 1989, best known for playing during all the Knicks and Rangers games. He has played more shows at MSG than Billy Joel.

[📹 newyorknico]
https://t.co/JZxire88me

[Massimo @Rainmaker1973]

How to fold an effectively flying paper fighter jet

[Stephan Hoyer @shoyer]

My team at Google is looking to hire a PhD student intern for research on AI-based coupled Earth system modeling. This would be a full-time ~3 month position in summer or fall 2024 working in-person in Cambridge, MA with @dkochkov1 and @janniyuval.

[Borriss @_Borriss_]

It is a historic day for AI.

The new text-to-video model by OpenAI Sora can generate videos indistinguishable from real ones.

7 examples:

[Massimo @Rainmaker1973]

These two squares are the same color

[Science girl @sciencegirl]

Slow-motion footage capturing objects descending into fluids with mesmerizing fluid motion effects
https://t.co/bBY6SCwQcR

[Stanford NLP Group @stanfordnlp]

A 2023 update of the CS224N Natural Language Processing with Deep Learning YouTube playlist is now available with new lectures on pretrained models, prompting, RLHF, natural language and code generation, linguistics, interpretability and more. #NLProc

https://t.co/6hb0EIaR5Z

[Soursos Anastasios @asoursos]

New book just arrived!!!

#ai #problemSolving #herbert_simon #allen_newell

https://t.co/QPf7sJRoNn

[Martin Fowler @martinfowler]

NEW §

In her 5th memo about exploring GenAI for software development, Birgitta Böckeler answers the frequently asked question if coding assistants are making the practice of Pair Programming obsolete. Spoiler alert: They don't.

https://t.co/NJZPrLebEb

[Kevin Schawinski @kevinschawinski]

The European Union has just agreed on a Compromise Text for its EU AI Act.

The EU AI Act will set the global standard for how AI systems must be built. Think of it as the GDPR for AI.

I'm going through the text to highlight important changes... 🧵 1/n

[Martin Kleppmann @martinkl]

One week remaining to register for our summer school if you're interested! More details now available on the website https://t.co/hFw63OJtgG

[Soursos Anastasios @asoursos]

When you find a code in typescript and you want to convert it to c#.. you want a transformer right?
Here you are:

#copilot
https://t.co/SlaNPAyHUV