Peculiar Musings

[François Chollet @fchollet]

In the initial 2020 competition, we set a SOTA of 20% on ARC-AGI. Four years later, the SOTA is 34%. I would be disappointed if we don't get to at least 45% (maybe even 50!) by the end of the year with the new competition.

Meanwhile, average human-level performance is around 85%.

[Jay Fenello @jfenello]

"Once a government is committed to the principle of silencing the voice of opposition, it has only one way to go, and that is down the path of increasingly repressive measures, until it becomes a source of terror to all its citizens and creates a country where everyone lives in fear."
-- Harry S. Truman

[François Chollet @fchollet]

General intelligence is *precisely* learning -- the ability to efficiently learn new things, beyond what your genes and past experiences prepared you for.

Current ML has near zero intelligence because static inference with a curve only yields local generalization, with zero ability to adapt/learn, and meanwhile fitting a curve via gradient descent is an extremely data-inefficient process (compared to e.g. discrete program search which can pick up complex, novel tasks from 1-2 examples). It requires a dense sampling of its operational space in order to generalize -- because it is limited to local generalization.

[Peculiar Musings @MusingsPeculiar]

@Swizec Have you considered that engineering might have a better vision of what is technically possible than the customer. Can engineering listen to requirements, read between the lines, then show don’t tell,

[Peculiar Musings @MusingsPeculiar]

Mind environment feedback loop. You create your own luck by preemptively solving future problems whilst remaining Buddhist non-attached about the outcome. Change your environment to change your mindset.

[Peculiar Musings @MusingsPeculiar]

Grayface is obsessed with constantly observing and collapsing the wavefunction for fear of Liminality quantum tunnelling into the uncertainty principle and summoning the unknown to overthrow the status quo

[Jason Wei @_jasonwei]

Moving from Google Brain to OpenAI, one of the biggest changes for me was the shift from doing individual/small-group research to working on a team with several dozen people. Specifically, working on a bigger team has led me to think more about UX for researchers. Some examples:

1. Great tooling accelerates research. Subpar tools hamper researchers by introducing unnecessary friction into thinking and analysis. Even small improvements like reducing clicks and scrolls can significantly increase researcher's productivity. Visualizations become particularly vital when working with multi-task models, helping to better evaluate tradeoffs between different models.

2. Simple design is key for a the success of an evaluation benchmark. For example, GLUE/SuperGLUE, as well as MMLU/GSM8K have a single number (higher is better), and everyone wants it to go up. They are easy to understand, download, and evaluate. Other benchmarks (e.g., BIG-Bench, probably one of the great benchmarks of the past two years IMHO) can have advantages such as much broader coverage, but are basically impossible to run and a pain in the ass to analyze. For Google's PaLM paper, I heard one engineer's full-time job was just to run BIG-Bench...

3. Strong documentation enables scaling communication without involvement. Imagine if you have to chat with someone to explain how something works. They have to wait for you to reply, and you have to stop your work to message them. This takes up two people's time. With good documentation, you don't have to be involved at all, and the other person doesn't have to wait for your responses, which accelerates both people a lot.

[François Chollet @fchollet]

The shortage of folks who actually understand how deep learning works and what it can/cannot do is gearing up to be even more dire than the upcoming GPU shortage.

[Runway @runwayml]

If you can imagine it, you can generate it.

Gen-2 is now available on web and mobile: https://t.co/KctIodJz2t

[Peculiar Musings @MusingsPeculiar]

"Order is just Chaos waiting to happen."
- Excerpt from The Honest Book of Truth, POEE

[Nathan Lands @NathanLands]

So...I thought Photoshop was in trouble because of AI.

But now, Adobe's new AI-enhanced Photoshop beta is out. And it's one of the most magical things I've ever seen. (part 2)

12 game-changing examples that'll blow your mind:

[Peculiar Musings @MusingsPeculiar]

@danielmarven He promised to buy her a horse!

[Peculiar Musings @MusingsPeculiar]

@elonmusk Adding to the Elon Musk AI Wife saga

[François Chollet @fchollet]

Google has a brand new landing page for ML developer resources: https://t.co/v3I9cwpcCA

[Peculiar Musings @MusingsPeculiar]

MidJourney is a new visual language, an ability to think and communicate visually.

As a newborn I have no reference guide to how different words and phrases appear visually or how to communicate thus I am performing AI baby babble for a new AI visual synesthesia language

[Peculiar Musings @MusingsPeculiar]

@fchollet Opps, I got distracted and drew an entire art gallery worth of cats!

[Peculiar Musings @MusingsPeculiar]

type Board=[[Bool]]
nextGen b=[[cell(p,q)|q<-[0..nc-1]]|p<-[0..nr-1]]
where nr=length b; nc=length(head b);cell(p,q)=let n=length$filter(b!!)(nghbrs p q)in(b!!p!!q&&n==2)||n==3
nghbrs p q=[(p+dP)`mod`nr,(q+dQ)`mod`nc|dP<-[-1..1],dQ<-[-1..1],(dP,dQ)/=(0,0)]

[Peculiar Musings @MusingsPeculiar]

Rust: the community will eventually adopt a fork name such as iron to distinguish themselves from core rust

https://t.co/BePtceHVW5

[Joseph Thacker @rez0__]

This morning I was hacking the new ChatGPT API and found something super interesting: there are over 80 secret plugins that can be revealed by removing a specific parameter from an API call.

The secret plugins include a "DAN plugin", "Crypto Prices Plugin", and many more.