mooneater

I work for a VC, in case you are seeking funding for your startup, Material Science is interesting to us robin@agfunder.com

If you fly backwards, its just all black

Good topic but I felt the types of disruption listed by the author were rather near term and short sighted.

Yeah I like transistor for my reinforcement learning show https://www.talkrl.com/ , their design has really improved

Because postgres and postgis are awesome!

I migrated to them in 2005 (from mysql) and was amazed by how much happier I was. And it never cost me a dollar.

I loved it so much I visited its authors in Victoria and bought consulting from them. They are amazing folks.

In those images I see: - complete graphs (the triangle matrices) - incomplete graphs (the circular ones with only some segments conntected)

And those are excellent ways to visualize these types of objects.

He built knowledge graphs.

Its... cool?

Alternative would be: clone the repo, do your changes, run gitk to be sure.

Which has advantage that result looks exactly the same as it will in production (no new diagram type b/c only using git and gitk), with zero risk that sim differs from reality.

Wrong. No way they could scale using humans. Humans were used only for training data.

ChatGPT itself says all kinds of things; journalists using it as a source are either badly uninformed or purposely misleading readers.

How much of this would be fixed if you lived in the woods and did the same job remotely (which is my situation with similar work).

You might try lions mane, a mushroom that noticeably boosts my mental clarity.

Also, being older we have a lot more on our mind. So methods to help organize and focus are more useful than ever to me now. Take notes, keep logs, etc.

Reminds me a bit of a social network concept of mine from years back: "Karachter"

"Imagine a social network where you only need, in fact you are only allowed, one friend. Imaging a messaging service so personal, so patient, and yes, so damned meaningful, that it only transmits one single character, to your one friend, each day. "

http://randomfeelsgood.blogspot.com/2014/02/announcing-karac...

Sounds similar to what Google's SayCan is doing. https://say-can.github.io/

They taught it separate skills. When a situation arises, the skills (which you could almost consider sub-agents) compete to decide who is most likely to be relevant here. Then that skill takes over for a bit.

They also have a version called "Inner Monologue" in which the different parts "talk to each other" in the sense of collaboratively creating a single inner monologue, allowing for reactiveness/closed loop behaviour.

I interviewed 2 authors of SayCan/Inner Monologue here: https://podcasts.apple.com/us/podcast/karol-hausman-and-fei-...

Ag and food tech industry news: https://agfundernews.com (disclosure: I work for Agfunder)

Is any of Waymo Driver's design published? Like do they use RL or how do they approach control.

I love the annotation interface, did you do that all yourself or use a tool? Ty!

I interviewed an author of WebGPT here, OpenAI co-founder John Schulman if you're interested, and he talks about it https://www.talkrl.com/episodes/john-schulman

20+ year fan of graphviz here, love seeing what you are doing with D2!

Any examples of larger diagrams (1000+s of nodes) and how does that perform in layout?

I'm glad they are thinking through the details on barrier design.

Lets not forget that is only necessary because the automotive industry is barbaric to begin with. Let's recall why trains dont require advanced crash barrier design like this.

Does anyone actually think this number is accurate to a decimal place?

I remember in grade 8 my geography teacher noticing I wasn't paying attention and asking: "What did I just say?"

The sound of his last couple sentences were in my recent memory (timestamped?), but it seems I hadn't fully parsed them yet. When he asked, I took a brief moment to parse them, then to his disappointment I repeated them verbatim to him.

I'll never forget the way my brain processed delayed audio that day!

there are things we the public NEED to do

I dont disagree, but my take is relying on public changing its behavior is not exactly a hopeful strategy.

So glad our high taxes are being put to good use, merci notre gouvernement fédéral

Love the vision, respect the dedication, thanks for the inspiration! A true gentleman and scholar tips hat.

not prediction based on anything statistical but based on one's understanding

It is still statistical. You still want to predict a distribution over expected next token.

However, the function needed to estimate the likely next tokens are not simple max over enumerated next tokens like in a language model. It's more like a transition model in reinforcement learning.

trying to work the concepts out in my own mind and to anticipate what the authors are going to say before turning each page. I usually fail to guess what the next page holds, but the fact that I've tried and failed makes me more ready to understand why the authors chose the paths that they did

TIL Donald Knuth operates a bit like GPT-3 but for research paper narrative.

they haven’t learned the fundamental ideas of algebra

I'd very much love to know exactly what Knuth considers the "fundamental ideas of algebra"!

Unneeded complexity is the greatest enemy of the software engineer.

Unnecessarily complicated is the default. Choose the elegant thing wherever possible (its not always possible, but often is).

Actively avoid complexity, or it will shackle you.

can you say more about how you detect instinct?

Downvoters: fine downvote me, but can you also name a perk that matters?

Go further, open it all.