jpfr

Hot rolling is a good example. For the last 30 years neural networks have been used to control the process. There’s a nice NeurIPS paper from 1991 [1].

And even with the non-deep networks from back in the day, people felt compelled to include prior knowledge and even used Bayesian priors to deal with a lack of data.

[1] http://papers.neurips.cc/paper/447-neural-control-for-rollin...

No it won’t transform manufacturing. Deep RL in physical systems is a pipe dream.

Deep RL is one of the most data hungry methods. And in physical systems you don’t get enough samples.

There are loads of data. But no manager of a shopfloor lets you produce scrap just to potentially learn.

There is an opportunity to combine ML with classical engineering models for manufacturing. Think differential equations for chemical processes. Then you end up with something like ML augmented control theory. That does work.

There were only 9 years between the 1998 Comdex and the iPhone release in 2007. Of the 23 years we have had iPhones for 14 years... Time perception is sometimes strange.

RISC-V really shines when the op-compression extension and macro-op fusion are taken into account.

https://news.ycombinator.com/item?id=25542963

Many of the early critiques of RISC-V have not considered that. And many points become entirely moot in light of these techniques.

It has become a rite of passage. You have effectively made it as an AI researcher if Schmidhuber decides that your breakthrough is actually derivative work of an idea he published 30 years ago.

The crazy thing is ... he is not entirely wrong. Only mostly.

> he was also not very kind to walter pitts and possibly contributed to his suicide

Wiener's wife told him a lie about Pitts. She invented an indecency involving Wiender's daugher. Both Wiener and Pitts never found out the truth about who ended their relationship.

Easy, you only have to add a handful of „battery included“ packages to the default system image.

That however means that some packages get a preferred status in the Julia ecosystem.

I have high hopes for the zz language. Especially the fact that it outputs C code. So interop is seamless.

https://github.com/zetzit/zz

For any systems language, interop with C is the litmus test.

With that in mind, this new language should not require 15,000 lines of standard library. A type-safe wrapper for libc should be enough...

Tough luck. Using 0-indexing would have lost Matlab users to the Julia community. So they consciously decided to rather lose you. :-)

I got used to switching between 0/1-indexing. The only place it actually hurts is when using modulo to remain within some range. And this is needed much much less in Julia than in C and C-like.

There is branch for elisp support in guile. https://git.savannah.gnu.org/cgit/guile.git/log/?h=wip-elisp

It works in principle but is slow due to some impedance mismatches that have to be bridged. Especially, Emacs has a special string format internally that is not native utf-8.

Andy Wingo is making a great job with JIT-support for Guile. We have to see whether emacs-guile can become fast enough to replace native-elisp. At this time it doesn't look like it...

This uses openCypher for queries, which is about to become an ISO Standard. In the mid-term the query language will be on equal footing with SQL.

http://www.opencypher.org/articles/2019/09/12/SQL-and-now-GQ...

If you wrote the story as fiction, most non-tech readers would find it boring. Somebody who is really interested in tech, driven and works all day, every day.

For the tech-savvy readers there's a lot of stuff pg did that is not directly related to his material success, but oh so cool.

For example his book on Lisp: http://www.paulgraham.com/onlisp.html

You don't write a book on Lisp because you want to get rich. You write it because you just do it for its own sake. And then he got rich immediately after publishing it.

What a cool move. For the nerds who can read the signals.

The situation the author describes is one of the reasons why we cannot (yet) get rid of Elsevier and the other scientific publishers. They are no longer needed for the dissemination of new research. But they are the keeper of the archive.

We don‘t know which works are going to be important for future generations. And just putting out a „PDF on the net“ will be gone in 20 years if there is no publishing house maintaining the original copy.

Arxiv could also take on that role. But not for the last 300 years of existing research in the publishers archives.

Its even worse. In many cases blockchains still require a central authority that controls the write access to the system.

A blockchain / distributed ledger can be easily flooded with spam. The only way around that is to make access expensive. That works for the "actual" blockchain because the effort to put something in the chain is the money.

In applications where you don't have want expensive writes, a central gatekeeper (or at least a cabal of power-nodes) is needed to restrict access to prevent spamming.

That is pretty readable C... for developers from that particular community.

The ngn/k implementation of the k language [1] is even more terse.

[1] https://git.sr.ht/~ngn/k/tree

Fabrice Bellard, who is quite famous on HN for his open source work, has founded a company for his own 5G implementation. And they are EU-based. [1]

I'm under the impression his work is closer to the "radio protocol" than the infrastructure management behind. Not sure if they are related to Open-RAN though...

[1] https://www.amarisoft.com/about-us/

Yes. Beamer LaTeX with the Metropolis theme and Fira Sans as the main font.

I teach a graduate course in optimization methods for machine learning and engineering [1,2]. Julia is just perfect for teaching numerical algorithms.

First, it removes the typical numpy syntax boilerplate. Due to its conciseness, Julia has mostly replaced showing pseudo-code on my slides. It can be just as concise / readable; and on top the students immeditaly get the "real thing" they can plug into Jupyter notebooks for the exercises.

Second, you get C-like speed. And that counts for numerical algorithms.

Third, the type system and method dispatch of Julia is very powerful for scientific programming. It allows for composition of ideas in ways I couldn't imagine before seeing it in action. For example, in the optimization course, we develop a mimimalistic implementation of Automatic Differentiation on a single slide. And that can be applied to virtually all Julia functions and combined with code from preexisting Julia libraries.

[1] https://www.youtube.com/playlist?list=PLdkTDauaUnQpzuOCZyUUZ...

[2] https://drive.google.com/drive/folders/1WWVWV4vDBIOkjZc6uFY3...

He knows too much about the topic to be bullshitting. Also, he does not pretend to have better answers than anyone else.

But IMHO Lex tries too hard getting deep/philosophical answers from his guests. And some guests clearly are not into “can we ever understand the brain and what would that even mean?”-type discussions.

If you want to see a real AI bullshit artist in action, look up recent appearances of Ben Goertzel.

I hope calling him a “con” wasn’t intentional.

He speaks French after all...

Looking at a small number of weather stations is "evidence by anecdote" and completely unsound.

You introduce bias by your selection of measurements. Plus a large portion of the earth is scantly inhabited with no weather stations in place. Like the poles or international waters.

Instead you should look at satellite measurements of the temperature across the entire planet. These are widely available. The scientific discourse and the available evidence is much more advanced than "hey it became colder in those 20 places".

https://en.wikipedia.org/wiki/Satellite_temperature_measurem...

Have the dominant TCP implementations of today discovered the same solution or is there still dormant potential to increase bandwidth?

I still have my files from 20 years ago.

I don't have my "media libraries" from 20 years ago.

As soon as a tool tries to move my files into an internal representatation I cannot control, I delete the tool.

From the last slides, they are turning VLC into another iTunes.

Too bad. I liked having a player that emphasized playing "files" instead of loading everything into some opaque "media library" first.

Let's see how long it takes before the monetization of the new "iTunes VLC" kicks in.

Emacs has a lot of cruft on the code-level. The worst part is the lack of separation between "backend" and "frontend".

There is support for many different "frontend" architectures: Win32/MacOS/Gtk/<Motif and many more old X11 toolkits>

But a lot of code was just duplicated when a new frontend architecture was added. And started to grow into different directions. The code is littered with #ifdef and there is no clear API for the fronted architectures to sit on top of.

For example, font handling is different for most frontends. And this difference leaks into the core part of Emacs where line-breaks are managed, etc.

There is work towards a "pure GTK" that uses Cairo as a drawing surface.

https://github.com/masm11/emacs

Imho the Cairo "drawing surface" and font handling should be the only implementation for what it does and reused for all frontend architectures. With the exception of terminal sessions...

But, Emacs devs are very conservative when it comes to "modernize". In my experience, it is a lot easier to add stuff than to clean up. That also explains the forest of #ifdef and accumulation of cruft.

He still wants Emacs to become a general-purpose word processor.

https://lists.gnu.org/archive/html/emacs-devel/2018-03/msg00...

https://lists.gnu.org/archive/html/emacs-devel/2020-04/msg01...

And not everybody is happy about it. [Careful, flame ahead.]

http://ergoemacs.org/misc/rms_emacs_tyrant_2018-03.html

Nice! I was wondering before whether Eliza, the 17th century spy girl from Neal Stephenson's "Baroque Cycle", was inspired by historical examples. It seems that was the case.

That looks really promising.

But development seems to have completely stopped earlier this year. Anybody knows what's going on there?

Usually you move the computation to the location of the data and not the other way around. That's the basic premise of Hadoop/Spark and many similar big data tools.

GAIA-X self-descriptions give visibility on whether your data is (would be) in a data center that also has GPU-capacities. And if not, what you would have to pay for interconnection.

See this demo for the basic idea of the self-description graph: https://www.youtube.com/watch?v=f17c3Vpi3rA

To be clear, this is not a new cloud-hyperscaler.

Rather, different providers can offer services and (compute) nodes with different levels of certification / geographic location.

So a government agency / hospital / critical-Infrastructure on can use cloud offerings without loosing „digital sovereignty“ by depending on single providers or outside actors.