Frederico Costa Galvão

Software, Final Fantasy, Semantic tech, Music. I want to build the next generation of personal digital assistants for truth-seeking people in Smart Cities.

[Frederico Costa Galvão @fredgalvao]

In a world of software that doesn't care about {context persistence and semantic analysis across service/app/source boundaries} nearly enough, which sadly happens to be our world, every Alt-Tab or Ctrl-Tab is a story never told.

[. @plevy]

« Knowledge accumulation is not a substitute for actual understanding. »

[Inrupt @inrupt]

Our Chief of Security Architecture Bruce Schneier outlines how LLMs and large-language-model optimization will transform how we all use the web. via @TheAtlantic

https://t.co/R6qnttGQR8

[near @nearcyan]

thoughtful post on attention economics + countermeasures: https://t.co/XNfCxDw2Yc

a few excerpts posted below

[Frederico Costa Galvão @fredgalvao]

On the same topic, another treasure trove of references and amazing work on multilingual AI, by @seb_ruder:

https://t.co/LGTq0te7eK
So many golden paragraphs.

(slightly sad conclusion that the state could be so much better, but hopeful of the huge potential for improvements)

[Frederico Costa Galvão @fredgalvao]

These links are total treasure troves for fans of linguistics, semantics, and NLP:
https://t.co/z1OTuddV5n
https://t.co/Ot7TzUc3yR
The amount of global+multilingual+collaborative research in WordNets/thesauri and NLP refuels my faith in humanity a bit.

Watch your back, metacrap.

[Frederico Costa Galvão @fredgalvao]

What an amazing "article" (how do I classify this?).

All of these small random questions, like:
>which soundtracks have a french-horn?
>How many sandwiches did you eat last year?
I love and hate them. I want to help fix the software problem.

cc @SWEJWO
https://t.co/MZi1dXHoZW

[Frederico Costa Galvão @fredgalvao]

@MatthewWSiu Context enrichment tools focused on "remembering" (but with a lot of "processing" too) is my dream. Twofold: I'd love to have these tools, and I'd love to work on them.

[Frederico Costa Galvão @fredgalvao]

https://t.co/hY1A3PN4O9

This is one of the doors I thought of: having LLMs process their candidate output in the realm of imagination (simulation), just like humans.

[Ruibo Liu @RuiboLiu]

Simulation is All You Need for Grounded Reasoning!🔥

Mind's Eye enables LLM to *do experiments*🔬 and then *reason* over the observations🧑‍🔬, which is how we humans explore the unknown for decades.🧑‍🦯🚶🏌

Work done @GoogleAI Brain Team this summer!

[Frederico Costa Galvão @fredgalvao]

Machines that learn should "suffer time" more.

I'm usually on the side believing knowledge graphs need to lead in NLP, but a couple of recent reads made me think of improvements to LLM and KG-less machine learning that might be valuable to all sides, even Stochastic Parrots™.

[Frederico Costa Galvão @fredgalvao]

One might say new versions of LLMs (1/year, at best) play that role, but that's still too frozen in time. Maybe an "evergreen/everlearning model" is needed.
Personal AIs in the form of digital assistants and chatbots also fit part of that description, and there's potential there.

[Frederico Costa Galvão @fredgalvao]

"The world is awful. The world is much better. The world can be much better."

And none of these statements would be verifiable without good and open data.

https://t.co/CFmX9zF4LJ