Twitter Feed: Posts

@relizarov ("Roman Elizarov") replied to a tweet by @relizarov:

This should give them the ability to produce better results on complex problems without being explicitly asked to write a step-by-step solution, and a proper training can teach them to automatically generate the intermediate steps without writing them down in the answer. /9

@relizarov ("Roman Elizarov") replied to a tweet by @relizarov:

So what would happen if we explicitly imbue LLMs with the human-like ability to selectively suppress the words they are about to write and keep them only "in their mind" as a further input to their word-generation process? /8

@relizarov ("Roman Elizarov") replied to a tweet by @relizarov:

If you've followed the news on the capabilities of modern LLMs, you might have noticed that they usually give better answers to complex questions when explicitly asked to write down a step-by-step solution. /7

@relizarov ("Roman Elizarov") replied to a tweet by @relizarov:

Of course, the human brain learns about the world using a very different and seemingly more efficient strategy than modern LLMs, but this is not the point. It is how humans come to a decision on what to say/write next that matters for my point below. /4

@relizarov ("Roman Elizarov") replied to a tweet by @relizarov:

Essentially, an LLMs is like a child, a very erudite one, but a child that still says aloud everything they think about. Like when you are asked to add two multi-digit numbers together and you say aloud all the steps you are making in your head to get the result. /6

@relizarov ("Roman Elizarov") replied to a tweet by @relizarov:

The key functional difference between the modern LLMs and humans is in the purely feed-forward nature of the former. Humans have an option to stop and think in their mind on what to say next, while LLMs do not have such a capability. /5

@relizarov ("Roman Elizarov") replied to a tweet by @relizarov:

We don't really know that much about the inner working of the human brain to be sure. But we do know that people rarely use their "slow" deliberate and resource-consuming part of the brain and mostly rely on "fast" intuitive thinking and decision making. /2

@relizarov ("Roman Elizarov") replied to a tweet by @relizarov:

This fast thinking and speech in humans might as well be based on the same principles of choosing the most statistically likely continuation of the previous interaction context using the learned data on statistics about the world, just as modern LLMs do. /3

@relizarov ("Roman Elizarov") wrote:

Lately, there are many popsci and philosophical articles about modern AI. Many of them describe the inner workings of Large Language Models, their statistical nature, and imply that this proves they are unlike humans. Who said the human brain is not doing essentially the same? /1

@steveklabnik wrote:

🤣🤣🤣🤣🤣🤣🤣

with quote tweet:

@austintribune ("The Austin Tribune") replied to a tweet by @daustinsanders:

@daustinsanders

@steveklabnik wrote:

"I think there's a pretty good rule of thumb in modern systems-ish language design: If Go and Rust and Zig all do a certain thing, that thing is probably a great idea. These languages have very different priorities, but often they overlap."

- https://news.ycombinator.com/item?id=34838613

@rabbix ("Carl P. Corliss") replied to a tweet by @rabbix:

Not only do I have a safe that was locked by the prev occupants, but the lockset on the room door is broken, causing us to be stuck in the room for a few until I could jiggle it the right way to open up. I know there’s construction going on, but I didn’t expect to end up in it.

@MattBaume ("Matt Baume") replied to a tweet by @MattBaume:

Absolutely perfect

@MattBaume ("Matt Baume") wrote:

I have just learned that Freddy Mercury was super into cats, so please update your list of reasons to love him

@homophiliapod ("Homophilia Podcast") wrote:

This week, we dig deep with writer/director Goran Stolevski on his new film #OfAnAge
https://podcasts.apple.com/us/podcast/homophilia/id1267538496?i=1000600131467

with quote tweet:

@FocusFeatures ("Focus Features") wrote:

One look is all it takes. #OfAnAge is only in theaters starting this Friday.

Get tickets now: http://ofanagetickets.com

Video:

@steveklabnik wrote:

i wonder if the onion will ever start pulling punches

https://www.theonion.com/it-is-journalism-s-sacred-duty-to-endanger-the-lives-of-1850126997

@LolOverruled ("Lolo") wrote:

I was surprised by how much I enjoyed The Fabelmans. Watching it felt like taking a warm bath

@rust_foundation ("Rust Foundation") retweeted:

@RustNationUK ("Rust Nation UK 🦀") wrote:

It's great to have all of our sponsors here with us at #Rustnationuk - part 1.

Thank you for sponsoring this year's event - @KeyrockTrading, @jfrog, @rust_foundation, @PrimeVideo, @SurrealDB

#Rust #Rustlang

@DemSocialists ("DSA 🌹") retweeted:

@railroadworkers ("Railroad Workers United ✊") wrote:

Precision Scheduled Railroading (PSR) is the operating model for increasingly greedy railroads. In 2019 Railroad Workers United highlighted problems PSR was creating leading to increased major derailments. These problems continue to be factors in unsafe rail conditions today.

Video:

@RikerGoogling ("Riker Googling") wrote:

holodeck games where you punch nazis

@DemSocialists ("DSA 🌹") wrote:

The people of Ohio deserve more than fibs from top Democrats.

https://jacobin.com/2023/02/pete-buttigieg-railroad-safety-procedures-brakes-hazardous-materials-derailment-risk

@reduct_rs ("ReductRs") wrote:

How to Communicate Your Dislikes During All Hands by Booing Loudly

@DemSocialists ("DSA 🌹") retweeted:

@TheDuhalde ("David S. Duhalde") wrote:

.@nycDSA showing up for @amazonlabor night at the @BrooklynNets game!

@DemSocialists ("DSA 🌹") wrote:

Join DSA members and ecosocialists from all over the country March 9, 8pm ET for @DSAecosocialism's Building for Power campaign launch. #ABetterWorldIsPossible

with quote tweet:

@DSAecosocialism ("DSA for a Green New Deal🌱🌹⚡️") wrote:

We're done letting the ruling class burn our towns, lives, and planet for a profit. If you are too, join us March 9th at 8PM EST to talk about how we can win the world we deserve by fighting alongside labor and Building For Power!
http://dsagnd.org/b4p-launch
#ABetterWorldIsPossible

@ProductPrincipl ("Lisa K. Wells") wrote:

#Kubernetes troubleshooting can be a real headache for dev teams. 😫 But fear not! @gostackstate's latest blog on guided Kubernetes troubleshooting will help you reduce toil and get back to what you do best. Check it out now! #SRE #DevOps #Troubleshooting https://www.stackstate.com/blog/guided-kubernetes-troubleshooting-how-to-reduce-toil-for-dev-teams/

@michaloleszczyk ("Michal Oleszczyk") wrote:

Nowy SpoilerMaster już jest, a w nim... TITANIC (1997) Jamesa Camerona! Zapraszam do słuchania!

https://open.spotify.com/episode/5NUsAMg2sCHGpnCZOvo0Qo?si=WxYSXnesQaurvDe_D7aJEw

@ewarren ("Elizabeth Warren") wrote:

So here’s my question: what does he have to hide? https://www.seattletimes.com/business/starbucks/starbucks-ceo-howard-schultz-declines-to-testify-in-sanders-hearing/

@rust_foundation ("Rust Foundation") replied to a tweet by @rust_foundation:

2022 Annual Report: https://foundation.rust-lang.org/news/rust-foundation-annual-report-2022

@rust_foundation ("Rust Foundation") replied to a tweet by @rust_foundation:

Rust Foundation blog: https://foundation.rust-lang.org/news

@rust_foundation ("Rust Foundation") replied to a tweet by @rust_foundation:

Community Grants Program information: https://foundation.rust-lang.org/grants