Victor Garcia

38 posts

Victor Garcia

@vmgdev1

Inscrit le Mayıs 2023

462 Abonnements9 Abonnés

""" One of the biggest misconceptions people have about intelligence is seeing it as some kind of unbounded scalar stat, like height. "Future AI will have 10,000 IQ", that sort of thing. Intelligence is a conversion ratio, with an optimality bound. Increasing intelligence is not so much like "making the tower taller", it's more like "making the ball rounder". """ Anyway, this is pretty clearly a pointless discussion.

English

Brandon Fishback@Bfish94·5h

@vmgdev1 @tobyordoxford @fchollet OP never made the claim you are arguing against so what’s the point?

English

François Chollet@fchollet·21h

One of the biggest misconceptions people have about intelligence is seeing it as some kind of unbounded scalar stat, like height. "Future AI will have 10,000 IQ", that sort of thing. Intelligence is a conversion ratio, with an optimality bound. Increasing intelligence is not so much like "making the tower taller", it's more like "making the ball rounder". At some point it's already pretty damn spherical and any improvement is marginal. Now of course smart humans aren't quite at the optimal bound yet on an individual level, and machines will have many advantages besides intelligence -- mostly the removal of biological bottlenecks: greater processing speed, unlimited working memory, unlimited memory with perfect recall... but these are mostly things humans can also access through externalized cognitive tools.

English

264

150

1.7K

211.7K

Victor Garcia@vmgdev1·5h

I was extremely clear, my argument is even boring. *I* was not trying to make a point about the how close a group of humans are to "optimal" intelligence like Fchollet was. The point I was trying to make is that there is no possibility of magical unbounded omniscient intelligence.

English

Brandon Fishback@Bfish94·5h

@vmgdev1 @tobyordoxford @fchollet Maybe you’re just not being clear. Also, no one is making the argument that AI will inherently grow unbounded indefinitely. The argument is that there’s no reason to think intelligence will only be marginally more intelligent.

English

Victor Garcia@vmgdev1·5h

@Bfish94 @tobyordoxford @fchollet No, it does not. I replied to a specific claim about the ability for intelligence to grow unbounded. You read into my argument that I was claiming AI intelligence will not be much smarter than humans. It's interesting how many people are misreading it though.

English

Brandon Fishback@Bfish94·5h

@vmgdev1 @tobyordoxford @fchollet Your argument has the same structure.

English

Victor Garcia@vmgdev1·5h

@Bfish94 @tobyordoxford @fchollet Like the other response, you are arguing with a strawman.

English

Brandon Fishback@Bfish94·5h

@vmgdev1 @tobyordoxford @fchollet “If travel speed grew unbounded, it would be faster than the speed of light, which is impossible. Therefore future tech can’t go much faster than a rocket.”

English

Victor Garcia@vmgdev1·5h

1. You're arguing against a strawman. 2. You need to understand that @fchollet's definition of intelligence doesn't include include knowledge, skills, or working memory. Stockfish can beat every human at the world at chess, stockfish is not more intelligent than humans. A more obvious example is a calculator can add numbers faster than a human, it's still not more intelligent than a human. Most humans don't actually care about intelligence but skills for obvious practical reasons. It's also important to understand that there is a maximal theoretical upper bound to intelligence because we can dismiss a lot of absurd predictions of how the technology will evolve.

English

Unrealrealist⏸️@PDoomOrder1·6h

@vmgdev1 @tobyordoxford @fchollet Sure it is possible that for many tasks of practical importance there are diminishing returns with increased "intelligence" to the point of saturation. That is an extremely different claim than the hairless monkeys that popped up a few 100k years ago are anywhere near the limit.

English

Victor Garcia@vmgdev1·8h

@tobyordoxford @fchollet If intelligence could grow unbounded at a certain point it would be equivalent to omniscience which itself is impossible because it would violate causality.

English

265

Toby Ord@tobyordoxford·12h

@fchollet This might be true, and would be very important if true. But your tweets don't provide much reason to think it is true. Do you have a particular extended piece that makes the case?

English

2.3K

Victor Garcia@vmgdev1·1d

@alth0u @postmilque The buildings from the midjourney photo are ornate (other than the one that looks like a heat sink) and don't look like the modern block apartment buildings. But they also look like office buildings not apartment buildings.

English

alth0u🧶@alth0u·1d

@vmgdev1 @postmilque wdyt!

Polski

alth0u🧶@alth0u·1d

insane reading failure op offers a preference guy who is calling for a new aesthetic then misreads an ornery reply guy appealing to authority saying they dont do old styles in a good way anymore, not that older styles are bad you think we get new aesthetics with this kind of critical thinking?

English

Victor Garcia@vmgdev1·1d

@alth0u @postmilque Do you think the buildings in the photo from Patrick's tweet are more like the left or the right from the original photo?

English

alth0u🧶@alth0u·1d

@postmilque it can read like it but consider

Patrick Collison@patrickc

@2024dion Midjourney, as always, is a source of thought-provoking ideas...

English

589

Victor Garcia@vmgdev1·1d

@can I can't tell if you are praising or insulting the project

English

can@can·1d

Honestly up there with the “Claude wrote a compiler” demo

Cheng Lou@_chenglou

The engine’s tiny (few kbs), aware of browser quirks, supports all the languages you’ll need, including Korean mixed with RTL Arabic and platform-specific emojis This was achieved through showing Claude Code and Codex the browsers ground truth, and have them measure & iterate against those at every significant container width, running over weeks

English

617

can@can·1d

woah what

Cheng Lou@_chenglou

My dear front-end developers (and anyone who’s interested in the future of interfaces): I have crawled through depths of hell to bring you, for the foreseeable years, one of the more important foundational pieces of UI engineering (if not in implementation then certainly at least in concept): Fast, accurate and comprehensive userland text measurement algorithm in pure TypeScript, usable for laying out entire web pages without CSS, bypassing DOM measurements and reflow

English

13.3K

Victor Garcia@vmgdev1·2d

@tanishqk @GergelyOrosz Are you stupid?

English

Tanishq Kancharla (tk) 🌊@tanishqk·2d

@GergelyOrosz why is it a problem to have models trained on your code?

English

1.8K

Gergely Orosz@GergelyOrosz·2d

If you use GitHub (especially if you pay for it!!) consider doing this *immediately* Settings -> Privacy -> Disallow GitHub to train their models on your code. GitHub opted *everyone* into training. No matter if you pay for the service (like I do). WTH github.com/settings/copil…

English

388

927

5.1K

538.2K

Victor Garcia@vmgdev1·5d

@ben_golub @alexolegimas @JesusFerna7026 Knowledge is only retained after repeated examination, a course without the pressure to study and get good grades isn't enough for people to learn.

English

179

Ben Golub@ben_golub·5d

extracting for broader discussion @alexolegimas @JesusFerna7026

English

215

25.1K

Victor Garcia@vmgdev1·22 Mar

@Ikryanov @theo Not inside a TUI

Français

161

Vladimir Ikryanov@Ikryanov·22 Mar

@vmgdev1 @theo JS doing rendering with a GPU for about 15 years already.

English

184

Theo - t3.gg@theo·22 Mar

T3 Code uses half as much RAM as Claude Code. Claude Code CLI: 635.5 MB T3 Code: 350.9 MB Yes, our Electron app is 2x more efficient than a CLI written with Bun.

English

180

2.6K

165K

Victor Garcia@vmgdev1·22 Mar

It should, I was just trying to point out that while most people consider the browser to be "not native", it is in fact very well optimized c++ with gpu acceleration for rendering. The obvious reason CC is slow is that they are using vibe-coded React which is inherently slow because of its design and has many foot-guns that can make it even slower.

English

325

Shinobu@Shinobu_uwu·22 Mar

@vmgdev1 @theo a cpu should do just fine in a 80x24 scene, it's just plain incompetence at this point

English

357

Victor Garcia@vmgdev1·18 Mar

@_Felipe I think you just have to let something fail or slow down and what you choose is based on the specific circumstances. Choosing buffer sizes is probably just based on how much you can spend on ram and measuring?

English

Felipe O. Carvalho@_Felipe·18 Mar

Is there no good canonical article on "Backpressure" applied to Computer Science? The wikipedia one talks about fluids and pipes. It's related but bad as a citation.

English

1.5K

Victor Garcia@vmgdev1·17 Mar

@kenwheeler @thekitze I only use festool and I use solid gold tenons.

English

180

patagucci perf papi@kenwheeler·17 Mar

@thekitze ryobi lmfao

English

131

6.2K

kitze 🛠️ tinkerer.club@thekitze·17 Mar

i'm getting into woodworking because god forbid i stick to one hobby for a week also fk u bougie mfs ryobi is the 🐐

English

114

21.4K

Victor Garcia@vmgdev1·17 Mar

@Duderichy Always has been

English

853

the Rich@Duderichy·17 Mar

there are two tech markets rn

Tenobrus@tenobrus

job market narrative violation: my weekly inbound from tech recruiters has ~doubled over the last 2 months

English

482

37.6K

Victor Garcia@vmgdev1·17 Mar

> You literally write that for you your codebase is something to "puzzle out" If someone else (AI in this case) wrote the code you need to puzzle it out, it's just how it is. I guess you just don't believe that I'm actually looking at the code and verifying it's doing what I think it does? It's significantly easier to read code after you have been told what it's supposed to do and why it's done that way.

English

Dmitrii@dmitriid·17 Mar

@vmgdev1 @SebAaltonen You literally write that for you your codebase is something to "puzzle out" and you don't even check to see which code is relevant to the code you are "reviewing" now, completely offloading your understanding to AI.

English

Dmitrii@dmitriid·16 Mar

Having seen the bullshit that LLMs produce even for the simplest CRUD web apps, good luck with your game engines. We'll yearn for the days of unoptimized Unreal games.

Sebastian Aaltonen@SebAaltonen

Now is the perfect time to write your own engine. The argument in the past was that you don't have time to write all the less important generic features. Now, the LLM will do that for you. You can focus on things that differentiate your product.

English

10.7K

Victor Garcia@vmgdev1·17 Mar

@dmitriid @SebAaltonen Also you're not going to be able to believe it, but I agree with you a lot more than you realize and I never tried to claim LoC was a good productivity metric.

English

Dmitrii@dmitriid·17 Mar

@vmgdev1 @SebAaltonen Of course I will not ask AI for code review because, as I showed with examples, AI bullshits even on tiny code bases, and I have enough actual knowledge about how things work to call catch that bullshit. 3/3

English

Découvrir

@tobyordoxford @fchollet @Bfish94 @alth0u @postmilque @can @elonmusk @BarackObama