News for hackers. Stuff that matters.
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by Y_Y on Saturday May 22, 2021 @04:33AM (#27244688)
I think hero-worship in any field is weird and unhealthy. That said, Bellard is astonishingly talented and prolific and it's hard to look at someone like that and say he's just a mortal who tries his best to write code, just like everyone else.

So what is it that causes this big gulf in productivity? I know that as a programmer I have all sorts of internal and external obstacles to just sitting down and churning out entire projects, let alone something useful with clever and high-quality code. Is a 10x rockstar just somebody who's been able to eliminate most of their obstacles, and has some amount of talent left over to solve problems with?

Re:NNCP: Lossless Data Compression with Neural Networks (2019) by chii on Saturday May 22, 2021 @05:02AM (#27244838)
> it's hard to look at someone like that and say he's just a mortal who tries his best to write code, just like everyone else.

would anyone look at Usain Bolt and say that he's mortal and just like everyone else? Why is it so difficult to imagine that there are people who are just so much better than you, that you (an average person) can never catch up?

Most people have no trouble understanding that they cannot reach the level of Usain Bolt. Why should programming be any different from sprinting in this case?

Re:NNCP: Lossless Data Compression with Neural Networks (2019) by abiro on Saturday May 22, 2021 @06:47AM (#27245289)
Usain Bolt doesn't run 10x faster than your average fit person. It's more like 2x. So it's fair to ask if there are factors other than innate ability at play when observing orders of magnitudes of differences in programmers' outputs.
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by ZephyrBlu on Saturday May 22, 2021 @08:25AM (#27245847)
The difficulty of tasks generally increases exponentially, not linearly though. 2x in absolute terms definitely translates to 10x or more in terms of skill.
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by nl on Saturday May 22, 2021 @07:43AM (#27245578)
The combination of better performance, more consistently over a long period of time is what makes 10x

If you put Manchester City up against a team consisting of average people who are fit and know how to play soccer it's pretty easy to see Manchester City scoring 10x the goals - at least.

Re:NNCP: Lossless Data Compression with Neural Networks (2019) by SirSourdough on Saturday May 22, 2021 @01:18PM (#27248489)
Isn’t the question here more like putting Man City (elite professionals) up against average professionals? I don’t see the 10x goal differential there necessarily. It would be unremarkable for a “10x programmer” to be 10x better than a hobbyist, which is what I see as the equivalent of someone “who is fit and knows how to play soccer”.
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by roenxi on Saturday May 22, 2021 @06:47AM (#27245292)
1) Yes, Bolt is also mortal.

2) If quantified the performance gap between me (unfit, untrained) and Bolt will be smaller than the gap between me (practised programmer) and Bellard.

Re:NNCP: Lossless Data Compression with Neural Networks (2019) by itsoktocry on Saturday May 22, 2021 @09:58AM (#27246584)
>If quantified the performance gap between me (unfit, untrained) and Bolt will be smaller than the gap between me (practised programmer) and Bellard.

This is quite the assumption. Bolt is the fastest man we've ever seen. You actually think you're more closely competitive to him than Bellard???

Re:NNCP: Lossless Data Compression with Neural Networks (2019) by vlovich123 on Saturday May 22, 2021 @11:35AM (#27247515)
It may be stated poorly but what OP is trying to say is that in terms of just output, the relative time difference in the 100m dash is smaller than the relative software output between them.

Obviously that’s silly because the physical result is approached asymptotically. World class athletes train and compete their whole lives of a fractional difference in result. Clearly the output by the human brain is not constrained by something asymptotic as what the human body can physically be trained to accomplish.

Re:NNCP: Lossless Data Compression with Neural Networks (2019) by jbjbjbjb on Saturday May 22, 2021 @05:42AM (#27245014)
I would say an innate talent for ‘coding’ is a relatively minor factor. What is very rare to see is people showing the level of responsibility and grit to get on by themselves and build something. Also in a business setting it is unheard of for someone to have the opportunity or even be trusted enough to do it not to mention all the other business factors.
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by taneq on Saturday May 22, 2021 @06:06AM (#27245112)
Just like Usain Bolt’s physical attributes are a minor factor and it’s his grit and responsibility that makes him so quick?
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by jbjbjbjb on Saturday May 22, 2021 @08:22AM (#27245820)
No it is completely different the analogy doesn’t work. Coding isn’t running.
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by ZephyrBlu on Saturday May 22, 2021 @08:28AM (#27245863)
> I would say an innate talent for ‘coding’ is a relatively minor factor

I would argue that innate talent is strongly linked to genetics. In the case of coding, I think intelligence plays a large role.

Re:NNCP: Lossless Data Compression with Neural Networks (2019) by Datenstrom on Saturday May 22, 2021 @03:39PM (#27249656)
> innate talent is strongly linked to genetics

I do not believe there is much of a spread in intelligence based on genetics. Short of disability, and given the same consistency, training, and opportunities I seriously doubt that anything even remotely close to a 10x gap would be seen.

I am certainly not keen on the science of the topic though, biology is certainly my worst subject. But I have heard of no research showing such a divide.

Re:NNCP: Lossless Data Compression with Neural Networks (2019) by jbjbjbjb on Saturday May 22, 2021 @09:54AM (#27246551)
It plays a role, sure. There will be significant diminishing returns though. What I’m suggesting is there are other factors at play here that are way more important than a few extra IQ points. Completing a significant project isn’t one dimensional.
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by Cybiote on Saturday May 22, 2021 @06:55AM (#27245337)
The trick is that even a tiny difference in ability can lead to a large difference in results as the number of decision steps involved in a task increase. The expert wastes less time in search space. This is a combination of good mental models honed from experience and attention to detail.

Something unusual about Bellard is the breadth of his knowledge. There are probably a fair few who can write a fast deterministic low level code optimized neural network library. A decent number who can write an autodiff library but few who can do both. Once that's done, I daresay that's the hard part.

Given such a library, implementing an LSTM or transformer model and combining it with Adaptive Modeling/incremental learning of data statistics using a neural network to feed an Arithmetic Coder is something that is within the reach of many more.

In summary, a 10x developer is not 10x smarter or 10x more skilled, they just waste less time investigating unproductive paths and are maybe 1% less likely to take wrong turns. This ends up making a huge difference for large complex tasks.

Re:NNCP: Lossless Data Compression with Neural Networks (2019) by pansa2 on Saturday May 22, 2021 @07:24AM (#27245481)
> Something unusual about Bellard is the breadth of his knowledge.

One thing I’ve noticed is that he chooses to apply himself to many different subfields of development, but always using the same technologies e.g. all his work seems to be in either C or JavaScript.

I’ve only worked in a couple of subfields but have seriously used at least 7 different languages, plus countless libraries. Compared to Bellard I’ve probably spent (wasted?) a large amount of time trying to choose, and then learn, the “best” tool for each job.

Re:NNCP: Lossless Data Compression with Neural Networks (2019) by ZephyrBlu on Saturday May 22, 2021 @08:37AM (#27245915)
To be fair, Bellard has the luxury of doing that because he's generally writing things from near scratch and he's not beholden to other people or outside requirements.
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by foepys on Saturday May 22, 2021 @04:57AM (#27244819)
Some people are just very intelligent and can grasp concepts quickly and apply this new knowledge efficiently. To them a person of average intelligence is often like a toddler that is struggling to put the cube into the round hole.
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by logicchains on Saturday May 22, 2021 @11:51AM (#27247659)
There are billions of people in the world, over seven thousand people with one-in-a-million intelligence, seven million people with one-in-a-thousand intelligence, but nobody else close to Fabrice Bellard in software output.
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by agumonkey on Saturday May 22, 2021 @09:36AM (#27246387)
What surprises me about bellard, beside the excellency, is the breadth. Who could predict his next topic...

Surely his intellectual life is diverse

Re:NNCP: Lossless Data Compression with Neural Networks (2019) by slim on Saturday May 22, 2021 @04:54AM (#27244808)
Why is hero-worship is unhealthy? I thought it's the contrary
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by user-the-name on Saturday May 22, 2021 @07:00AM (#27245358)
Because it decreases self-confidence and initiative. It makes you think of the things that the "hero" does as exceptional and out of your reach, instead of something you could do as well if you put your mind to it properly.
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by zaphirplane on Saturday May 22, 2021 @07:23AM (#27245472)
Interestingly enough seeing powerful role models inspired schwarzenegger, bolt martial artists, poets, architects even Tom cruise inspired Air Force pilots
Re:NNCP: Lossless Data Compression with Neural Networks (2019) by opsy2 on Saturday May 22, 2021 @09:35AM (#27246377)
It seems small, but it's a big distinction between hero worship and role model!

'Hero Worship' is about elevating someone to a categorically higher level. This can lead to thinking you could never be as good, taking everything they say as gospel, defending even their poor opinions, etc.

If we -worship- 10x developers, you can maybe better see how with the emphasis on that word things can get pretty toxic pretty quick. For example, I think many would agree HR disputes should not take talent level as an input, but if one of those people isn't even a mortal...

---

None of this is necessary to have a role model, or precludes celebrating someone for doing excellent work as is this case.