HNHacker News
TopNewBestAskShowJobs

Herval_freire

97 karma · joined March 30, 2023

https://linktr.ee/herval

Contact info:

Address: 6363 Christie Ave, APT 2302, Emeryville, CA 94608

Phone: (917) 704-0699

email: hervalfreire@gmail.com

Media relations: Michelle Veronese

(https://www.linkedin.com/in/michelleveronese)

submissionscomments
Herval_freire··on From deep to long learning?
Someone in a previous comment on LLM research said that according to what he knew about LLM research that we were in a local maximum that no further improvement was likely possible.

I disagreed with him, and this article is evidence that is in favor of my point. If research like this continues to move forward LLMs will improve at a rapid rate.

Different threads attract different groups of people with different areas of expertise. So I will sort of reiterate this topic here as I'm interested. What are most people's thoughts on this "local maximum" thing have we actually hit a dead end? Especially given the proliferation of effort towards producing research like the one showed in the topic here.

Herval_freire··on I wish GPT4 had never happened
Right so if I'm a janitor and I use chemicals to clean things which is the result of hundreds of years of human ingenuity and chemical engineering I suddenly owe humanity money?

No. That is the perverted idea. I'm simply talking about fair as in fair trade. As in what is defined as a fair trade and a fair transaction according to economic theory and common sense. I am using the absolute most normal and most common sense meaning of the word "fair."

You are the one in your words: "perverting" the meaning. Perversion is actually too strong of a word. I would say you are definitely twisting the meaning and using an uncommon and sort of made up definition.

According to the most common usage of the word fair those who own AI only owe what they paid for. If they give away money then the economy is becoming more socialist or in another words an even distribution of wealth, but certainly unfair in the eyes of capitalism and fair trade and the fair exchange of goods.

Keep in mind though I'm not blindly supporting capitalism. The future of society may rely on a more practical but more unfair distribution of wealth. But we have to face the fact that such a distribution is fundamentally unfair.

Herval_freire··on Why and how we retired Elm
What company. Worth avoiding.
Herval_freire··on Why and how we retired Elm
They had an acquisition where the code became 75% react over night due to the acquisition. This was the major turning point. Purely business, purely popularity. It wasn't technical superiority.
Herval_freire··on I wish GPT4 had never happened
What do you mean by fairly distributed? If all my machines are better than you intellectually and all my machines are stronger and faster than you physically...

Then it is completely fair for me to not give you any wealth because the machines out perform you.

People fear AI because the fair outcome is in fact detrimental for a good portion of humanity.

That is the paradoxical irony. In theory to save people you must function as a welfare state with things like ubi. You must deliberately distribute income unfairly. The practical course of action goes against our moral instincts.

Herval_freire··on How to Design Programs 2nd Edition
Yes. one of the things rust sacrifices for high performance is modularity. Rust is a better language for safety and high performance, however it is not the best language in terms of modularity.

Haskell takes the crown for modularity and safety but as a result it is less performant then rust.

Move by default is not modular. Why? Because it is a mutation. The state of your program changes after a move and your program must be structured around that.

That variable that was moved can no longer be reused in the same context so your program is less modular as a result as the definition of modularity is high reusability of modules.

Clone just means pass by copy. Make a copy and then pass the value. Pass by immutable reference works too. Just keep everything immutable for maximum modularity. But if you want to mutate things and keep things relatively modular then pass by copy (aka clone) is the best alternative.

All in all rust is typically harder to program for and harder to make elegant because of all the anti-modularity features used to increase performance and safety.

Herval_freire··on How to Design Programs 2nd Edition
Rust sacrifices modularity for performance. You're going to have to clone everything if you want to follow along with rust.

Additionally you have to use generics and trait bounds extensively for some equivalent function in rust.

Herval_freire··on U.S. and China wage war beneath the waves over internet cables
History is littered with empires. Nothing is ever stable. Both situations of multitudes of nation states and a single monolithic empire don't have staying power.
Herval_freire··on U.S. and China wage war beneath the waves over internet cables
The scenario you declare here is basically option 2. Some sort of cold war stalemate between two super powers.

The US doesn't want this. China wants more then a stalemate.

You are right, none of these are options by choice, they are only possible outcomes.

Herval_freire··on U.S. and China wage war beneath the waves over internet cables
Yes. It's still the underdog though. But china is trying to build something to rival the US.
Herval_freire··on U.S. and China wage war beneath the waves over internet cables
You can also ask chatGPT for some good recs. I say this here because it looks like this was already done.
Herval_freire··on U.S. and China wage war beneath the waves over internet cables
This is definitely a realistic take . Many people don't agree with you and prefer to see china as the enemy though.

I would say that the view point is a little biased as it doesn't mention all three possibilities.

1. US world empire

2. Equal international relations between china and the US

3. Chinese world empire

3 Is the thing not mentioned in your passage. Like the US, china itself is not some morally idealistic state, when given an opportunity for dominance they will take it.

Herval_freire··on China plans to ban exports of rare earth magnet tech
Yes agreed. That's a huge benefit. But it also comes with a downside. The fix for the mistake can immediately by unfixed in a subsequent election.
Herval_freire··on How to be a -10x Engineer
>Pick one. Can't be both.

Pick what? I didn't pick anything. I'm just laying out examples.

>Not backwards. In terms of value, 1x is a value produced by 1 typical developer. -1x is a decrease of value.

Yeah so my point is -1x is Not arbitrary. You said it is. I simply said it isn't.

>Organizations lose knowledge all the time because people quit because of e.g. toxicity in the workplace (brought by -10x employees).

That's one way of looking at it. It's valid. But by itself I was referring to an honest employee acting to the best of his capability. Like IQ this part cannot be negative. Things like side effects of a bad actor deliberately damaging the company is not something I was referring to. Overall though this type of thing is hard to measure and anecdotally rare.

I mean if you want to factor in terrorists who bomb the tech and kill everyone in the company into your model, sure. Terrorists have -10000x productivity.

>I guess I'm done with this conversation, it no longer interests me.

Sure. That's fine. The conversation is getting into the nitty gritty of interpretation of words which is pointless imo. I think you're pretty clear about what I'm talking about. All in all though my statistical interpretation is still ultimately correct.

Herval_freire··on China plans to ban exports of rare earth magnet tech
Exactly. The answer is not so clear cut black and white. Centralized governments tend to be more efficient but lack the checks and balances of a democracy. Singapore, for example, is a good showcase for centralized government.
Herval_freire··on How to be a -10x Engineer
Bring a project to a stand still with the term -10x? The term itself doesn't make sense.

I know the article is generally just talking about bad engineers. However the topic I started here is more about the term 10x.

>And I have seen this multiple times. People who in one way or another managed to slow down other people. Toxic people can create such an environment that the productive people quit. They don't bring smaller value, not even zero value, they bring negative value to the team/project.

Sure but negative progress or -10x progress or 10x progress is just not realistic. The 10 in the term 10x is the thing that is arbitrary. Not only is it arbitrary it's completely unrealistic.

>I don't know where you come with this arbitrary interpretation of exactly -1, but yes, that happens to projects.

Arbitrary definition of 1x? 1x is default -1x would be the default but backwards. It's anything but arbitrary as it's the smallest whole number metric for negative performance. The 10 on 10x is again the arbitrary constant here. Why 10? is it because we have 10 fingers or use base 10 in our number system? 10 is Arbitrary.

Also, Losing knowledge and intelligence is the equivalent of brain damage so are people banging their heads against walls and shooting up on brain damaging drugs? Doubt it. Productivity stops at zero, negative progress is so rare it's basically non-existent, that is unless you think people are regularly suffering from brain damage.

People who use the term 10x or -10x as some kind of relatively accurate metric (like many in this thread are doing) aren't applying good intuition or statistical knowledge to the situation. 2x for really good engineers and 0x for really bad engineers are the more technically accurate term. I get the usage of 10x as purely an expression but that is Not what the topic of this thread as evolved into.

Herval_freire··on China plans to ban exports of rare earth magnet tech
Except this isn't true. Trump was voted into power by the lower middle class simply for having rhetoric that was different from the status quo. He was "real" and his style largely resonated with an america who was tired of politicians acting fake and tired of digging through all the complexity of the economy and politics to figure out what was wrong with the country that caused them to work two jobs just to support themselves.

Trump instead lowered taxes for the rich. He was one of the rich, and he served himself. It was his rhetoric and dramatic conversational style that essentially caused people to unknowingly vote against their self interest.

I apologize for using the word "stupid" here. People and voters aren't stupid. Trump, however, was indeed an example of people not using the best strategy in the name of their self interest.

Herval_freire··on China plans to ban exports of rare earth magnet tech
Hence the term 2nd hand. China won't be able to get chips directly. They have to go through secondary sources.

Similar to how the US illegally obtained titanium from Russia for development of the sr71.

Herval_freire··on China plans to ban exports of rare earth magnet tech
I lived in both the US as a citizen and china as well.
Herval_freire··on China plans to ban exports of rare earth magnet tech
>There are deposits of these rare earth minerals elsewhere. There are always more deposits, it’s just that they’re untapped until we deplete our cheaper sources. If China withholds its minerals, we will use more expensive sources, but we will still get the minerals.

Same thing if you with-hold semiconductors. Somebody for sure is willing to sell general purpose chips to China second hand at a more expensive price.

>I am less convinced that China is really good at playing the long game, and more convinced that China is subject to many of the same political struggles that we see in the US.

Unlikely. China is subject to different struggles but they are much better at playing a long game given that their power is more centralized and has longer terms. Xi Jing ping is likely here to stay. Bidens term is almost up and the next guy may just undo everything he did.

Herval_freire··on How to be a -10x Engineer
I didn't say it has to mean that. 1x is the average. We agree on that. The std can be anything. But the max possible std is 1x. that's all I'm saying.

You can now define the std to be whatever you want up to 1x. So I take the most ludicrously large std then interpret what the term 10x means in terms of the maximum std.

10x is in this case, 10 std from the norm IF 1x was the std. So the 10x concept is ludicrously improbable. The real std is likely much lower than 1x this making the existence of a 10x engineer even less likely.

This isn't arbitrary. This is the definitive interpretation using statistics. Science so to speak. If you like to interpret 10x under some crazy uneven scaling or use some other methodology other then science, be my guest. But such actions wouldn't be conducive of a 10xer would it? Which further supports my point.

Herval_freire··on How to be a -10x Engineer
Negative IQ? The test should be the definition of IQ, the statistical distribution doesn't define IQ. If the statistical distribution is saying you can have negative IQ but it is fundamentally impossible to score that amount on the test then the statistical model is the one that doesn't make any sense.

In other words the statistical model is not an entirely accurate model of the IQ test. Which to me is what IQ is. If negative IQ exists what does it even mean? These numbers have to have some physical actualization. It doesn't make sense for IQ to be some measure of some factor that's impossible to ever actualize.

That's just my interpretation though. Regardless of this though, it is highly highly unlikely for engineering productivity to go that deep into the negative. It's also fundamentally impossible to have infinite productivity too. The real domain is for sure some fixed range of numbers which makes the std for sure limited by the distance of the mean to the first number in the range.

By this logic, A 10x engineer is analogous to someone with 1000 IQ or 58 feet in height. We have intuitions about this that is inline with the statistical outcomes I outlined above... and it makes sense to apply these intuitions to the concept of 10x engineers.

Herval_freire··on How to be a -10x Engineer
It's not arbitrary. Theoretically the max range of the std is infinite. However because the input values are limited in range the std also becomes limited. You're thinking of it in terms of an infinite range from negative infinity to positive infinite. In terms of what we are dealing with practically things like IQ, height or engineering effectiveness are on a scale of 0 to infinite.

So for IQ, or height, no one can have negative or zero height. This limit effectively limits the std. It is not arbitrary. Try it.

For a distribution of integers on a range from 0 to infinite, let's say the mean is 1. The std will also have a maximum value of 1.

If you add extreme data points to change std, you will also change the mean.

This makes sense right? If you add say 9999999 to a distribution of numbers to keep the average the same you have to counterbalance it with a -9999999 or something like that. But however because the domain is limited from 0 to infinite you can't do that. So effectively the maximum value of the std is the mean itself assuming that the system is normally distributed.

Herval_freire··on Mindset shifts for functional programming (with Clojure)
It doesn't make any sense to convert any arbitrary recursion into tail called optimized.

If the recursion can be tail called optimized then yes the loop is the optimized performant implementation.

But if the recursion fundamentally utilizes the call stack then the reverse is actually true. The recursion is now the performant implementation of a for loop. So a loop conversion optimization actually doesn't make sense here. That's partly why a compiler won't optimize this.

Why? Because in recursion you utilize the call stack, in the for loop you're going to create a heap allocated stack and allocate on it repeatedly. Allocation slows down the iterations.

The only advantage of the for loop in this case is that there won't be stack overflow, but overall the recursive version will actually be faster.

Herval_freire··on Mindset shifts for functional programming (with Clojure)
>Nobody has yet convinced me that recursion has any sustained advantage over looping.

Recursion gives you a stack by default. You don't have to explicitly think about the stack. In looping the stack must be explicit.

Recursion and looping are the same thing. Recursion can be mechanically translated to a for loop and a stack, the concepts are isomorphic.

Herval_freire··on How to be a -10x Engineer
Another thing to think about is how basically every human trait follows a normal distribution.

If mean = 1x (by definition 1x should be the mean) and the std is the maximum possible value (also 1x). Then 10x would be 10 std past the norm.

This makes my numbers not arbitrary. 2x or 3x would only be 2 or 3 std past the norm. Which is reasonable (only if the std = 1x which is likely not the case and is most likely lower).

If a 10xer existed, That's like saying you found a human who runs 10x faster then the mean (10mph) or you found a human 10x taller than the mean (5'10"). 10 is arbitrary but 2 or 3 is less arbitrary. Ever heard of the company 3sigma? The name represents Rarest non arbitrary anomaly from the norm.

It is actually fundamentally illogical to think of human ability in terms of a multiple of the mean because no other measurable human quality follows that kind of distribution. Almost all of these measured things follow the normal distribution.

You are therefore highly highly unlikely to find a 10x engineer or a -10x engineer. This applies even to things like engineering leadership.

When viewed from this lens you can actually see that it's actually very unlikely for even the best leader to improve the productivity of every member of his team by even 2x. So I take that back. Likely what we are seeing is something like a 1.1x.or 1.2x increase and our minds are exaggerating the effect like an optical illusion.

The concept of measuring any human trait along multiples of the mean via 1x or 2x or 3x or 10x is fundamentally unrealistic when you take a data driven viewpoint of normal distributions. Everyone on this thread is erroneously following this model and while my instincts were more tempered, to a certain extent even I'm following this erroneous model as I declared 3x to be within the realm of reality.

We are all going about this all wrong (including me), the data driven conclusion is that at best you can find 2x engineers and they are freaking rare (because the actual std is nowhere near 1x). The maximum likely delta between the best engineer and worst engineer is 4z, where in this case the z represents 1/2x or an engineer half as good as the average.

Likely people are anecdotally seeing this 4z delta and their mind is exaggerating the delta to 10x. From what we know of statistical reality an actual 10x is unlikely to happen at all.

Herval_freire··on How to be a -10x Engineer
Every competent and qualified engineer should be able to do the job of 10 competent and qualified engineers?

The statement that every engineer can be a 10x engineer is recursively paradoxical. If a 10x engineer existed, in my mind, he would immediately be able to identify this logical error.

Herval_freire··on Playing with Fire – ChatGPT
I'm sorry, but that statement is not accurate. Oxygen is an element that exists naturally in the universe, and it was not invented or created by humans or any other living organism. However, it is true that oxygen has played a crucial role in the evolution of life on Earth. Photosynthesis, a process by which plants and other photosynthetic organisms produce oxygen, has had a profound impact on the composition of Earth's atmosphere and the development of complex life forms.
Herval_freire··on Playing with Fire – ChatGPT
>I also don't agree about ChatGpt being the inflection point. The capabilities of Gpts were known for years, but chatGpt popularized them because it made it so easy to use

You mean LLMs not GPTs. chatGPT was an inflection point in terms of publicity but also in terms of additional Reinforcement training that made the model highly highly usable. chatGPT was the first model that was incredibally usable (and I don't mean usable in terms of like a GUI or UI, usable in the sense that the AI was actively trying to assist you).

>That is not evidence that AI will destroy humanity. The fact that Ai is increasing in capabilities also means that it is increasing its capability to align with humans, no? I don't get why the reverse is considered as the sole and inevitable conclusion

The article didn't say humanity will be destroyed by AI. It's more saying society will dramatically shift and it implies that a lot of humans will suffer as a result. The shift is dramatic enough that while maybe apocalyptic is a bit too extreme of a word it's not entirely unfitting.

Herval_freire··on How to be a -10x Engineer
Oh man. My eyes just skipped over that negative sign every time. Now it makes sense. I thought the article was sarcastic. I'm blind.
← PreviousPage 2 of 4Next →