Leo's amazing predictions

[Michal__]

51 minutes ago, Leo Gura said:

It costs $2+ per query.

When using cursor one question can take 10+ minutes to complete and can cost like 10$.

I have spent 50$/1 - 2 hours with Cursor. 

[Leo Gura]

3 minutes ago, Michal__ said:

When using cursor one question can take 10+ minutes to complete and can cost like 10$.

I have spent 50$/1 - 2 hours with Cursor. 

Interesting.

I hope it's worth it.

Edited 4 hours ago by Leo Gura

[something_else]

12 minutes ago, bazera said:

Yes me too, but how will new people become seniors without going through that junior experience first.

Strange times.

Yea, exactly. It's going to be interesting to see how this pans out.

We've completely restructured our entire team now as a result of AI. Instead of having two squads of 4 devs each working on one feature, now every dev is working on one feature in a 'pod' with a designer and a product manager. Even if the feature is massive, it gets one dev. Because of AI assisted coding this is manageable, and actually we're all loving it because we get a ton of autonomy and ownership over features that we didn't get before.

For a while, when we were working in squads of 4, it felt like devs became meat interfaces to Claude. The product team came up with a feature, designer designed it, tech lead broke it down and wrote tickets for us and then we basically pasted the tickets into Claude.

Now each dev has complete ownership over the lifecycle of their feature, manage their own tickets, breaking it down however they like, architecting a solution, working closely with product and designers etc. I think this may be how the future of software engineering is going to be, moving away from the complexities of writing code and more towards system design, consultation, and collaboration.

[Leo Gura]

The problem I find with using AI to do creative work is that you end up wasting way too much time trying to get the AI to create what you want, and it never really creates what you truly wanted. But you wasted 10 hours fiddling with the knobs. By the time you invested 10 hours into it, you're hooked. But it isn't really creating what you wanted.

Such a process is the death of creativity and spirit. Rather than being truly creative and skillful, you are just manipulating knobs to get half-assed approximations of something good.

Using AI is almost like gambling. At least that's the case for image generation. It takes hours of trial and error just to create one usable image.

Edited 4 hours ago by Leo Gura

[Leo Gura]

Correction: Earlier in this thread I said the gold EFT is called GDX. That was a mistake. It is called GLD.

Edited 4 hours ago by Leo Gura

[something_else]

14 minutes ago, Leo Gura said:

The problem I find with using AI to do creative work is that you end up wasting way too much time trying to get the AI to create what you want, and it never really creates what you truly wanted. But you wasted 10 hours fiddling with the knobs. By the time you invested 10 hours into it, you're hooked. But it isn't really creating what you wanted.

Such a process is the death of creativity and spirit. Rather than being truly creative and skillful, you are just manipulating knobs to get half-assed approximations of something good.

Using AI is almost like gambling. At least that's the case for image generation. It takes hours of trial and error just to create one usable image.

This is very much the case for image generation, it is almost good enough to be amazing for specific applications but always falls just short. I promise you that this is not the experience with coding agents in most cases. They're flexible enough to do exactly what you want much quicker than you could without them. They make mistakes, but if you point them out and give them more direction, they'll adapt and do what you want.

Right now I'm working on a quick prototype for a new editor interface we are building at work and it's the most creative I've felt during working hours in a while. I have ideas for features I want to add, I ask the AI, it implements them, they work. My mind is entirely focusing on ideas, and implementation is secondary. This code will never reach production, but in ONE day I have an extremely detailed, functional prototype that proves the tech is viable for a production-level implementation. This prototype alone would have taken a team of four people 2-4 weeks to complete three years ago.

[Leo Gura]

23 minutes ago, something_else said:

This prototype alone would have taken a team of four people 2-4 weeks to complete three years ago.

Okay, but if everyone can do that, it has very little value.

Anything that's easy becomes worthless in business.

The market will get flooded with mediocre AI slop creations. Why would they be worth anything?

Today you still have an edge because AI is so new that 99% of people don't know how to use it yet. But in 10 years every fool will be using it to create slop.

I don't think you guys quite appreciate how serious this value-add issue is. If AI can truly do what humans do, then humans have no value-add, which means humans are no longer necessary. You better hope for your sake that AI cannot replace you. Because if it can, you're as good as dead. Not because it will kill you, but because you are useless.

If you think that you will get to be useless and still survive, you're kidding yourself. Useless things don't survive for long.

Do you really think these billionaires care about helping useless things survive?

Edited 4 hours ago by Leo Gura

[Natasha Tori Maru]

@something_else honestly, you sound blessed to be employed in a workplace that has adapted and restructured quickly as a result of advances in AI. Do you imagine your job would be less satisfying without such quick evolution? Cheers for sharing too

Edited 4 hours ago by Natasha Tori Maru

[Elliott]

It's kind of like low cost manufacturing, it will be utilized for a lot but there's still going to be frontier or critical niches for more hands on coding and new people learning. More things should be automated too, so we should see more programming, 'Jevons Paradox'.

You're not going to vibe code a Mars lander, all weather models, all geology models,.... things ai can't 100% test.

I know they are vibe coding weather models, but you'll still have manually coded ones too. You'll have companies that just troubleshoot bugs, rather than turnkey development, new people can learn in that business.

Edited 3 hours ago by Elliott

[something_else]

1 hour ago, Leo Gura said:

Okay, but if everyone can do that, it has very little value.

Anything that's easy becomes worthless in business.

The market will get flooded with mediocre AI slop creations. Why would they be worth anything?

Today you still have an edge because AI is so new that 99% of people don't know how to use it yet. But in 10 years every fool will be using it to create slop.

I don't think you guys quite appreciate how serious this value-add issue is. If AI can truly do what humans do, then humans have no value-add, which means humans are no longer necessary. You better hope for your sake that AI cannot replace you. Because if it can, you're as good as dead. Not because it will kill you, but because you are useless.

If you think that you will get to be useless and still survive, you're kidding yourself. Useless things don't survive for long.

Do you really think these billionaires care about helping useless things survive?

This is why a lot of people are scared, and unconvinced that this is a bubble where the outcome is going to be a stock market crash. AI seems to be good enough where it's very possible that it will upturn the entire economy by wiping out 50%+ of our white collar jobs. Is that a stock crash? I guess in a sense.

Edited 2 hours ago by something_else

[Elliott]

37 minutes ago, something_else said:

This is why a lot of people are scared, and unconvinced that this is a bubble where the outcome is going to be a stock market crash. AI seems to be good enough where it's very possible that it will to upturn the entire economy by wiping out 50%+ of our white collar jobs. Is that a stock crash? I guess in a sense.

The bubble is the rapid and oversaturation in data centers, not that it won't be utilized. There's only so fast that the world can adapt, and the data center boom is much faster than that where we're seeing skyrocketing electricity and natural gas rates, these rates hurt every other industry, and consumers. Electricity is expected to go up 20% in cost, natural gas 30%.

I think they're building data centers as fast as possible because they'll probably get banned because of energy use when the economy crashes.

Edited 3 hours ago by Elliott

[Leo Gura]

1 hour ago, something_else said:

by wiping out 50%+ of our white collar jobs

That is not happening with current chatbot.

Only if AGI.

[StaraX]

@Leo Gura What is your prediction for the emergence of AGI? Have you thought about it?

[Leo Gura]

10 minutes ago, StaraX said:

@Leo Gura What is your prediction for the emergence of AGI? Have you thought about it?

I don't think anyone knows. It's all just guess-work.

AGI could be 5 years away, or 50 years away, or 500 years away. Or maybe computers will never be able to do it. I don't know.

If AGI happens, you should expect bad shit to happen. I don't mean that AI will hurt us. I mean humans will eat each other alive. So I am hoping AGI will not happen soon. Our political situation is not mature enough for something so serious.

The bubble popping is the best case scenario here.

Look, in the end, your guess is as good as mine. On this AI topic I have no particularly special position to speak from, like I do on the topic of God. Mostly I just want to avoid losing a bunch of money in a hype cycle.

Edited 2 hours ago by Leo Gura

[Natasha Tori Maru]

I don't personally see a development / evolution path from current AI to AGI.

[Butters]

3 hours ago, something_else said:

But you can get around this by asking Fable to do the planning, and then ask it to spawn Opus 4.8 subagents (or even Sonnet or Haiku) to actually implement the code. This way you get the brains of Fable coordinating cheaper models to do the bulk of the token-heavy work.

Wait, what? Aren't the sub agents that it spawns already those?  I thought the sub agents is the reason it eats so many tokens. 

I've just been asking fable to write plans into .MD files and sending those to Opus 4.8 /w ultracode myself.

[something_else]

35 minutes ago, Leo Gura said:

That is not happening with current chatbot.

Only if AGI.

Calling it a chatbot at this point is disingenuous. It functions as a very capable reasoning engine in most real world applications beyond consumer level

[something_else]

3 minutes ago, Butters said:

Wait, what? Aren't the sub agents that it spawns already those?  I thought the sub agents is the reason it eats so many tokens. 

I've just been asking fable to write plans into .MD files and sending those to Opus 4.8 /w ultracode myself.

It can spawn subagents with any model, Fable, Opus, Sonnet or Haiku. If you don't specify, I think it tries to pick a sensible value. But Fable may decide it wants to put Fable on the job and if you have a few Fable subagents running under a Fable session agent it will burn through your quota very very quickly. I like to play it safe and specify that Opus should be used for subagents.

Though I also do what you do sometimes as well; ask Fable to create an MD file and then put that into Opus. This works better for really big tasks where having a papertrail of MD specs can be useful.

[Leo Gura]

16 minutes ago, something_else said:

Calling it a chatbot at this point is disingenuous. It functions as a very capable reasoning engine in most real world applications beyond consumer level

I would argue that calling it intelligence is disingenuous because it has memorized 2000 years of human knowledge. That allows it to fake intelligence very very well.

It is like you trained a computer to memorize the answer to every multiple choice test that has ever been written. And now it sure seems like it can pass every test. But that is disingenuous because it doesn't actually know how to pass a test.

There is a fundamental disconnect between intelligence and cramming LLMs with more data. These are not the same thing at all. No amount of data makes you intelligent. But if you memorize all of Wikipedia, you sure can pass as intelligent in most circles. But not in my circle.

Edited 1 hour ago by Leo Gura

[Andrey]

1 hour ago, Leo Gura said:

I just want to avoid losing a bunch of money in a hype cycle.

 

Here is the mindfuck, by trying to avoid losses you will actually lose money