I don’t know why it makes me feel so calm looking at this painting. Something just feels normal and boring but a great sense of peace. Fast forward a few weeks, after a multiple week trip to Tokyo, I think I know why: a sense of solitude I haven’t experienced in the busy streets of Tokyo.
I don’t feel that Safari on iOS inferior to other browser engine (comparing to Android). It has great battery utilization and performance. But I do think if they separate the browser engine update from the entire OS update, it will open the flood gate for other browser engines.
Agree. I think for the limited scopes such as in a mod/function, using Rust's borrow checker semantic makes sense as it is easier to fit in our mental model. If you have to pass those data structures around to the outside of your limited scopes, I would prefer Rc/ARc more. I also think the style guide has a catch all phrase there when it says "OK to use Rc/ARc depending on the complexity...". For me, having to pass a shared reference around is enough complexity to reason about.
I just don’t understand the use of a decoder model to make a classifier. Jev model will give you probability scores for each of the classes/selections you ask for. The probability is actual statistical probability that a selection is right.
Using a decoder model for this comes down to picking the most probable class based on the probability what the next token assigned to a class is. They are all probabilities but semantically mean totally different things. Am I right?
I also didn’t see Ms mentions that they use Qwen as the base model. Somehow I saw a comment here and thought it was the case. Regardless, my parent comment is about other decision models that based on decoder model.
I assume Mike use the creative mode instead of survival mode. I tried survival mode and it feels too stressful for me. Even in creative mode, I struggled very much to build such elaborate landscapes a lot of folks can do. I’m wondering if this Minecraft is something you either love or hate it and nothing in between for you to learn?
What platform of Minecraft (console, pc, mobile) are you playing on?
I mean, PC is fine. I don't know what would be stressful in creative. Either you feel like creating something elaborate, in which case it's easy, or you don't, in which case what's the problem? There's no reason that you 'should' be motivated to make stuff like that.
I’ve been coding for over 20 years. I may not be the perfect coder, but for me, the real goal has always been solving problems and coding is just a side effect. If I could solve an operational problem with Excel, I would! Code, once created, is a liability I have to carry.
So I’m embracing AI-assisted coding with open arms. It’s not that I don’t enjoy coding. it’s that this approach lets me solve problems faster than ever before.
Ok but you do realize that AI can also solve problems? The thing is it's not just taking away "the coding" it's also taking away the problem solving (of course I'm not talking about all problem solving).
That's the real risk here is that we are getting to the point (probably already there for a lot of stuff) where you can just point it to your already existing codebase and tell it your problem and it can basically do it.. I mean I've tested that in practice, sometimes I'll have a clear direction of where I want to go or what I would do.. but I don't really give it any hints of how to solve the problem, and many times it comes up with a similar result or at least one that also works completely fine.
So I think part of the issue is that people like you and me I feel like are going to be so much less needed because an AI can just do it instantly for basically free.
Anyway, I know this is a slight hyperbole.. but it's also not. It kind of sucks to realize that a simple prompt can replicate your work 95% of the time and do it basically instantly.
Currently we have C-level people saying "I want a website that sells widgets". Then product managers make a plan for all the pages/functionality on that site. Designers style the site. Managers divide up the work into smaller tasks. Tech leads steer a group of software engineers, and software engineers do the actual implementation.
2 years ago AI was a novelty and essentially just "fancy autocomplete". Today a skilled software engineer can use it to mostly one-shot the whole website with a detailed technical prompt and then iterate on it from there.
What happens 2 years from now? If progress continues, I expect we'll start walking up that chain of command. First you eliminate the software engineers and just have the tech leads do the prompting. Then you eliminate the tech leads and have the managers do the prompting. Then you eliminate the managers and designers and eventually you end up with star-trek-esque programming where some C-level person is just telling their computer "I want a website that sells widgets. Good! Now make the navigation bar blue. Could you make the font bigger?"
So what are your plans for when the wealthy no longer need you?
I look at this and I feel the reverse: what do the wealthy bring to the table now?
previously we had to work for them cause anything worth doing that could become a business needed a bigger team. now, a developer doesn’t need anybody above him to make stuff, and can potentially work on doing things they find valuable which imo something we always strived to but couldn’t afford to. in my mind, it’s really about ideas and execution now. and execution i bet a developer who wants to make something can do better than anybody higher up the chain for the forseeable future.
sure there’s things you still need investment and a physical presence for. but i reckon there is money and careers to be made in purely software based businesses still, and i think that value will mostly be captured by individuals and small teams moving forward.
The wealthy bring the capital required. The problem is fundamentally the commoditization of intellectual labor. It will simply be a game of whoever spends the most wins.
even in software historically, larger teams/more money have never produced better software. sure, not disagreeing that may be they’ll have all the best LLMs and the hardware to use them for a while. But I don’t foresee open source weights to the public to lag that far behind, and hardware that can run them to become more affordable.
you know who _own_ the means of production more than ever? the rich. they control your labour environment. they control politics. so you're the opposite of independent from them.
i could setup a local llm that’s nearly frontier level if i wanted to. i doubt i’ll ever need to though since tokens are becoming so cheap. although i am well off, i am not wealthy. and i only foresee this becoming more affordable in the future. may be i am wrong but imo an era when capital is less important is already here for the purposes of what i described.
my point was that for anyone who wanted to make software they couldn’t due to lack of free time or funds, they likely can give that a go now.
Then I will find other problems to solve. Do you think AI will solve every problems for humanity? it has nothing to do with coding as I mentioned earlier.
Just because there are open problems AI cannot solve, it does not mean you can make meaningful contributions. AI might become more capable than you in every way. Say AI cannot solve P vs NP. You can dedicate your life to the problem and will not make progress. Similar for more worldly problems like new treatments. Or societal problems. You are the cow.
The internet enabled the massive complex systems we have today (public cloud being the largest in scope probably, but there are sprawling libraries like everything around JavaScript, modern web browsers themselves maybe).
Without the internet these systems can not be maintained, we need all information about them available instantly. We also need people sharing how to solve problems at a global scale which the internet enables.
Personally I think LLMs will just allow even more complexity. This will create more situations that need human intervention to solve problems. And also new problems we have never seen before.
Hah I have no idea. If we reach AGI in 10 years we will still have problems. Otherwise we all would be enlightened by then, right? Until then I think mental health would still be a problem. Maybe I should get a therapy license now.
> Then you eliminate the managers and designers and eventually you end up with star-trek-esque programming where some C-level person is just telling their computer "I want a website that sells widgets. Good! Now make the navigation bar blue. Could you make the font bigger?"
Those C-level people won't want to do that. Plus in a competitive economy everyone else can do the same so there will be competition to remain relevant.
1 website won't be enough, you'll need 10 and each site needs 10 times the features or your buyers will look elsewhere.
At the end of the day the C-level person delegates to a techy person, it's not the C suites job to care about font sizes or navigation bar colours.
> Ok but you do realize that AI can also solve problems?
Why have companies with infinite amounts of AI and expertise not deployed comprehensive end to end solutions for business problems?
Every solution they've come up with is "here's a tool. The solution to make money from it is left up to the reader". Why would they leave all that money on the table?
What you said is valid but I think you are still constrained in thinking that the code is the solution, though. I wish I could word it a better way. Maybe we need a word calculator /s
It's not right a lot of the time (yes, even the latest models) based on the problem you're trying to solve for your users. (That's one thing we can keep doing).
It doesn't always know the right thing to fit the need, so you can now figure out what that looks like and design it appropriately (and try 10 options in production and see which one wins!)
But also you can just do a lot more work. If the prompt can replicate 95% of your work, now you can do keep moving forward faster and do a lot more work.
This does mean, of course, that there will be less of a need for software engineers that are only capable of reasoning about the edges of applications or one narrow part of the whole discipline will get squeezed out. If a computer can do the job, it's now less valuable of a skillset than it used to be in the market, and that hits wages hard if your only contribution to the market was "I know how to create React components when someone tells me what is in them, and how they need to interact with other components"
Happily, despite the amazing productivity of AI, there is still so much absolute CRAP software out there, and so many different places we can apply better options. Bureaucracy and poor management will continue to create new opportunities.
My guess is that they use an encoder-only model as the foundation and then do RLDC. Why? 1) it doesn’t need text generation 2) limited context window (40k last time I check)
reply