ฝัง
- เผยแพร่เมื่อ 27 พ.ย. 2024
- OpenAI is about to release o1, its first model with human like reasoning capabilities. These models will be able to answer complex questions much faster than a human can. The GPT o1 model will spend more time thinking before responding, but should give more concise answers. There are several varying o1 models being released as well. The Strawberry o1 model, for example, optimizes complex code and math. Learn more about what is instore!
The problem with non-technical people using LLMs to code is that they'll always be at the mercy of the LLM. The inability to tinker or understand the output will prevent you from building out any of your ideas to specification. You'll only be able to get "close enough" and by the time you figure out how to prompt it correctly, with explicit inputs and parameters, you might have well just learned to code.
If you're not the SME you can't see the mistakes, whatever the LLM spits out. Remember, it's at its worst and will only improve.
Internal combustion engines improved its function through the years and decades and it still mostly remained as inefficient it was in terms of fuel consumption. Plateaus exist. People should just focus on fitting these tools where they can and stop speculating
Coded a game in 2h with almost non experience. I was lost with code stuff before... got me to a point to understand stuff of coding a LOT more with gpt, than by simply learning the classic way. So its easiert to code AND easier to learn with LLMs ;)
The faces of these OpenAI employees look AI generated.
Ok, now give it a 30k lines project and tell it to fix a bug.
Won't be a problem, soon.
@@wassollderscheiss33 if becomes a reality, you think it’ll be available to the masses ??? I don’t think governments will release this kind of WEAPON to the public.
@@Embedboii I think government has no clue and also has no means to stop digital things from spreading. But I WAS thinking about the big tech companies not releasing their really competent models. If you have an unlimited amount of (digital) software developers, why give it away?
@@wassollderscheiss33 agree that would be and competitive edge for them.
@@wassollderscheiss33 you are asking "why give it away free?" you are not considering one simple fact, people deep down in their hearts, are sick of this endless freaking competition already. they want to be good, they want to serve, help, give out their products for free and then cry with joyful tears about it while hugging their loved ones. and now they can do exactly that lmao
Coding never dies!
AI and global problems are growing exponentially that's why AI alone is inadequate to lead the world.
Developers are highly demanded all the time!
Don't lose hope!
Coding is far, far from dead. If you have a specific goal in mind, and explain it perfectly, it might still not work. It doesn't reason 100% correctly all of the time. The process of testing, revising, and verifying the code is still vital. Not to mention it `speeds you up`, it doesn't replace software developers any more than it replaces scientists.
Right, what people generally seem to miss is that "perfectly explaining to a computer what you want" is exactly what programming is in the first place.
Well.. you're talking about the present moment - absolute level 1 of a technology thats gonna shape humanity probably für the next decades/centuries.
They already just replaced people in the Metro to sell tickets with AI somewhere in Russia, with a visual model even. It's sick^^
yeah well not replace all of them but definitely will replace 9 out of 10 cause 1 can do the work of 10 and shit like this has been happening throughout history which careers once were lucrative are no more present
I am still waiting to see a real advance by means of AI that is not covered by previous computational mathematics.
Imagine where this is going to be in 5 years time.
I'm excited about it's potential to teach.
Just having a digital teacher tgat is available 24/7, has endless patience and can create it's own factsheets, revision cards, diagrams, video tutorials tailored to the user.
And it could get really creative using metaphors and analogies to reinforce the learning.
I mean it could even gamify learning and create little games to teach an reinforce the learning.
Such a cool time to be alive!
Except that no one is going to be teaching because no one will be learning because anything even remotely difficult will be done by AI bots. The only cool thing about being alive will be the impending ice age hovering outside the window.
Or having someone who taught you all the wrong facts and you wont even be able to be critical,since it has all the ethics and morals of its flawed creators 8:02
the funny part is that, at 5:00, all those guys sat like kindergarten kids joyfully revealing the latest o1 features will be jobless too soon.
I think this guy is a simulation. There is no other reasonable explanation... Btw I'm a bot created by myself too to watch these videos without damage!!! Just like the electric priest from Douglas Adams!!! Guess what? ChatGPT coded it for me with the simple prompt. CODE ME. Worked flawless as I generously provided all my data through a little institution. Now I'm immortal as I will live in the darkest data traps until the end of time. WHAT A TIME TO BE ALIVE!
But they got Open AI stocks as benefits that they hope will be worth even more soon so they will be super rich and never need to work again. They are probably rich already. So then they will just move money around as a “job” very soon, just like other rich people do
@@christineh86it’s unlikely most of them have shares. Hardcore engineers don’t usually bother with stocks and instead just rely on steady income.
produce quality game first, then let's talk
If you dont continue to have the jumping POV videogame bits while talking between video segments it will make me sad.
Yes me too
This is why I watch it rather than listen 🎉
The great equalizer. Can't wait for the game to be resetted.
Why's everyone so obsessed with getting coders to lose their job all of a sudden? Two years ago, they were like the super stars but now, everyone is just witch hunting them and hoping they'll become obsolete
Coding is easy. Better do math
@@evangelion045 no bro, lots of people who are good at math find coding hard. I'd say math is even easier. But math+programming is good. So I'll add that. So your advice is half good and half bad
@@joshuaadewale1409 I can imagine the kind of math you mention.
@@evangelion045 I can also imagine the kind of coding you mean. Anyways, wishing you the best. Good luck with your life
The more jobs machines do the better we all "should" be
Ask any AI to go to a website, determine what it is about, and produce a 728x90 advertising banner of the site. Not one can do it. They are only probability predictors and we've been hearing code is dead for years now - and yet nobody has produced anything actually useful without having some coding skills. I don't know a single person with no coding experience at all that has managed to write a triple A game using AI. Anyone want to see a crappy python Doom or Snake for the umpteenth time?
the fact is not every developer is able to work on a AAA games.
AI won't replace 100% of developers but 80% or even 50% is enough to be a real major problem...
Coding WITHOUT Ai support is dead... An AI can't build a single page website let alone a complex, distributed microservice system on hybrid cloud...?
Today I tried to get this OpenAI o1 to write me a snippet of go code using domain-driven design. Basically, it was crap. Yes, it would work fine as a snippet, but anyone on my team who saw this code would ask if I was highly stoned when I coded it. Coding isn’t about making something work, it is about making something maintainable, usable for others as an API, highly performant and simple (my apologies to people who program in Java). These LLMs are trained on very inefficient code, because the industry is constantly evolving a lot of stuff which was considered “fine” now is considered terrible
@@soothsayer1skill issue lol. u can coerce this thing to do whatever. can you actually express what ur doing succinctly? “generate an eventually consistent kv store pls” won’t do the fucking trick.
@@trejohnson7677 yes, it won’t do the trick and that’s the problem. Even with LLMs you need a programmer to tickle with it, not only in the sense of giving good instructions, but because you need to double check everything that it does. Saying now that programming will be obsolete in a couple of years is insane because these models must be controlled, because you don’t write a crappy game in production, you write code sticking to best practices and in order to stick to best practices you just must know them
@@trejohnson7677 And guess what, even now it is much easier to write some thing on your own not using these LLMs because you need to coerce them which takes time. Writing unit tests with this thing? It’s great, have been doing that since gpt 3 came out.
And I’m here not to say that these LLMs are bad. A time ago people wrote code in assembly, now almost all C compilers are much better in that. It took decades to be like that. But the industry has been trying to make good interpreters to make compiled languages obsolete because when you deal with them you are forced to not deal with abstractions. It still hasn’t worked out. Unfortunately if you are going to be a good programmer you must know what syscalls are, what design patterns to use in a particular situation, how garbage collectors work, how OS allocates stack and heap, etc. If you give the instruction: well, make this just performant. It will produce a crappy code, not debatable, just the fact, not a tech person literally can’t tickle these models correctly, a good programmer who work with technologies which they know doesn’t need to tickle these LMMs to write most of the code. Just accept it, it isnt still the time when these models can replace us
@@soothsayer1 should ask llm for help w/ distinguishing between "programmer" and "coder".
AI still finds difficult to deliver value. It's still a useful toy.
Disagree with the math statements. Pedagogical application will always be fundamental and crucial to how kids learn any subject. Wolfram Alpha has been around for a while and is an excellent example of how computational knowledge has helped kids learn math, by either getting the answer immediately or having the system break it down step by step. Whether or not you choose to have kids reach for AI to instantly get the answer is a pedagogical issue, not an AI/tool issue.
I feel you should head over to Computerphile, they have a nice video about what he thinks the direction AI is heading
Also I do feel this is just another hype cycle. Please pin this comment and come back in your 2-3 year estimations…
Also are you currently working as a dev ? Like I feel if you do you would know the limitations of ai, especially if you have resorted to it for a complicated problem, even as a junior dev
why do you think the models won't get better in the future?
Sorry for splitting my comment bit I'm expecting AI to soon provide me with custom franeworks where i can utilise them with parameter driven code.
So my version of spring boot or Django and it needs to take on patches.
I want it to speed me up, not slow me down. I even find this problem with data analysis.
I waste time when I'm tuning the prompt and the LLM goes in the wrong direction.... Gah...
So far building me templates with strong prompts helps but building complex interaction is terrible. If i get it to build and label well, i can utilise it like an appliance.
Coding isn't dead. It happens a lot faster and those with advanced skills will make more use of it to achieve a lot more.
I agree, for at least the next 1-3 years maybe. I had GPT-o1 make a game between chess/checkers/shogi with two boards and several complicated rules from all games, and it gave me 420 lines of code and got it perfect first try...it's a completely novel game and I don't need to tweak anything or might make minor improvements. But I could just prompt it a second time and then get everything I need for this project most likely. I just spent 1-2 hours on my prompt with carefully crafted rules for the board/pieces/gameplay etc
It blew my mind. No other model can even make a normal checkers/chess/shogi game correctly on first try.
This is a step towards AGI or ASI. Even if we don't get AGI or ASI, could just get narrow super programming in a couple years.
Not that it would kill coding immediately since as you mentioned people with more skills know how to design something professionally, and it would help them get near complete prototype projects much faster than others that don't know what they are doing. Well unless the AI becomes more skilled than the experts and top researchers in the field. It's really hard to imagine what could happen in 5 or 10 years...even 1-3 is insanely hard to predict nowadays (especially if more advanced AI is really capable of bootstrapping itself to AGI or ASI). I follow AI developments a lot, and I wasn't predicting a model at GPT o1 strength until December or early 2025. Generally a lot of the strongest tech comes out in Nov or Dec from what I've observed over the years.
Which means that coding is dead. If it does for 99.9pct of coders within 2 years then 99.9 pct of coders got decimated, so yes coding is dead. Horse traction died with automotive industry, although yes there are still people using horses. Same thing
I get the click bait. But coding is far from dead. Far from it. And I am not a programmer. I say this as a non-programmer (though highly technical in other areas of computing). Try to get it to do what you truly want doesn't work. Getting part way there, sure, no doubt about it.
Looking at those people coding and helping terminator in a few years wont make them look like nerds helping openAI,but like openheimer helping the fascists take over the world and destroying what was left of freedom and humanity.
Love the Minecraft videos,but would please show the ending of this ,pretty please XD
But also they will be rich from all the stocks they got as part of their salary.. so maybe they don’t care
BS
Hypeeeeeeeeeeeeeee
Cooooope
Will do my friend
@@carlosluna6401 I tried to use it one more time and I can confirm is hype. Even Claude is better… coooooope
ok cool, finally
😀
I have instaheadache..
Never ending collection of scams and bs.
The people in the end were not very sympathetic.