It's crazy how much I've used o1 the last days for coding tasks. And now just before I went home from work it said "You have 25 answers left, it resets October 1".. Damn how am I even gonna survive that long. I guess I'll have to think more about what I ask it, to get many features into one prompt.
i make a 4o thread where i just dump my ideas, any relevant info and ask for a concise prompt for o1 to get as many answers in 1 reply from o1 as is feasible. It also helps to ask it for a step by step, so it links the info together, if that's what you're after.
Claude for programming works great but it takes more prompting when creating a new web app that hasn’t been created before. There are so many “snake” “flappy bird” and “Tetris” clones online so using that as a test is not helpful.
I never tried o1-mini because I just assumed it would be worse than o1-preview. It was quite stressful to have only 30 requests per week. Thanks for showing that I can simply use o1-mini!
Thank you for this! Especially the o1 demo. I use cursor myself now to create a cool AI app (of course), and I just told o1 to review what I had of infra and code, and then make a battle plan for the continued development. Needless to say using your trick of a good elaborate prompt, the answers were great! Could you share some insight on how you put this together?
Why do people keep referring to the history of ai by saying things like "brings me back to the days of gpt3" All of this has only been accessible for two years. Content creators keep doing this in their videos and its mental.
Def not gpt3 days. o1 is price gated by insane api prices, OpenAI going fully closed and for profit. If people that worked on gpt3 would hear this they would call them out for sure. O1 is crazy good but now it's more and more about the ecosystem and monopoly building around openai than research.
The problem with using API to break out of chatgpt, each individual user needs to insert their personal API and hard $ will start to tick so it is just not feasible and accessible enough to offer it to the end user. We need to wait for microsoft OS to offer llm support before it is actually worth developing the tool without the use of API
15:40 "whoa calm down jamal, don't pull out the 9". what you mean to say is, everyone can write some code that they don't understand and don't know if it's buggy or not, or how it should work.
01 alone can't code any apps. It's solely the person guiding it that's actually building it... a.i is just spitting out small bits of code. I know firsthand a.i sucks with medium+ size codebases. Its only as good as you are at understanding the code base... if that makes sense lol. Even then, it's not really building it. It's cipying code off of github line per line lol.
I am an electrical engineering graduate. I started my career as a full stack web developer, with one year of experience, just tell me should i continue my career in software development?
I am senior dev (not in faang ) and I can tell you that nobody knows for sure will ai be able to do devs job in future. Right now it can't do complex coding tasks with big codebase. As for job market I am pretty sure it would become significantly better in a year or year and a half (not a covid times good but like 30-40 percents better.)
Has anyone else gotten a response from o1 mini saying it’s training cutoff date was September 2021? I asked it a programming question and it’s responded that it’s training cutoff was September 2021 and has no knowledge past that time
That's a bit exagerated. I can still programm better with claude sonnet. I teste o1 with setting up Gitlab with nginx in Docker and it could not do it. It was confused with the ports and nginx setting with tls. I am sure there are some programs and games that it has code of it it's training data, that's why it can do it. In my usage I have not found anything that shocks me. There are too many over hypers out there.
BS, I tried to work with a small 80 file typescript codebase, and o1-mini flops completely, while sonnet3.5 keeps producing better code then o1, no idea why, just facts. 5 days of daily iterations, with cursor composer and o1-mini is not as good as "benchmarks" show, no idea why.
AI is a hair above useless. Cant even sort and group my downloaded spreadsgeet of bank activity based on examining prior transactions I already categorized in a prior spreadsheet. Cant sort photos into logical categories based on content of photo, cant monitor the stock market and give me winning trades 80% of the time..., ..., The number of things AI cannot do is astounding
I think it's a valid comment. I like the interview in whole. But it is slightly distracting. Those moments were he is 'monitoring' as you suggested. Can be edited.
o1-mini isn't avaialble in my account. I'm a paid account user. Is there something I have to do to or is it a hurry up and wait for them to roll it out situation? o1-engineer looks legit!
🔍 Join the New Society and get a personalized AI analysis of your job and career: www.skool.com/new-society
Few people are honest with themselves and intellectually honest. Am I projecting or is this guy lowkey greatness
“You can code any app” that is honestly such an absurd claim. Stop the cap
It's crazy how much I've used o1 the last days for coding tasks. And now just before I went home from work it said "You have 25 answers left, it resets October 1".. Damn how am I even gonna survive that long. I guess I'll have to think more about what I ask it, to get many features into one prompt.
i make a 4o thread where i just dump my ideas, any relevant info and ask for a concise prompt for o1 to get as many answers in 1 reply from o1 as is feasible.
It also helps to ask it for a step by step, so it links the info together, if that's what you're after.
Claude for programming works great but it takes more prompting when creating a new web app that hasn’t been created before. There are so many “snake” “flappy bird” and “Tetris” clones online so using that as a test is not helpful.
Yes I'd like to see it actually aid in the conception and development of a unique game idea
Then why don’t you build actually usable products and make profits from them instead of youtube monetization
Because it takes a lot more to build a usable product than this.
You're far greatest than even ASI 😅😅😅
Hahahaha I love you
Because he sell courses 😂
I never tried o1-mini because I just assumed it would be worse than o1-preview. It was quite stressful to have only 30 requests per week. Thanks for showing that I can simply use o1-mini!
Such a great talk - keep up the live demos
Thank you for this! Especially the o1 demo. I use cursor myself now to create a cool AI app (of course), and I just told o1 to review what I had of infra and code, and then make a battle plan for the continued development. Needless to say using your trick of a good elaborate prompt, the answers were great!
Could you share some insight on how you put this together?
2 ads per 2mins btw
Ad block btw
YT Premium
If you don't know how to block ads in 2024 then you deserve to see them, it takes a couple seconds to figure it out 😂
If anyone needs YT premium for dirt cheap DM me!
Use brave browser 🧠
Great demo: thank you very much!
U must hav a dif version of o1 than me
Yup, o1 didn't work better than sonnet3.5 for me... Typescript codebase
Why do people keep referring to the history of ai by saying things like "brings me back to the days of gpt3" All of this has only been accessible for two years. Content creators keep doing this in their videos and its mental.
Def not gpt3 days. o1 is price gated by insane api prices, OpenAI going fully closed and for profit. If people that worked on gpt3 would hear this they would call them out for sure. O1 is crazy good but now it's more and more about the ecosystem and monopoly building around openai than research.
Love listening to two passionate individuals discussing the latest achievements in artificial intelligence. Great content!
Best vid I've seen! Unbelievable!
I'm so excited to try this tomorrow
I made a custom language model using o1. Currently training to see what happens.
Let me know, so it's capable of that? 😮
@Vbastin-p2r kind of. Gpt is only as smart as you are able to use it. Still had to learn a lot of math's, programming, and theories myself.
But it is SOOOOOOOOO much better than 4o or 4
The problem with using API to break out of chatgpt, each individual user needs to insert their personal API and hard $ will start to tick so it is just not feasible and accessible enough to offer it to the end user. We need to wait for microsoft OS to offer llm support before it is actually worth developing the tool without the use of API
it seems software engineering has become all about developing snake games and pong games.
15:40 "whoa calm down jamal, don't pull out the 9". what you mean to say is, everyone can write some code that they don't understand and don't know if it's buggy or not, or how it should work.
Please, show what you've cut out
Is Cloud Engineer a prompt or a code or an Api or A Model?
Very inspiring!
@Steinberg making VST's should be this simple
o1-engineer doesn't work for mere mortals on tier 1 that don't have access to the o1 API
it does, you just need to use OpenRouter
I am sure, with a bit of tinkering, it would work with OpenRouter...
01 alone can't code any apps. It's solely the person guiding it that's actually building it... a.i is just spitting out small bits of code. I know firsthand a.i sucks with medium+ size codebases. Its only as good as you are at understanding the code base... if that makes sense lol. Even then, it's not really building it. It's cipying code off of github line per line lol.
Thing is can it code an app like twitch or bigo live ?
We love you dadusak
I am an electrical engineering graduate. I started my career as a full stack web developer, with one year of experience, just tell me should i continue my career in software development?
I am senior dev (not in faang ) and I can tell you that nobody knows for sure will ai be able to do devs job in future. Right now it can't do complex coding tasks with big codebase. As for job market I am pretty sure it would become significantly better in a year or year and a half (not a covid times good but like 30-40 percents better.)
Will the programmers be replaced has anyone any doubts?
The programmers will change but not be replaced.
Has anyone else gotten a response from o1 mini saying it’s training cutoff date was September 2021?
I asked it a programming question and it’s responded that it’s training cutoff was September 2021 and has no knowledge past that time
o1 model is trained with data until oktober 2023
Is it better than all those coding models like deep seek?
How do I get the o1 ai app
That's a bit exagerated. I can still programm better with claude sonnet. I teste o1 with setting up Gitlab with nginx in Docker and it could not do it. It was confused with the ports and nginx setting with tls. I am sure there are some programs and games that it has code of it it's training data, that's why it can do it. In my usage I have not found anything that shocks me. There are too many over hypers out there.
is there a workflow step after everything is done to then implement and see/use the end result? (audio app example)
What models wouks you use to develope a web3 wallet with very littlw coding wxperience but solid conceptual understanding?
How long is the training?
18:55 min what website is that?
Legit convo
Tony Montana?
the next thing will be multiple simulations happening at the same time of thinking for weeks not just one thing
Not gonna watch. Just came to comment that no you cannot. It helps but it still creates many errors.
BS, I tried to work with a small 80 file typescript codebase, and o1-mini flops completely, while sonnet3.5 keeps producing better code then o1, no idea why, just facts. 5 days of daily iterations, with cursor composer and o1-mini is not as good as "benchmarks" show, no idea why.
Its still not great, but better.
AI is a hair above useless. Cant even sort and group my downloaded spreadsgeet of bank activity based on examining prior transactions I already categorized in a prior spreadsheet. Cant sort photos into logical categories based on content of photo, cant monitor the stock market and give me winning trades 80% of the time..., ...,
The number of things AI cannot do is astounding
First!
stop looking back and forth man. It's distracting. Focus.
You don’t even know what he’s monitoring while he makes his videos bro calm down
I think it's a valid comment. I like the interview in whole. But it is slightly distracting. Those moments were he is 'monitoring' as you suggested. Can be edited.
BS
o1-mini isn't avaialble in my account. I'm a paid account user. Is there something I have to do to or is it a hurry up and wait for them to roll it out situation? o1-engineer looks legit!