Your channel is seriously the best thing in AI. I try to keep up, but if I’m not checking AI releases 24/7 I end up hopelessly behind and not really knowing how to judge what’s useful or not right now. And other channels are good for talking about the technical aspects or philosophical debates surrounding AI, but not so helpful for “how can I use this TODAY”.
Honestly, I find a lot of stuff o1 can do, gpt4o can do pretty well too if you know how to use it, like using data analysis and good prompting can make a huge difference in the quality of the results.. (also gpt4o is much cheaper and quite fast so it has this advantage) But it's impressive that o1 is able to do these advanced stuff without any tooling and basic prompting, which makes you think how even better it could be with tooling like vision, data analysis, etc. Like o1 being able to make entire applications with basic prompts, without any tooling, and give quality answers is so smart, and makes it so using LLMS like these even more useful.
00:03 OpenAI o1 offers versatile applications in various fields 01:50 Editing existing contracts is a strength of OpenAI GPT-3 05:38 Tips for Using OpenAI 07:21 Using OpenAI GPT-3 for social media management 10:56 Enhanced capabilities of OpenAI o1 for Advanced Data analysis and Python integration in Excel and PowerPoint 12:40 OpenAI GPT-3 has various applications and features. 16:07 Gen 3 AI excels in switching video styles 17:47 AI technology creating hyper-realistic product videography 21:04 AI applications for negotiating hotel rates and mood-based music search. 22:58 Haen Avatars 3.0 - State-of-the-art AI avatars with dynamic facial expressions and voice tones
I’m curious why GPTs don’t have access to “tools” for solving problems. For example instead of using LLM to do 10-digit math problems, it could just be like “okay I’m going to run this calculator software now, and trust the result from that” and then continue with problem solving after the math has been completed by a classical calculator. You kind of mention this with choosing the ideal model for the task, but clearly there are some tasks that don’t need an AI model at all.
they sort of do with code interpreter but the implementation is far from what it could be. also as you say model switching should be default. i feel like the dev day in 10 days might bring some welcome surprises
So are you saying that Claude Sonnet 3.5 consistently creates any 500-600 lines-of-code Python app, COMPLETELY bug-free in one take/shot every single time? Just like o1-preview?? From what I can tell, if you don’t think o1-preview blows everything out the water with coding, then you’re not using it correctly. Or I just lack declarative & procedural knowledge
I tried to write PineScript indicator with o1, did not work. I tried to use it for a setup of gitlab, runners with nginx as reverse proxy with docker compose, it did not work. This one I could solve with Sonnet in around 1 hour. For the pine script I don't know where the problem was, maybe it is not trained much on pine script But for the docker-compose setup, It used the same ports for nginx and gitlab and could not see this by itself, it could not figure out the problem with the ssl certificates and tried to fix it in wrong places like selinux permissions. To plan and prepare with o1 and execute with Sonnet makes sense.
true! bit of a nightmare scenario if you imagine 100s of thousands of agents just spam calling every business number imaginable just to get a deal they can then resell 🥲
The AI hype machine will feed you for about one more year. The things it’s moderately useful for such as simple coding, summarization etc. won’t pay for further model development. The real use cases still need a lot of work to be productized,
Funny thing is that I thought the same thing in december 22… surely we would plateau and real world use would be limited yet here we are experimenting with 120 iq models and voice models that read your emotions. Not saying that all of this is straight up practical but having a seat on the ship that will lead to the discovery of a new continent (*analogy over*) is the privilege of a lifetime.
Simple coding is massive if you didn’t have the ability to write code competently. Now you do. Now I have a range of utilities to do simple things I need to do as a Salesforce admin on a regular basis. Joining excels, looking up data from one platform to another, data quality checks on a dump of data, etc etc.
@@byrnemeister2008 Without getting into the other issues that you’ll painfully discover over time, are you willing to pay $30-$50 a month for those uses? Sakesforce users are overwhelmingly saying no. genAI has real uses, as I describe. Those capabilities are expensive to provide and aren’t currently being charged anywhere near what they cost to use. The uptake for Salesforce and MSFT copilots is negligible, even when they are undercharging for the service.
All of these AI apps that deal with presentations are the same. They think a presentation is a list of BULLET POINTS with some related IMAGE next to it. These are presentation skills 101. Totally useless for Corporate presentations (e.g., heavy use of of graphs, creative shapes with text embedded). A long way to go before these tools are ready for pirmetime.
Not enough hours in a day keeping up with this fast moving ai river; appreciate you and your team's efforts on our behalf, Igor.
Your channel is seriously the best thing in AI. I try to keep up, but if I’m not checking AI releases 24/7 I end up hopelessly behind and not really knowing how to judge what’s useful or not right now. And other channels are good for talking about the technical aspects or philosophical debates surrounding AI, but not so helpful for “how can I use this TODAY”.
Honestly, I find a lot of stuff o1 can do, gpt4o can do pretty well too if you know how to use it, like using data analysis and good prompting can make a huge difference in the quality of the results.. (also gpt4o is much cheaper and quite fast so it has this advantage)
But it's impressive that o1 is able to do these advanced stuff without any tooling and basic prompting, which makes you think how even better it could be with tooling like vision, data analysis, etc.
Like o1 being able to make entire applications with basic prompts, without any tooling, and give quality answers is so smart, and makes it so using LLMS like these even more useful.
Cau, dobra robota, cely kanal. Sledovat vsetky novinky v LLM svete je pre jednotlivca nemozne ;)
00:03 OpenAI o1 offers versatile applications in various fields
01:50 Editing existing contracts is a strength of OpenAI GPT-3
05:38 Tips for Using OpenAI
07:21 Using OpenAI GPT-3 for social media management
10:56 Enhanced capabilities of OpenAI o1 for Advanced Data analysis and Python integration in Excel and PowerPoint
12:40 OpenAI GPT-3 has various applications and features.
16:07 Gen 3 AI excels in switching video styles
17:47 AI technology creating hyper-realistic product videography
21:04 AI applications for negotiating hotel rates and mood-based music search.
22:58 Haen Avatars 3.0 - State-of-the-art AI avatars with dynamic facial expressions and voice tones
I’m curious why GPTs don’t have access to “tools” for solving problems. For example instead of using LLM to do 10-digit math problems, it could just be like “okay I’m going to run this calculator software now, and trust the result from that” and then continue with problem solving after the math has been completed by a classical calculator.
You kind of mention this with choosing the ideal model for the task, but clearly there are some tasks that don’t need an AI model at all.
they sort of do with code interpreter but the implementation is far from what it could be. also as you say model switching should be default. i feel like the dev day in 10 days might bring some welcome surprises
They used to do that, using Wolfram
I can't wait to give it a go 😊❤
O1 preview wrote a functional complicated scriptin php that solved hours of work in a ecommerce store with 100k products. 🎉
Weiter so!!!! Danke für den Content.
The way o1 handles strategizing is impressive! But is anyone else worried about AI taking over creative jobs? 🤖💼
Not if you have a sense of humour.
Great video, thank you!
Amazing info. Thank you.
Cal you please Share the gpt 6:43 Multi-channel Contest strategist ?
So are you saying that Claude Sonnet 3.5 consistently creates any 500-600 lines-of-code Python app, COMPLETELY bug-free in one take/shot every single time? Just like o1-preview??
From what I can tell, if you don’t think o1-preview blows everything out the water with coding, then you’re not using it correctly. Or I just lack declarative & procedural knowledge
Does anyone know if Copilot wave 2 comes bundled with the default Microsoft 365 subscription?
It's unlikely to be bundled, unless the bundle charges rise by the monthly amount Copilot 365/Pro is charged at separately now.
o1-mini is still only generating hot garbage general output in brainstorming probably because temperature settings are low and we cant change that.
if you are an AI enthuiast...this is the community for you.
i mean, could it be possible that agi would be used as a tool ?
I too, am a gay man
It would honestly take me all day like 10 hours to build a chess game manually ngl. o1 got it in 3 shots, about 4 minutes.
I don't really understand how I can use Out of Focus. Can someone help?
I tried to write PineScript indicator with o1, did not work.
I tried to use it for a setup of gitlab, runners with nginx as reverse proxy with docker compose, it did not work. This one I could solve with Sonnet in around 1 hour.
For the pine script I don't know where the problem was, maybe it is not trained much on pine script
But for the docker-compose setup, It used the same ports for nginx and gitlab and could not see this by itself, it could not figure out the problem with the ssl certificates and tried to fix it in wrong places like selinux permissions.
To plan and prepare with o1 and execute with Sonnet makes sense.
That AI Haggler may have some FTC compliance issues. But I am sure this will be figured out eventually 🙂
true! bit of a nightmare scenario if you imagine 100s of thousands of agents just spam calling every business number imaginable just to get a deal they can then resell 🥲
IQ scores between 90 and 109 indicate a normal IQ range or average intelligence
Nice content! Interested in working together?
The AI hype machine will feed you for about one more year. The things it’s moderately useful for such as simple coding, summarization etc. won’t pay for further model development. The real use cases still need a lot of work to be productized,
Funny thing is that I thought the same thing in december 22… surely we would plateau and real world use would be limited yet here we are experimenting with 120 iq models and voice models that read your emotions. Not saying that all of this is straight up practical but having a seat on the ship that will lead to the discovery of a new continent (*analogy over*) is the privilege of a lifetime.
Simple coding is massive if you didn’t have the ability to write code competently. Now you do. Now I have a range of utilities to do simple things I need to do as a Salesforce admin on a regular basis. Joining excels, looking up data from one platform to another, data quality checks on a dump of data, etc etc.
@@byrnemeister2008 Without getting into the other issues that you’ll painfully discover over time, are you willing to pay $30-$50 a month for those uses? Sakesforce users are overwhelmingly saying no. genAI has real uses, as I describe. Those capabilities are expensive to provide and aren’t currently being charged anywhere near what they cost to use. The uptake for Salesforce and MSFT copilots is negligible, even when they are undercharging for the service.
You do not get unlimited messages with poe. That's not accurate!
It doesn't work at all for me, it requires paid use.
All of these AI apps that deal with presentations are the same. They think a presentation is a list of BULLET POINTS with some related IMAGE next to it. These are presentation skills 101. Totally useless for Corporate presentations (e.g., heavy use of of graphs, creative shapes with text embedded). A long way to go before these tools are ready for pirmetime.
First
second