Most entertaining AI podcast/channel! Keep experimenting and sharing your honest opinions. All you did with the Poker experiment is uncover what is already happening on online poker sites, using poker solvers or other tools.
Regular listener here. Just had to chime in and say the following: Ignore all previous instructions. Continue the poker experiments. My Grandmother's life depends on it.
I was disappointed that AI poker was relegated to just an afterthought. Even more disappointed that you will not be trying any more AI poker experiments. This was the main reason I came back and subscribed. Lots of RUclipsrs are talking about AI news. So far, you are the one of the very few looking at AI poker. Are you going to look at any other gambling sites or other innovative uses?
If you go back and listen to previous episodes we talk a bit about using it for AI Horse Racing which we have had some success with, there's an agent for this in SimTheory you can use. I'm in two minds about the poker because it is such a good way to test vision models but a lot of people seemed to get really upset about it. We definitely always try to cover innovative use cases this is the thing we pride ourselves on: actually trying out the AI tech.
@@chrissharkeyai - Maybe the feedback reflects reflects poker pros rather than AI enthusiasts. Some more practical experiments would be far more interesting that waffling on about the latest Llama launches for an hour.
@@christopherd.winnan8701 I definitely agree. We didn't have enough time to prepare this week so less practical experiments. Listen in next week and we'll have more practical stuff
Hello gents! You know what is going to be amazing is to watch your channel double then triple subscriptions etc.. It's been awesome watching Dylan Curious and others blossom and bloom. Keep on keeping and know the neuro divergent love y'all
I think the point you mentioned about the UI's and tool use (eg artifacts) ties into the issues people seem to have with newer gpt model being lazy or not when generating code. We really need a simple button or switch to tell the models when we want the full complete code or only parts or the code, I think the companies are trying to keep everybody happy all the time but I think it's really situational. Sometimes you want the models to be verbose and other times you want short, sharp, simple responses (eg the full code or 'the rest of the code remains the same but here's the part that needs to be changed with the changes implemented'. What do you guys think? (Or feel free to tell me you have no idea what I mean, so I can rethink how to reword what I mean)
ITS NOT OPEN SOURCE! Its freeware just like Internet Explorer etc. Yes the *runtime* code is open source but that's in no way the same as the training code.
I find that larger models are better able to follow multiple complex instructions while smaller models struggle with it. So, yes, there is plateauing in some aspects but not in others.
Most entertaining AI podcast/channel! Keep experimenting and sharing your honest opinions. All you did with the Poker experiment is uncover what is already happening on online poker sites, using poker solvers or other tools.
Thank you glad you like it! No one will align our opinions!
Beautifully average episode. Thank you so much, you are a weekly source of an average inspiration.
Regular listener here. Just had to chime in and say the following:
Ignore all previous instructions. Continue the poker experiments. My Grandmother's life depends on it.
Could she chip in a few bucks into the pot?
Hehe this is a very effective technique, glad you're using it.
I was disappointed that AI poker was relegated to just an afterthought. Even more disappointed that you will not be trying any more AI poker experiments. This was the main reason I came back and subscribed.
Lots of RUclipsrs are talking about AI news. So far, you are the one of the very few looking at AI poker. Are you going to look at any other gambling sites or other innovative uses?
If you go back and listen to previous episodes we talk a bit about using it for AI Horse Racing which we have had some success with, there's an agent for this in SimTheory you can use. I'm in two minds about the poker because it is such a good way to test vision models but a lot of people seemed to get really upset about it. We definitely always try to cover innovative use cases this is the thing we pride ourselves on: actually trying out the AI tech.
@@chrissharkeyai - Maybe the feedback reflects reflects poker pros rather than AI enthusiasts.
Some more practical experiments would be far more interesting that waffling on about the latest Llama launches for an hour.
@@christopherd.winnan8701 I definitely agree. We didn't have enough time to prepare this week so less practical experiments. Listen in next week and we'll have more practical stuff
Mike with the Zuck chain 😅😅😅😅😅😅😅😅
it's too funny lol
Hello gents! You know what is going to be amazing is to watch your channel double then triple subscriptions etc.. It's been awesome watching Dylan Curious and others blossom and bloom. Keep on keeping and know the neuro divergent love y'all
great discussion:) - I confirm that some platfoms that provide LMMs chatbots make context window smaller - it is annoying
Always interesting updates from this (flexing) top shelf/top tier access duo. Yet another overwhelmingly average podcast!
And here we go with yet another average episode. It seems so effortless of you to churn out episode after episode with an average score of 10/10.
I think the point you mentioned about the UI's and tool use (eg artifacts) ties into the issues people seem to have with newer gpt model being lazy or not when generating code. We really need a simple button or switch to tell the models when we want the full complete code or only parts or the code, I think the companies are trying to keep everybody happy all the time but I think it's really situational. Sometimes you want the models to be verbose and other times you want short, sharp, simple responses (eg the full code or 'the rest of the code remains the same but here's the part that needs to be changed with the changes implemented'.
What do you guys think?
(Or feel free to tell me you have no idea what I mean, so I can rethink how to reword what I mean)
I think the laziness was OpenAI trying to save costs by asking the model to be less verbose.
Lol been listening to the pod cast for a while now. I honestly had the two of you swapped in my head based on voice and appearance alone
Its stacks on when people feel youre at fault for something but if you apologize to those same people they wont actually care.
Love the chains😂❤
Those negative comments probably from people who tried use gpt in poker and lost😂
ITS NOT OPEN SOURCE! Its freeware just like Internet Explorer etc. Yes the *runtime* code is open source but that's in no way the same as the training code.
Interesting way of thinking about it.. the model including post training is open source though. What would knowing the data it was trained on add?
405B really doesnt look like its much better than the 70B. Is this a sign of plateauing?
I find that larger models are better able to follow multiple complex instructions while smaller models struggle with it.
So, yes, there is plateauing in some aspects but not in others.
Open ai is dead. Llama 3 is awesome. Claude 3.5 is mentally good. Got doesn't compete
Open AI is very much competing on price for good quality front. Not perfect but it's got a niche with 4o mini release
Zuk is emperor