Another demonstration of chatGPT 4.0 capabilities

[Laird]

Update: This week, I’ve wasted at least 10 working hours on the new 'ChatGPT-01' wonder! I can’t say it enough times – it can’t 'think'. Don’t believe in the hype.

What about DeepSeek? Do you think it's doing something very analogous to thinking in this video? It certainly looks like it to me.

DeepSeek R1 Fully Tested - Insane Performance

What about DeepSeek? Do you think it's doing something very analogous to thinking in this video? It certainly looks like it to me.

DeepSeek R1 Fully Tested - Insane Performance

Fun video. The reinforcement learning simulates human 'lines of thought' quite well, but it is still fundamentally based on the same underlying algorithms that drive other LLMs. It doesn’t 'understand' the words it outputs in the same way that we inherently grasp their meaning.

As I understand it, the buzz around DeepSeek is that the Chinese have demonstrated that a state-of-the-art model can be trained on a limited amount of standard commodity hardware, rather than requiring thousands of state-of-the-art GPU accelerators.

[Laird]

Yes, that's my understanding too. I've seen some skepticism expressed as to just how much they've reduced hardware requirements versus how much is propaganda, but it seems to be generally agreed that whatever they've achieved is impressive and game-changing.

[Laird]

I've just similarly tested the version of ChatGPT at chat.openai.com, which it self-identifies as being based on the GPT-3.5 architecture. I tested it on a deceptively simple problem, like you did in the OP for version 4.0, and it didn't get it right after several attempts, although it wasn't totally out of the ballpark. I'd be interested in trying on version 4.0. Maybe you can feed the problem description in and see how you go? Here's the relevant part of the transcript:

[Snip]

[Note: the correct answer is 51, as I calculated here]

I've tried this on the current (free) version of ChatGPT, and it didn't even raise a sweat, providing the correct answer with correct reasoning within a few seconds.

That's unsurprising given how relatively easy the problem is compared to what it's capable of solving, but it does go a little way towards showing how much this tech has advanced in just three years.

Transcript attached.