Views : 46,336
Genre: Science & Technology
Date of upload: Apr 29, 2024 ^^
Rating : 4.731 (61/846 LTDR)
RYD date created : 2024-05-15T21:08:18.725192Z
See in json
Top Comments of this video!! :3
This is probably a smaller, less resource-hungry version of gpt4 chat. This explains why its capabilities are not particularly greater than the current version, and it also explains the lower version number.
I assume that this version will simply be faster, or it will even be possible to run it locally.
6 |
The example of the "PULL" door@9:40 is solved incorrectly. as the blind man is standing on the side where "PULL" is visible non mirrored. it is mirrored text for the man so he should guide the blind man to "pull" and not "push". Am i missing somehting here??
3 |
I would have gotten the Tommy apple question wrong. That is a riddle more than a math problem. I think what is interesting is that the LLMs get the problem wrong lol! Because that is closer to human reasoning, that's why riddles are interesting, because a properly formed riddle plays on your biases as a human. Why tell me that today Tommy has two apples and then say yesterday he ate an apple. That makes it seem like a subtraction question when its not. Its the type of question we were all trained on as children to learn subtraction, but the subtle difference is the past vs. the future. Very deceptive. Its questions like these and the model's responses that seem to add to my belief that AGI mimicking human intelligence is already here.
1 |
After reading Sam Attam's tweet stating, "i do have a soft spot for gpt2," alongside his previous comment, "GPT-2 was very bad. GPT-3 was pretty bad. GPT-4 is bad. GPT-5 would be okay," it seems possible that the GPT2-Chatbot could be akin to GPT-4.5 or GPT-5.
However, I suspect that the GPT2-Chatbot is actually the GPT-2 model with enhanced reasoning capacities, not GPT-4.5 or GPT-5. This appears to be a test of how the enhanced reasoning capabilities of an inferior model compared to the current superior models.
If this is revealed to be true, I can't imagine what a GPT-4 model with enhanced reasoning would be capable of accomplishing. š¤āØ
1 |
@SkateboardDad
3 weeks ago
It would be so sick if one of these videos actually was what the thumbnail looked like.
62 |