SimpleBench coming in clutch as usual. I absolutely love that you made your own benchmark for testing.

I find it fascinating that GPT 5 outperformed 5.1 and 5.1 outperformed 5.2. Why     See More ny theories? Seems like a potential topic for an entire video? Perhaps I'm missing something obvious in regards to what types of questions are on SimpleBench.    See Less

The problem is the assumption that counting the sheep achieves the goal rather than simply clearing another arbitrary hurdle as you realize the scope of the land outside the paddock.     See Less

17:15 yes “eventually it would count all the sheep” but it still wouldn’t necessarily be able to relate each sheep     See More er.

AI has a strong atomistic intelligence—it’s a savant—yet is still an “idiot savant,” it has a weak holistic intelligence… and I’m not sure LLM’s will ever get a good holistic intelligence as I see little improvement on that level. A different type of breakthrough might be needed to get a positronic matrix. Until then we are stuck with Lore.    See Less

Where is the video on Opus 4.5, the best model of the year and best coding model ever? This is becoming heavily biased towards Google and OpenAI.     See Less

Did your march for Palestine strike out? Does British law exists or is it Sharia now?     See Less

Great work with Simple Bench     See Less

For me, brute forcing the problem by more computing power that a lot of people wouldn't be able to access (cheaply at least) is not impressive.     See Less

I have yet to find a better workflow and experience than I had with 4o/4.1 combo... these benchmarks don't mean shite.     See Less

for programming, i've tried to daily a range of models but always come back to claude. all these recent benchmark results feel extremely disconnected with my actual day to day use of the     See More it used to just be claude (in a good way, overperforming).    See Less

It’s score on benchmark makes me question the relevance arc agi.     See Less