GPT-4 Turbo vs GPT-4o in Reasoning TEST #gpt4o

New GPT-4o, means new GPT-4 OMNI, with new video and voice functionalities, but what about its classical reasoning performance?

Tested w/ my personal test suite, maybe you should hold on to your trusted GPT-4-TURBO for causal reasoning and logic deductions.

Not a statistically relevant test or results, since not performed 10000 times on multiple machines on multiple days. Just my personal test impressions.

It helps me to get a feeling for the new GPT4o. When to use it, and when to switch back to TURBO.

#airesearch
#newtech
#gpt4o

[pin]

One thing I'm finding with GPT-4 is it doesnt seem to expand much on topics when I try to dig in deeper. It basically just rewords what it already said.

[Charles-Darwin]

I think 'o' actually stands for orbitofrontal cortex.

To mimic our own structure, It could be a smaller/narrow receptive input network that doesn't really retain or memorize beyond simple and critical pathways, and a much larger network that assesses the weighted inputs - for bottom-up top-down approach. Because of this, I think 4o is a double ended model that are working together/in tandem for distilling input and assessment.

This region of the brain is multimodal, but just as our organic builds, vision is the primary input where the other modalities also largely construct to visual representations (hear a garbage truck outside, visualize what that truck looks like in your head). This region is also extremely low latency by necessity as responses to visual input needs near-automatic responses (driving a car, walking).

All things considered I think this is the analogue of our orbitofrontal cortex and perhaps the applicability extends far farther and wider than theorized prior to implementing the solution. Shy of having the equivalent biological function to survive, I think this is AGI and we've only seen the baby brother. I don't think we'll get the whole enchilada this year or the next, rather what they've been saying, an agentic version of Jr to do biddings to paid subscribers will come this year then next year will be an incremental improvement and prescense in robotics for sure. They'll keep the big one privately running to bolster its abilities and maybe out of precautious reasons. This kind of a breakthrough also aligns with the primary scientists (and alignment conscientious) taking their leave as the management has turned on the primary objective, allocating infrastructure resources to press forward with the model's expansion over creating safety for it.

I think those scientists are spurring a company dedicated to alignment.

[hl]

Perfect. Just what a needed. A reasoning comparison that’s not coding related.

[dasistdiewahrheit]

Reminder: Don't give ClosedAI power!

[IdPreferNot]

Purely anecdotal, but when i run some random work "think in steps" prompts on the new model, that i would have assumed the old model in practice didnt have a problem with in past workflow examples,
Other than the entirely improved speed (at apparent real costs), im not sure what actual prompts its IMPROVING for !?!?

[propeacemindfortress]

I bet both will argue that marketing is a vital component of adoption 😂

[lighteningrod]

Omni is a Omni channel patch :( probably for training gpt5, and I’m paying for their benefit.