Is it just me, or does GPT-4o Omni suck? I’ve been using it with my Voila browser extension, and I am shocked at how stupid GPT has become since the transition to GPT-4o. It takes things too literally, gets things wrong, and won’t do what I tell it to do.
For example, when I asked GPT-4o to rewrite a product description in the PAS marketing format, it started with the product info and then appended a made-up math formula using the variables P, A, and S instead. Then, when I explained that PAS was a marketing formula, it wrote a description of the formula instead of rewriting the product description using the PAS format. PAS, by the way, stands for Problem-Agitation-Solution and it’s a common copywriting formula.
When I asked GPT-4o to rewrite a sentence fragment, it kept the fragment in the sentence, and even when I asked it to fix the fragment, it still wouldn’t fix it. I finally had to switch back to GPT-3.5 to get it fixed. (At this point I was going to do it myself but I was curious to see if 3.5 might work better and it did.)
I really hate GPT-4o, and, unfortunately, Voila has removed GPT-4 from its options, probably because GPT-4o is cheaper. I've been using Mixtral more and more on my Merlin extension, and that actually works pretty well despite not being super advanced supposedly. Plus, with a 1x credit system for Merlin, I probably won’t use up my credits quickly, even if I asked questions all day. Unfortunately, Voila currently only offers GPT 3.5 and 4o - though I just emailed them to request GPT-4 back.
I just wonder if there is a bit of diminishing returns, where the more they tinker with the LLMs, the worse they get. I do have a friend with a friend who claims to be “in the know,” who is saying that behind the scenes they have a really advanced “GPT-14,” but we’ll see.
So far, I’m not impressed with Omni. What a disappointment.
OK, full disclosure, I have not and will never use any of this crap, but to each their own.
I think many who are in this arena - consumers - myopically see these AI creations as being intended for them and to make them smarter, faster, or give them those abilities, by proxy.
I submit two premises:
1) These platforms are not for you - you are just the training dummies for them.
2) They are meant for the next two generations, that've not been born yet and are in daycare right now - respectively.
In the coming decades, these next two generations and those after, will conform to what they are fed daily, by their AI friends/buddy/tutor. The goal is not excellence or aptitude - but mediocrity, indoctrination and control.
This can be seen in the analog examples of affirmative action and CRT. Those programs were not meant to turn out better, highly educated minority students -but ones that had ingested the regime propaganda and hence been activated to take their place in the agenda wars.
AI and chat GPLMNOP is no different. Yes, you may benefit from the apps and programs now, but the real audience has yet to use them. You are the ones helping to tweak the "just dumb and acceptable enough" algorithm, that will teach, inform and indoctrinate those future generations. It will be their norm and there will be little to no push back, because they won't know anything different.
I'm using ChatGPT 4o that I have a subscription with OpenAI.com. So far, I'm finding it working quite well and haven't run into instances where it didn't provide a good response. I wonder if accessing it through an OpenAi.com subscription makes a difference as opposed to accessing it through an aggregation service?