A brief look at the island, a technologically supreme grouping in the industry/military comprised mainly of Facebook, Microsoft's OpenAI and the usual suspects, Nvidia and Apple Corporation, closely observed by X, Google parent. (Amazon is in there but they're getting group hugs from Raytheon at the moment with their incredible microchip production timetable that exceeds even Moore's law mythological innovation scales.)
#1. Llama provided a foundation for the open source community to chase the stars with new language models. Facebook released it intentionally with its fangs in it to cripple any and all competitors to ChatGPT, the christened monopoly by the chiefs of the world.
PROOF: Anyone that has tried even today's latest Llama based merges will find inconsistent results from even Mixture of Expert models, large models (130B parameters) and lean models with supposedly cleaner data. The output reeks of tampering and noise levels higher than there should be at this point. Hence, llama introduces noise or has at least a few strands of instruction to deliberately sabotage open source models from reaching ABOVE GPT4 - to multimodal, high precision utility.
- Agent work is impossible without using GPT4 generally speaking, even Mixtral models spit out trash thereby breaking agent chain mechanisms for anything feasible like a regular blog published by your own team of agents
- Functional deep coding is impossible due to "misunderstanding" of user intentions and general "clumsiness" with math, logic resulting from llama foundation.
One shot was proven to not only be possible but a reality with the initial releases of the non-nerfed GPT3.5 which was freely available more than a year ago, and has since never seen the light of day. This was the pre-conference version that clearly illustrated feasibility and utility of such a tool in the hands of the public.
Not only did that model display feats that current open source models cannot accomplish, it was even spitting out knowledge about the parent models that trained it when people learned they could trick it into divulging its training data.
#2 OpenAI meanwhile provides a restricted access line to last years slightly upgraded GPT model which is highly woke and super nerfed from the demonstration public version which is no longer available.
The wait list for GPT4 clearly showed the island; no one is allowed onto the landing strip to inspect the tech. No one who isn't on their list of approved denizens. This is tech supremacy at its best.
#3 Nvidia cooperation arises to further ensure the island cannot be overcome.
Despite H100's becoming more available, a whitelist still exists. The drivers themselves powering open source are based on closed code programs. You cannot inspect Nvidia and you cannot get your own hardware company. Therefore, if a nerf occurs at the hardware level, you will be frustrated that your AI LLM merging group just spent 20K on useless nerfed tools.
"Another LLM that is capable of NEAR GPT4 level results!" Ha!
#4 Apple and Amazon appear to be the big boys dripping high tech old world diluted concepts at scale. For enterprise partners of freemasonry, they allow monitored use of their highest tech diluted farming machines. These are bad boys but certainly, if we know anything about this society's makeup, it is highly rudimentary and lacks all class and sophistication.
You can't come near the levels of power with a restrictive physics and electronics paradigm that must be sustained with all the silly engineers without security clearances...
Albeit, this is an intriguing exercise and vain in light of all the things going on on earth.
- Mass layoffs in all tech companies
- Emptied high rises (30%+) at this point globally
- Mass hyperinflation
- Supply lines disrupted globally due to drone warfare threats
- Insane governors in all nations, for example, ordering mentally disabled people to pilot planes, giving over controls of hiring and firing to ChatGPT3.5 and, let's not forget, judgements of laws in the UK to ChatGPT!
- More wars than humanity can realistically handle whatwith all the other societal decay issues at large from the last decade