Artificial Intelligence thread

E

Eventine

Junior Member
Registered Member
https://twitter.com/i/web/status/1883622600028668367

Google is gradually becoming the strongest AI player in the West.

When the race first started, most people expected Google to win it, only to be surprised when they seemed to rush out failure after failure. With Gemini 2.0, however, it seems that Google finally found its strategy and are going from strength to strength.

Granted, Google's Gemini Flash is a smaller model than both O1 and Deep Seek, but its metrics are consistently near the top of the leader board and the free 1500 requests per day makes it, by far, the most generous of available thinking models - even above Deep Seek R1, which is limited to 50 requests per day.

Google is probably burning money - although indications are their model is an order of magnitude smaller than competitors' models - but they are easily the most competent Western company right now in recognizing that "there is no moat" and that the winner will be whoever gets the most adoption, NOT whoever has the "best model." Besides which, Google is still doing top class research - being the first to offer a context window of a million, and with the recently published Titans paper, perhaps even an "infinite" context window.

If I'm Sam right now, I'd definitely be sweating bullets. The next Open AI model better blow the world away, because AI is going to get cheap.
 
Last edited:
M

magmunta

New Member
Registered Member
Here is my take; of course, I may very well be wrong. Deepseek released the paper, weights and models to be downloaded. In the paper, they say RL has created R1 and explain how in the paper. They have not published the data and codes for implementing training using the data. So, the paper could be deceptive to make American competitors waste their recourses on the futile method that the paper says. In other words, deepseek could have been developed by a different methods than the one outlined in the paper. Perhaps, those who are behind deepseek have mathematically found that this method that the paper says is futile while hiding the real method in order to make American competitors divert their research to futile methods. This sounds like what cia did to soviets, deceiving them with pion or muen - don't remember exactly which one-bombs . Soviets actually swallowed that. Who don't know that until others successfully develop R1 level LLM with the technique that the paper asserts. Being open to criticism...
 
You must log in or register to reply here.
Top