It doesn’t beat thudm nor Deepseek, Nemotron also torches it     See Less

Is this an ai voice and some kind of an agent? please mr Aigrid confirm that you are human. I am not so sure people say you are not     See Less

16:50 you CAN complain about free open source LLMs that aren't good enough to use in a professional way. Otherwise,     See More just a toy and won't be used.    See Less

I don’t understand what’s in it for these companies to spend all of that money to buy the compute necessary to train a model, pay for the power necessary to train a model, release it all     See More then also allow people to run their questions and cover the cost of all that inference compute.

How do they monetize this? Or do they spend 100s of millions to be kind to the community?    See Less

Actually you can get drunk if you take a sip every time you hear the word “actually “ in this video.     See Less

DeepSeek has an order of magnitude more knowledge and much broader abilities. Qwen3 was trained to maximize test scores and performance on a handful of select tasks, but its general knowledg     See More lower than far smaller models like Gemma 2 9b. In other words, it's grossly overfit and a piss-poor general purpose AI model. But what it does, it does well (e.g. math, story telling, and thinking).    See Less

Open source is two words.     See Less

Be careful what you input into Alibaba’s AI. Your data belongs to them and you cannot opt out from data sharing.     See Less

Do you get paid for each time you say "actually" 😅 I think you have good content. But I struggle to listen all the way through when you overuse words, excessively     See Less

4:28 theres been recent findings that base models without CoT are outperforming the reasoning version in a long compute al     See More theres going to be interesting trade-offs to be spotted soon i think!    See Less