You are making AI usable in everyday life. Natural means talking hearing touching and seeing immediately. typing and reading is in reality too slow. Awesome!!
Amazing. I respect Nvidia and what they have done behind their wizard of OZ GPU veil / "Moat of CUDA," but this is what the future of AI will be built on. Definitely know what we will use some of our home refi on for investments once this company has stock options.
Groq builds specialized hardware that greatly speeds up inference, I.e., token prediction carried out by a fully trained LLM. BitNet 1.58 is a special representational technique used during pre-training (so it is not a compression technique like quantization that is applied to the fully trained model). With BitNet 1.58, each parameter in the LLM takes on only the ternary values -1, 0, and +1 rather than a 16-bit floating point value. It will be interesting to see what happens when LLMs with ternary parameter values are executed by specialized hardware such as a variant of the current groq LPU, which would know how to deal with the special encoding of the parameter values.
The company makes a great product but the guy just strikes me as the annoying boss most people dislikes. I hope he is only like this in public appearances and not at work.
![ARGENTINA vs. BOLIVIA [6-0] | RESUMEN | ELIMINATORIAS SUDAMERICANAS | FECHA 10](http://i.ytimg.com/vi/C3wPtKNbAOA/mqdefault.jpg)
You are making AI usable in everyday life. Natural means talking hearing touching and seeing immediately. typing and reading is in reality too slow. Awesome!!
So its basically Jarvis
Amazing. I respect Nvidia and what they have done behind their wizard of OZ GPU veil / "Moat of CUDA," but this is what the future of AI will be built on. Definitely know what we will use some of our home refi on for investments once this company has stock options.
wow!
Why is Qatar and UAE hosting so many of the current high level meetings?
putting so much money
1.58 bit? No?
Groq builds specialized hardware that greatly speeds up inference, I.e., token prediction carried out by a fully trained LLM. BitNet 1.58 is a special representational technique used during pre-training (so it is not a compression technique like quantization that is applied to the fully trained model). With BitNet 1.58, each parameter in the LLM takes on only the ternary values -1, 0, and +1 rather than a 16-bit floating point value. It will be interesting to see what happens when LLMs with ternary parameter values are executed by specialized hardware such as a variant of the current groq LPU, which would know how to deal with the special encoding of the parameter values.
Little did the host know, his job will be replaced by AI very soon...
that is a real female human acting a bot for demo purposes
Amazing
The company makes a great product but the guy just strikes me as the annoying boss most people dislikes.
I hope he is only like this in public appearances and not at work.
His comedy has really gone downhill 😂
Well at least two of us understood that joke 😂😂
thats what a deserving soon to be billionair looks like
@@SeanWald0 Unfortunately, 1-bit LLMs might delay the payout. No need for proprietary hardware.