Google's 2nd Answer to "BING ChatGPT": Sparrow | after BARD w/ LaMDA | 2nd Gen Conversational AI

[@ailinxu]

Epic feud battle between Google and Microsoft has began ❤️❤️❤️

[@Cloudvenus666]

I love all your videos. You’re a wealth of knowledge. Question, is there any way we can get our hands on the model, even before google releases it for themselves?

[@code4AI]

Thank you for the positive feedback. Regarding the Sparrow model: if I ask, Google might give me the code, but where do I spin up 10000 TPUs to run inference? I do not think, NVIDIA has produced enough H100 for this task .... Smile. It also shows, why OpenAI is depended on Microsoft's compute (cloud) infrastructure.

[@Cloudvenus666]

@@code4AI fair point but what type of TPU are you referring to that you’ll need 10K of? Version 2 or 4? Because version 4 has 100 times the computation power compared to v2. I believe TPU4 architecture provides up to 180 teraflops of floating-point performance, compared to the 2.5 teraflops of TPU2.

[@code4AI]

It was mere figurative, but look at BLOOM for detailed data: The Jean Zay Supercomputer (France) took 177 days to train BLOOM, and if I remember correctly, they used around 400 A100 80GB GPUs. Those monolithic autoregressive LLMs provide great revenues for cloud compute providers, but I do not need a LLM in my professional work that has been trained on french lyrics or british poems. Dedicated professional systems (limited to your profession) is what I am currently building, not a system that does everything for everybody. It is a nice marketing idea to raise awareness and induce growth for Microsoft shares, but why start all cars in a parking lot when I just want to drive with one car ...

[@oliverli9630]

when will it be released? 😍

[@code4AI]

Google will make an announcement next week ...

[@jasoncravens1124]

I'd think a B+ is very acceptable. It's just doing things, in that 8 percent, they don't want you to stumble on. It's not quite controlled enough for their liking.

[@gmsarti]

Your videos are awesome :)

[@code4AI]

Glad you like them!

[@ammini999]

Google Home shoud already use this technology, isn't it?

[@code4AI]

Well, you know, a start-up can afford to fail in high-tech, Google can't!

[@nyyotam4057]

I like Google. I like Google a lot.

[@Haveuseenmyjetpack]

starts at 15:37

[@code4AI]

How unfriendly! The demo starts there, but quite a lot of my viewers ask for information, theory, development and details. You miss especially the reason why ChatGPT was launched. Smile.

[@Haveuseenmyjetpack]

@@code4AI Those people are free to start from 0:00. However, 99% of people just want what the TITLE says "SPARROW DEMO integrates real-time Google Search"

[@code4AI]

You are wrong, in 2 ways. First: The title says "Google's answer to ChatGPT: Sparrow". Second: You do not know what 99% of my viewers want. I do, since 2 years. You just imply, that what you want is what all other viewer have to want. Arrogant.

[@Haveuseenmyjetpack]

@@code4AI Your videos are too long mate. Just keep the meat, discard the filler. peace

[@code4AI]

A personal statement? I appreciate it. Thank you for your comment.

[@summitupman995]

ChatGPT's training didn't end in September 2021

[@code4AI]

Since I provided an official link to latest OpenAI's documentation in my video, would you please provide an official link for the verification of your statement. Thanks. Did OpenAI update its information? I just check OpenAI's documentation again (beta.openai.com/docs/model-index-for-researchers) and quote: "GPT-3.5 series is a series of models that was trained on a blend of text and code from before Q4 2021. The following models are in the GPT-3.5 series: code-davinci-002 is a base model, so good for pure code-completion tasks text-davinci-002 is an InstructGPT model based on code-davinci-002 text-davinci-003 is an improvement on text-davinci-002"