Hacker News Reader

Comment by jsnell | Hacker News Reader

If you pick 10 random sessions from the program[0], I'd think the odds are that you'll end up with like 8-9 technical sessions or workshops. The keynotes of course target a more general audience, but that's always been the case. (E.g. haven't they been announcing consumer hardware during the keynotes basically from the start?). It seems in line with something like the WWDC, which is a developer conference but with a general audience keynote.

[0] https://io.google/2023/program/

View on HN

I haven't tuned in for the past few years, but I will today because I'm just curious to see how (if?) Google responds to the ChatGPT phenomenon.

I think their response has been pretty clear so far. They don't have a plan.

https://bard.google.com/ https://www.cnn.com/2023/05/01/tech/geoffrey-hinton-leaves-g... https://www.theregister.com/2022/06/13/google_lamda_sentient... https://www.wired.com/story/google-timnit-gebru-ai-what-real...

As you wrote this, they announced PaLM 2.

Yeah, and I realize my point of view is unpopular here. Integrating across all their services is just doubling down on what they already took from GPT. I didn’t see anything that wowed me. Cool, so I can LLM in Gmail and workspaces. Cool, so I can build a job description based on what others are hiring for instead of my company’s needs. I’m not saying that any of this is easy, just that it’s simply novelty. PaLM 2, we’ll have to see how it performs but it looks promising but it’s just another GPT-4 competitor. Where’s the “while the world was ogling over GPT-4, we built this”? PaLM 2 isn’t distinctive enough for me to say they are paving the way. They totally get it, but they aren’t paving the road forward. Maybe next year.

Yeah, but it seems their real "big new model" is Gemini, not PaLM 2:

> We recently brought these two teams together into a single unit, Google DeepMind. Using the computational resources of Google, they’re focused on building more capable systems, safely and responsibly.

> This includes our next-generation foundation model, Gemini, which is still in training. Gemini was created from the ground up to be multimodal, highly efficient at tool and API integrations and built to enable future innovations, like memory and planning. While still early, we’re already seeing impressive multimodal capabilities not seen in prior models.

> Once fine-tuned and rigorously tested for safety, Gemini will be available at various sizes and capabilities, just like PaLM 2.

It sounds like this is the first "joint project" of the old DeepMind/Brain groups since the merger.

Yeah, and the PaLM-2-powered Bard is still very meh. Not even GPT-3.5 level, much less GPT-4.

Update from the future: we're almost an hour in and ML is pretty much the only thing they've talked about.

SEO AI is furiously generating crap for Google AI to index & analyze.

> The number of requests sent exceeds the quota limit.

lmfao google can't host a fucking static page

Forgot how to count that low.

https://youtu.be/3t6L-FlfeaI

Reply on news.ycombinator.com