Hacker Read top | best | new | newcomments | leaders | about | bookmarklet login

PALM2 is from then Google Brain I think, DeepMind didn’t bet big on LLM until much latter.


sort by: page size:

Curious why you think this? PaLM2 looks great, and Google has been productizing cutting edge AI pretty fast for years.

This work is based on a fine-tuned Google Palm Model from 2022. I'm not sure if this is a fair comparison to the latest groundbreaking series of LLMs

I am sure Google is working on some great new models. PalM was a disappointment, but they have been leading the charge in deep learning for decades.

Beautiful website and nice fonts!

but back to the topic: I’m quite shocked that PaLM gets outclassed by much smaller models on a regular basis. I would have thought that Google, despite not having a moat, at least had enough talent and focus to get LLMs right. But what I’m observing is that startups like ClosedAI, Anthrophic, etc. constantly beat big players like Google in their own game.


Palm2 tells me that it is most powerful LLM in the world, but it isn't Gemini yet. LOL! I don't need AIs that are parroting marketing crap.

it uses PaLM 2 - which is an LLM trained from scratch. https://en.wikipedia.org/wiki/PaLM

this is for investors. while some companies actually build useful models, some like inflection are still catching up with palm-2, an old, forgotten, measly model.

They have their own LLMs such as PaLM

Wasn't that PalmOS which got acquired by LG?

When they make it based on PaLM vs LaMDA

All this is pure speculation:

I have absolutely no inside info or actual knowledge to back up this hunch, but part of me thinks Google released PaLM2 just to get something out the door that was somewhat comparable to OpenAI's models.

That way they can say they put something out and that they're "competing", but internally, maybe they're really looking forward to (and working on) their next-gen AI models. Something to compete with a hypothetical future GPT-5, but maybe they're looking to get their version out faster than OpenAI can. I feel like Google is really going to give it their all on the next-gen model they release (not PaLM 2, but whatever they're working on after it).

Curious to hear other people's thoughts on this.


PaLM 2 on Vertex AI which Google just released yesterday has fine tuning the large models as a core part of their offering.

PaLM was basically a larger version of GPT-3. It was not similar to ChatGPT. It was a pure token predictor. It was not fine-tuned for chat.

As you wrote this, they announced PaLM 2.

Of course palm rejection (and many other features) are powered by ML. THis doesn't mean they run on the ANN though.

Yes, PaLM and Bard have existed internally for years but Google just didn't release them. Imagen, which is better than Midjourney or Stable Diffusion, also hasn't been released.

Which one? If you're talking Palm 2, I couldn't say, I spend my days trying to convince everyone at Google to prioritize and support ME first ;)

Given how PALM2 had less params than the first one (540B) I think that makes sense.

I can already tell you that PaLM is not anywhere near as good and PaLM-2 is at least not as good before RLHF.

Not going to keep replying, believe what you want about Google's capabilities

next

Legal | privacy