PaLM API and MakerSuite | Hacker News Reader

PaLM API and MakerSuite

developers.googleblog.com

95 pointsnwoli38 comments

> Select developers can access the PaLM API and MakerSuite in Private Preview today, and stay tuned for our waitlist soon. [1]

So we're currently waiting for the waitlist?

Edit: ah, from this hn thread [2], seems the waitlist has been published:

https://cloud.google.com/ai/earlyaccess/join?pli=1&hl=en

[1] https://blog.google/technology/ai/ai-developers-google-cloud...

[2] https://news.ycombinator.com/item?id=35207693

Google is doing what OpenAI did 3 years ago. They should have just made it open to try.

I know they don’t want to get another bad PR, especially when people start comparing results with other generative AI providers, but it’s too late to play safe.

https://medium.com/@pravse/the-maze-is-in-the-mouse-980c57cf...

>Does anyone at Google come into work actually thinking about “organizing the world’s information”? They have lost track of who they serve and why. Having worked every day at a startup for eight years, the answer was crystal clear for me — — I serve our users. But very few Googlers come into work thinking they serve a customer or user. They usually serve some process (“I’m responsible for reviewing privacy design”) or some technology (“I keep the CI/CD system working”). They serve their manager or their VP. They serve other employees. They will even serve some general Google technical or religious beliefs (“I am a code readability expert”, “I maintain the SWE ladder description document”). This is a closed world where almost everyone is working only for other Googlers, and the feedback loop is based on what your colleagues and managers think of your work. Working extra hard or extra smart doesn’t create any fundamental new value in such a world. In fact, in a bizarre way, it is the opposite.

Sounds like something that would struggle and crumble pretty easily if a serious contender came into place

That’s pretty much every big company or public organization.

The conquest is done and the moat is in place, they live and work for themselves.

The few cases when their moat starts to fail, they just take it easy first then they step on the gas pedal to fix it and realize nothing happens.. then they panic and star to learn their organization is a giant mount of mediocrity incapable of anything new.

Then they just increase prices, buy some competitors and everything is fine again in the kingdom.

Compared with what OpenAI and Microsoft have done over the past few days, it's hard to get excited about vague press releases.

I don't see how Google can get any kind of market share if all they offer is an API behind a waiting list.

OpenAI understands this. They provide a simple chat interface to interact with their language model that even my grandparents can use.

Midjourney understands this too. The Discord interface is a bit awkward to use, but the fact that any Discord user can instantly become a Midjourney user was a brilliant idea.

Google, on the other hand, is only targeting a tiny subset of all people, namely those who can make use of an API and want to sign up for its cloud offerings.

Is there any (ex?) Googler here who can explain why Google keeps squandering its immense technological advantage over its competitors again and again?

I won't give you the welcome to google spiel. There are a couple of reasons for this imo:

* Most things are built off the internal codebase/repo (only android and a couple of others afair has it's own repo/independent setup). So when exposing something to the public not only does it have to go through layers of legal/compliance processes, it also has to go through "uniformity" processes.

* Engineers don't just build mvps, they try to build the perfect thing where every thing integrates with everything (or there needs to be a road map for this).

* Everything has to be an API first, uis are thrown on top of things.

While these aren't bad engineering philosophies these add up and need to intrinsically build and maintain umpteen layers of decorator/bridge patterns everywhere.

Now combine all this where you don't have the luxury of being the first person in the market but are reacting to a sudden competitive pressure. Lots of processes to undo or get around!

You're comparing a startup to a large corporation. My assumption here is that Google is terrified of the legal issues of raw interactions with an LLM and it saying something wrong or hateful. In terms of search it has a lot to lose if they mess up and start saying wrong things. Bing on the other hand is not the core of Microsoft and can be more tolerant of risk.

That being said I've seen plenty of instances of Bing supposedly being hateful, wrong or gaslighting folks. IE: https://www.theverge.com/2023/2/15/23599072/microsoft-ai-bin...

I think LLMs will be commoditized before these companies are able to make money out of selling APIs. This is not like cloud infrastructure, every decent datacenter can copy their services and steal their users. It's more like web hosting

I hope you are right. Getting locked in to one vendor due to lack of choice is risky.

This article doesn't mention pricing, but I'd be interested to see how this measures up to OpenAI's costs.

I'm gonna make a wild, flying leap prediction here.

There's a thousand pound gorilla that hasn't made a peep about LLMs. That same gorilla has digitized the largest corpus of text on earth, and controls the first or second largest private sector server infrastructure on the planet. They also have talking devices and assorted IoT stuff everywhere. They have the monitoring and live feedback infrastructure (read: Eavesdropping) to improve performance on a much faster timeline.

My prediction is that Amazon all but killed Alexa to focus on building a usable home LLM-type assistant for the masses, and is holding its fire waiting to see how the first generation of essentially toy LLM demos is digested by the press and the public during this Google/Microsoft war.

I don't think Amazon is that good at keeping secrets though. Every big project they start leaks immediately.

EDIT: I wonder how the total size of Amazon's ebook library compares to the pile. Back of the envelope calculations suggest it should be the same order of magnitude.

  We've taken care of everything
  The words you hear, the songs you sing
  The pictures that give pleasure to your eyes

  It's one for all and all for one
  We work together, common sons
  Never need to wonder how or why

  We are the priests of the temples of Syrinx
  Our great computers fill the hallowed halls
  We are the priests of the temples of Syrinx
  All the gifts of life are held within our walls

Ah, fellow greybeards, recognize the AOR master? https://www.youtube.com/watch?v=RtdKhwhAcd4

At least in late 2021 - Google by itself was substantially bigger than Amazon + AWS in terms of data centers.

In terms of private text, FB is orders of magnitude ahead of Amazon.

Amazon is not a 1000 pound gorilla in this space. It's just another player in a game for 5.

It was never clear to me how many books went into training GPT et al. One area where Amazon might have an advantage is if they are able to train on their library of eBooks.

Amazon has been unable to hire LLM/transformers talent for a good 5 years now.

They might have terrific data, but Amazon has always been a little behind on NLP research.

Curious... Why would they be a bad place for that talent to go? Just their corporate culture? Facebook managed to build a LLaMa, and they're notorious...

Amazon did release a 20B transformer called Alexa last year: https://www.amazon.science/blog/20b-parameter-alexa-model-se...

The description sounds fantastic, especially the extra tooling around fine-tuning and prompt creation… but where is it? I can’t find a demo, signup or download link.

That’s outside of scope. Google has been publishing “papers” and writing press releases about their AI tech and how good and smart they all are for the last decade.

That’s the business they are in.

They might get lucky branding it PaLM but TCL own the trademark on the word Palm (Since the original Palm went out of business, and HP sold it off) in the software space

Wonder if it will end up like the IOS/iOS thing where Apple ended up having to pay Cisco

Which leads us to the astonishing failure of OpenAI to give GPT/ChatGPT a user friendly name.

What is going to be the household name for talking to an AI/LLM that replaces Googling?

It’s user friendly enough to be a household name. Google’s supposed software can’t even be used yet.

I guess for computer scientists. My dad calls is ChatGDP half the time.

No pricing and no waitlist ...

https://cloud.google.com/ai/earlyaccess/join

Thanks!

Side note, I love that I'm on android phone using Google's browser to access a page by Google about Google tech products and it doesn't render well enough to read the options I have to pick from.

I assume this is only usable by online applications? If your application doesn’t have an internet connection, you can’t use PaLM?

Where's the Link to the actual API?

Not available yet.

This baffles me. I signed up for the GPT-4 API on Tuesday and got access to it yesterday. Not to mention being able to interact with the model through ChatGPT almost immediately after they made it available.

This is interesting to see, is there not much demand for GPT-4? It feels like the pricing is 10x GPT3 and not as much of an advantage to use and the pattern of giving access to almost everyone who signed up (not sure if there was even a need for a waitlist to be there) tells me things are plateauing here?

I think they probably just prepared for the load, given their experience with the original ChatGPT release.

The irony that OpenAI used a google form to launch their waitlist.

Occam's Razor says that's just an ordinary contemporary tech user/worker, who has no concept that they might not want to do it that way.

But if it happens to be OpenAI strutting in front of Google, like, "We can even hand you all our hot leads, and still beat you," that'd be something.

Reply on news.ycombinator.com