Hacker News

points by bilalq 2 days ago | 0 comments

Your comment was likely focused on dysfunctional parts of the culture there (and that seems to be how most replies to you interpreted it as well). There's truth to that, sure. But I think respect for data sensitivity and security is also a big part of what makes the culture ill-suited for frontier AI lab work.

When I was at AWS, we had data access locked down very tightly. I was trying to train a model to predict how long various dynamic workloads running on behalf of customers would take. Just getting access to anonymized production data from a DB that my team owned took well over a month and had me compromise heavily on what attributes I could access. I also spent a decent amount of time in meetings with lawyers and compliance folks to explain and justify all our data classifications for PII, retention, access controls, etc. We would review our data policies quite frequently. We needed break-glass procedures to access prod data in our own systems, and it was uncommon to need to resort to that. Even when you broke glass, certain classes of sensitive data would still not be visible to you. On the plus side, We did get quite good at operating and maintaining systems without needing to see sensitive info.

If you contrast that with Meta's disregard for privacy or the way frontier labs disregard the law, it's easy to see why Amazon would struggle in this space.

[–] JohnMakin 2 days ago

I read your comment as “amazon acts too responsibly to be a frontier lab,” and that worries me. That they even bothered to do all that stuff is good. That others don’t is bad.

[–] bilalq 2 days ago

I did indeed mean it that way, but I agree with you. It sucks, but it's just the unfortunate reality of the world. Many of the most disruptive and successful startups made it there by ignoring legal or moral limitations.

[–] rkagerer 2 days ago

Frankly reading your comments was a relief. I'd rather they take my privacy (and security) seriously. I don't need or want any AI features or technology from Amazon, thanks very much.

[–] 22c 2 days ago

Amazon has a history of fostering off-shoot companies that are given more freedom to innovate[1][2]. Keeping AI development in-house under Bedrock was probably their mistake. There's a big difference between running AI and developing new AI models.

[1] https://en.wikipedia.org/wiki/Annapurna_Labs

[2] https://en.wikipedia.org/wiki/Amazon_Lab126

[–] bilalq 1 day ago

I don't know where the Strands team sits in the org structure, but there's been a lot of stuff put out by them lately too.

[–] torginus 2 days ago

Their core selling point is customer trust that their product works as advertised.

From my interactions with them, is that they really care about customer success, and everything including AI is a means to an end to them.

I think building a cool product with undefined reliability and capabilities, then selling fantasies about what it will possibly be able to do down the line isn't really what they do - or what their customers are paying for.

[–] JohnMakin 2 days ago

I agree, from my personal experience. AWS support at most tiers I’ve used is excellent. Granted, the larger account you have, the friendlier the service, but even when using it for smaller personal projects, I’ve always been pleased with the level of support and devotion to customer experience (except the console experience, that is not fun).

Don’t know to comment about the other areas of their business but AWS has always been excellent for me.

[–] andai 2 days ago

Incentives can produce locally rational and globally harmful behavior. (See also: "Moloch", although I'm less of a fan of the concept lately since it focuses exclusively on the harmful aspects of emergent systems, deleting the neutral and positive ones in the process.)

A relevant metaphor is the nuclear arms race. Nobody wants a world full of nukes pointed at each other, but if the other guy is gonna do it...

The argument in this case is, we can't slow down, because China certainly won't, and do you want those guys to build the machine God first? So it's "rational" to maximize p(accidental_skynet).

[–] Arbortheus 2 days ago

If the people that do stuff well fail, what does that leave us with?

[–] piva00 2 days ago

With the need to enact political change so incentives align for the benefit of society at large.

[–] Refreeze5224 2 days ago

This indeed. Economic democracy is as important as political democracy.

[–] shimman 1 day ago

What do you think of the idea of workplace democracy?

[–] apercu 22 hours ago

Or even a little skin in the game for staff that are interested in it.

[–] realo 2 days ago

Careful there... people who cannot even describe or define what communism is will tag you as a communist.

[–] lnkdinsuxs 2 days ago

what is communism and are there verifiable reports of it accomplishing what it promises?

[–] Refreeze5224 2 days ago

Where is there verifiable proof that capitalism is not rule by the rich and powerful?

[–] nsagent 2 days ago

I interned there for a summer working on evaluating text generation and despite the fact that I used no consumer data whatsoever, but rather a well-known public research dataset, I also couldn't egress from the machines I ran training on. Not even reports on the training (admittedly, this was 2021 when they had just introduced this new system). So it was even more of a damper on research progress than you might imagine.

[–] lnkdinsuxs 2 days ago

> That they even bothered to do all that stuff is good. That others don’t is bad

good and bad isn't what Meta and AWS is running with. If AWS trains willy nilly on user data like Meta does, AWS is toast. AWS is already taking a huge risk by now using their own take of AWS FDEs competing against their established ecosystem of consulting companies that used to recommend all 583 of AWS products to host a static webpage. Turning of more customers by training on their data is too much of a risk. That's the "good" - a risk analysis.

Meta doesn't care if you stop using Facebook, Whatsapp etc. Frankly, if you are embedded in the Facebook, Whatsapp ecosystem, you don't have a way out as a normie. Even if you are technically capable, good luck publishing your custom Mattermost build for iOS.

Meta has a significant amount of investment in hardware, originally due to Metaverse. The way to monetize that investment is inference and training. That's a do or die decision for Meta. Meta isn't releasing open-weight models because they are "good".

Meta is releasing open-weight models for the same reason Alibaba, NVIDIA or IBM does. Or Google releases Gemma.

There's no "good" or "bad" - the moment Alibaba, NVIDIA, IBM or Google's objective functions change, "Do no evil" will quickly morph into "Do as much evil as necessary without the costs of getting caught exceeding E(outcome)"

[–] SoftTalker 2 days ago

> Meta has a significant amount of investment in hardware, originally due to Metaverse.

Metaverse was 4-5 years ago already wasn't it? That hardware is verging on obsolete.

[–] jnaina 2 days ago

this. the culture that makes AWS the gold standard in cloud security (yes, it is still the best in the business despite the one-off embarrassing CapitalOne incident) with a strong emphasis on locking down data, minimizing PII, and saying “no” by default, is the very opposite of what building frontier LLMs is all about, which is really around hoovering up vast amounts of other peoples data and moving fast before the rules catch up.

[–] everfrustrated 2 days ago

There's a reason companies tend to prefer to acquire over in-house. This is part of the calculus.

[–] sheepscreek 2 days ago

That explains a lot. They are anti “move fast and break things” which is great news for enterprises or anyone seeking stability but terrible for innovation.

I don’t condone the use of pirated content for training. But anything that was bought is above the board as far as I’m concerned. Can let the lawyers fight out the minutia and inevitable licensing specifics after the fact (book publishers are hawks).

Fun fact, did you know that every library in the world has an independent contract with every major book publisher?

In cases I am familiar with, they added a condition that a physical book can only be lent X times. After that, the library must purchase a new copy (even if the original copy is in pristine condition). Yeah, hawks.

[–] djtango 2 days ago

Seems like what you would get if you were customer-centric and building a cloud platform for enterprise and govt to trust which was all set up pre-AI

[–] marginalx 2 days ago

What you are citing appears to me as a business use case of classical machine learning models not necessarily LLMs, and demand forecasting. Frontier AI is clearly not dependent on such data, and has many outside data sources and I'm even sure sensitive AWS data would be of good use as training data, not to mention the legal/contractual obligations AWS has in protecting that data, what am I missing?

[–] bilalq 1 day ago

That was just an example anecdote to highlight the culture around respecting privacy and security. Obviously AWS would never allow for training on customer data, but even for metadata or data of our own internal systems, things are super strict.

Public datasets that are liberal with licensed access is one thing, but the best frontier labs clearly went well beyond that.