Hacker News Reader

Comment by null_investor | Hacker News Reader

Meta is killing it. Google seems to be lagging behind them in AI research and useful things that is shared within the community.

I'm sure this, LLAMA and the other projects that they have will help drive up new creations, companies and progress.

I'm also sure that this kind of openly sharing code and research will drive up business value for them. It may be hard to say right now what it is, but I'm sure it will.

That's the difference of a founder-led company vs. market led. Google is mostly concerned on short-term goals so they don't report a single bad quarter or has too high CAPEX on a project without profit on its sight (like VR).

Once Meta finds the killer app for VR, all the other companies will be so many years behind that they will need to buy software from Meta or not take any market share in this new space. Similar to what happened about AI chips and Nvidia. Nobody was investing enough on it.

View on HN

Google still leads in AI research. They are doing the opposite of short term. The reason why it does not seem like it may be because most of their work is either basic research or related to chemistry, physics or things that are not public like Facebook.

They are behind in productization of the research. So far they seem to do the minimal effort from the trained model to product.

Also don't forget Waymo.

waymo is pretty far behind in terms of self driving tech

I was under the impression Waymo was the leader. Who are they behind and how?

i guess it depends on how you measure it.

1. how many miles does waymo drive compared to competition?

2. how many people use waymo versus competition?

3. how many miles of road does waymo work on?

the answer to all three is thousands of times less than Tesla's FSD, which granted is not free of disengagements and still supervised. But then you have to ask what consumers want:

1. self driving that needs intervention once every few hundred miles and works anywhere

2. self driving that works only in low volume and predictable traffic, in a small section of a single city, sometimes still gets stuck and blocks intersections, breaks as soon as the road conditions change, but is entirely hands off

i think most people want #1, and the usage stats agree with that

Sure, but the major difference is that Waymo already operates a real service accessible to absolutely anyone in certain cities that actually transports passengers from point A to point B with no driver present in the car. Tesla FSD is not at that point yet.

I am not even saying this as some Tesla FSD hater, I used it a ton over the years and am generally happy with it. But claiming that, in its current state, it has a lead over Waymo is a bit questionable.

waymo is using an approach that is inherently limited and not scalable. they went for short term gains at the expense of knee-capping themselves long term. it's a fundamentally less robust approach.

FSD is really, really close. The recent 12.5 software is a big step forward for them, and their rate of improvement is really, really fast these days.

i appreciate the level headed discussion

Fair point about recent software updates, I can definitely believe that (I moved to a place where a car is not a necessity at the beginning of this year, so I didnt get to test the recent updates myself).

However, I can still see someone claiming that Waymo might have an edge in some form as a fair take, given they indeed have something that FSD, at the moment, doesn’t (which is full operation with no driver). Whether that is a viable long-term approach compared to FSD (or whether it is hitting the ceiling that FSD doesn’t have) is a solid point though.

Pretty far behind who?

Yeah I have noticed all of the actually useful ML papers I have read come from Deep Mind and Google because they seem more interested in stuff that actually matters like industrial control systems and other boring stuff that makes money

What should we take away from the AI Overview snafu? Is it an example of an ill-conceived concept for an AI product? Or, is it an example of a technically inferior model hampering an otherwise good product?

Insufficient red teaming, presumably due to release pressure from higher up, and lack of work processes requiring more red teaming.

I can't even fathom being an engineer on that project and not throwing it a few curveballs to see what happens. Someone must have known.

why aren't they a research institution then? We live in a funny world where academia is ridden with paper mills and fail to do its job, and a for profit company like Google is doing academia's job and... fail to do its job on applications and commercialization also.

They kind of are. They do research and spin of the results as companies or products. Waymo, SandboxAQ, Isomorphic Labs, Aalyria, Niantic, Verily,

https://x.company/

I know nothing here, but isn't that also the difference: Meta is just trying out things, and can find that killer application later. Google meanwhile existentially feels that search must be the killer application, and tries to shoehorn everything into that. And in doing so put the bar for success so, so high, ignoring where things are really at.

Problem is that Google core product is not search, but ads. Over years they were slowly degrading search experience because making search better, faster and more efficient means we would spend less time watching ads.

Situation is the same on YouTube where Google have insane market monopoly, but still depreoritize actual creators who made their platform successful to appeal to brands and push more and more ads. Demonetization of everything ever slightly controversial. Somehow Google index whole internet 100 times a day, their OS run on majority of smartphones, their browser is monopoly, they know everything about everyone, but can't even filter spam efficiently.

I never liked or used Facebook / Instagram much, but at least Meta core products are not deteriorating at same insane pace Google products are. I can still use WhatsApp daily without ads being fed down my throat.

> Over years they were slowly degrading search experience because making search better, faster and more efficient means we would spend less time watching ads.

That honestly sounds like revisionism to fit your narrative. Occam's razor would decree that the degradation simply came through SEO, and it's inverse correlation of quality (SEO means essentially pushing terrible content higher for advertising revenue)

It most likely wasn't Google's aim to reduce their search quality, it was just a side effect of them becoming too successful, that it became unreasonably profitable to game the system

Google actually have a daily history of internet available on palm of their hand as well as insane access to both user behaviour (Android, Chrome) and website statistics (Analytics, Adsense, Google DNS, etc).

They had the power to not succumb to "SEO" content spam, but this wouldn't improve their ads revenue baseline. Pushing their own services and centralised internet on other hand...

The feedback loop with Google is they got better and better at indexing everything, so more "indexable" content got created to take advantage of it. There was probably some sort of tipping point where normal human behavior and content-generation couldn't keep up with Google's increased capability, so the increase came in generated/copied/llmd/spam content which outpaced the natural growth of real content. This further pushed google to get better and better at "indexing" everything, making the issue worse.

Cherry on top for Google: Due to the sheer "volume" of content this loop ended up creating, it posed a larger moat for any new entrants into the search market thus strengthening their monopoly position.

There will be a subsequent tipping point: Where due to the sheer volume of generated/spammed/seo crap on the internet, the volume of "real" content will be so small that it'll become within reach of being indexable again by smaller players. We're already seeing this to some extent, and it'll only get better.

> Where due to the sheer volume of generated/spammed/seo crap on the internet, the volume of "real" content will be so small that it'll become within reach of being indexable again by smaller players.

Will it be discoverable, though?

It wasn't SSO that put ads before the real search results.

https://news.ycombinator.com/item?id=40133976

SEO didn't get rid of the yellow backgrounds on ads either. A few times during the peak of LLM fever I found that I could search an LLM-related term and get zero organic results above the fold. Every visible "result" was actually an ad disguised as a result.

For a long time Search was what Google needed you to use in order to collect massive amounts of personal information about you. Today they have Android, and Gmail, and youtube, and Nest, and fitbit, and Maps, and Chrome all feeding google a steady stream of data on the most intimate details of your life. Google doesn't need Search anymore to track what you're doing online or to learn about what you're doing offline, so Search gets neglected and filled with an ever increasing amount of ads and trash.

If you have a website, all the spam and AI garbage filling up Google's search results can make it very hard for people to find your website using Google. Don't worry though, because you can pay Google to fix the problem Google has made for you, in fact, you might increasingly feel like you need to pay Google to fix the problem they're causing you or people won't be able to find your website at all. You can try to tell people your domain name, but Google has trained the pubic to search from the address bar and they've mostly forgotten what a URL even is.

Not sure how meta is killing it. Their AI integrations into their apps like whatsapp, instagram are beyond useless. They were stuffed in there just to fool markets into beliving meta is some sort of ai player.

> That's the difference of a founder-led company vs. market led.

Zuckerberg is the most unimaginative CEOs of the them all. Meta never had a single original product ( save for portal device). All their products are acquisitions. Meta is phenomenally bad at innovation.

Zuckerberg seems to have pulled some sort PR campaign to cleanup his image. But we all know FB is still a shady company run by a shady man, nothing has changed in its rotten core. It was fined billions of dollars just this week by Texas.

Meta is far from a "founder led company" . All the founders of the apps he buys leave immediately and are in turn run by managment consultant types like Adam Mosseri .

> Once Meta finds the killer app for VR

Sad to see that ppl are still buying into metaverse hailmary zuckerburg scammed to convince markets that meta is still an innovative company amid slowing user growth. I still don't understand how metaverse fraud pulled by zuckerberg wasn't some sort of SEC violation.

Meta can still very much be an innovative company regardless of user growth though, right? They are very much innovating in the AI research space, and are absolutely leading in open* source research/models, even if their AI integrations are flopping.

Sure, none of this research might be obviously contributing to their products in a share market sense, but it's still innovation which is contributing to the (entirely separate to their "product") AI research community.

> Meta can still very much be an innovative company regardless of user growth though, right?

only till market tolerates it . they will shut it down once patience runs out like they did with metaverse nonsense

Meta has insane distribution. Even if they don't match the best AI integration in a couple of months, they can push it to 1 billion people in days when it hits.

Same thing for AR/VR, they can afford to invest massively in different plans, use cases and form factors. Fitness, work, social, metaverse, gaming, hardware... It's a collection of bets and in this day an age, one great product-market fit can scale into billions of revenue.

In a recent interview Zuck stated that he was surprised by how well the RayBans worked. It's also a great synergy with this new wave of AIs. Because they invested so much they might well get ahead in this race

> That's the difference of a founder-led company vs. market led.

Are they really so different?

Facebook throws things at the wall, even expensive things without a clear path to profitability, such as Llama.

Google throws things at the wall, even expensive things without a clear path to profitability, such as waymo, google glass, google fiber, stadia, and everything on https://killedbygoogle.com

Facebook made a big companywide re-orientation to deliver a vision that flopped - the metaverse.

Google made a big companywide re-orientation to deliver a vision that flopped - google plus.

Facebook renamed themselves to Meta. Google renamed themselves to Alphabet.

Facebook has an AI research division founded by a French-American computer science professor and Turing award winner. Google has an AI research division founded by a British-Canadian computer science professor and Turing award winner.

Facebook released a widely used open source Python ML library with a camelcase name, PyTorch. Google released a widely used open source Python ML library with a camelcase name, TensorFlow.

Perhaps they're following the same playbook, it's just luck that Facebook's gambles paid off recently.

Well, Meta is not as arbitrary with closing projects. Google closes things which are getting traction, that's a really disturbing aspect of interna politics and incentives

Also you mixed up tensorflow and pytorch.

afaik tensorflow is by google brain team and pytorch is by FAIR (Facebook AI Research)

* That's the difference of a founder-led company vs. market led. Google is mostly concerned on short-term goals so they don't report a single bad quarter or has too high CAPEX on a project without profit on its sight (like VR).*

It really depends on the founder. Some are very unhappy to see the stock price drop even if they don’t really need the money this second.

Also, it’s a mixed bag. I personally think Zuck is wrong on VR but right on AI.

> Once Meta finds the killer app for VR

The killer app for VR is porn.

If Meta doesn't care for money, then I really don't understand what Facebook has devolved into. It used to be a chronological feed of what your friends are doing. Now all I see is an infinite steam of anger-bait, thirst traps and ai-generated spam. How can it still be controlled by the same guy?

>Now all I see is an infinite steam of anger-bait, thirst traps and ai-generated spam. How can it still be controlled by the same guy?

The stream isn't controlled by Zuck, it's controlled by AI, an AI trained to "give the people whatever they want" to maximise engagement. The more deep learning you put into a product, the harder it is to directly control the output.

They can easily control the output. You just need a filter.

They don't care about the output

They care a great deal about the output, but their priorities and goals are very different from that of their users.

The same holds true for Google. It's not that Facebook could design and build social networking sites that aren't terrible for the health or their users and society, or that Google could build a better search engine/search engine UI. They absolutely can, but that's less profitable.

If Google can't solve the SEO problem what makes you think Facebook "just needs a filter"?

Im answering inside a context. Telling me that Facebook can't control its output because of AI / complexity, than thats just wrong.

Why doesn't Google have higher quality results? It's about just controlling the output too.

No its a different thing or argument.

If facebook can't 'control' the output because of AI or whatever as an argument, thats facebook specific.

Can google control its output? Yes for censoring etc. but Google shows you external self indexed and self analyzed content, Facebook doesn't do that. It has logged in users using their tools to create content and Facebook uses their internal algo to show whatever they want to you.

I find people's faith in the AI disturbing. I'm a Canadian who's never set foot in the U.S. or a striper bar, but for a long while it was intent on showing me ads for bikini bars in Florida or Hooters. Even if you assumed I loved titty bars, I'm not even in the right country!

Plus, it showed me what I want so much I completely left the platform.

Please do not use the term "AI" as a way to deflect responsibility. People use AI. The people who use it are responsible for its use.

> The stream isn't controlled by Zuck, it's controlled by AI

The laziest possible excuse, one rung below "just following orders" even: the "it wasn't me, it was the machine I built" defence.

Google is notorious for in creating and sunsetting projects who's only purpose is yo showcase talent without regard for any other business model outside of their primary money makers. No support, no open sourcing, just something to show off and show that they are in the innovative app space.

I don’t think this is true at all. Alphabet has Waymo and Isomorphic Labs, trying to commercialise solutions to problems with very, high impact. Time scale is at least a decade

That's a lot extrapolation from a few open models.

> Once Meta finds the killer app for VR

I like that it's taken as a premise that billionaires do have some kind of superhuman vision, that it's not even a matter of "if" but "when" Zuck's VR gamble pays off.

they are both, they have a different focus though. E.g. AlphaProof and AlphaGeometry 2 are "killing it" also in their fields.

But for consumers Meta seems to have a lead indeed...

Google Gemini recently enabled 32k context in their free tier.

Yet they still locking access to reading files from Google Drive to some "advanced" plan or might be it's not implemented at all. Just 6 months ago or something ago "Bard" could do it and it was one of few actually useful features to summarise PDFs.

I really dont want to have any dependencies in my workflow on Google products and features because they'll surely break them.

Hmm, I just tried it and it works fine. I had to enable workspace access under https://gemini.google.com/extensions

It's enabled for me. It can now even access list of files on Google Drive and print them. Yet every time I trying to make it read PDF it's just doesn't work:

> Sorry, I can't access your Google Drive files. You'll need to upload the file or describe the content of the pages to me.

Any example of prompt you using that is working for you?

Yes, I had to prefix it with @Google Drive. E.g. "@Google Drive Hi, Can you extract the table at page 3 of the ti lm317 datasheet?"

Wow. Okay actually it does seems to work now.

When they initially moved from Bard to Gemini nothing worked for sure, but with your trick it does work.

Also find out right now that it can actually work if you ask to look for a file on Google Drive and within the same command ask to do something to specific file.

Thank you.

I still find Gemma 2 9b to be better than Llama 3.1 8b often

Yeah, Google Deepmind is doing useless stuff like protein folding and math.

While Meta excels at chatbot of the day

Totally agree with you

Facebook has nothing besides their social network. They care very little about a lot of stuff (after all they didn't care about fakenews and other things).

The metaverse was a huge waste of money.

Their AI Character thing is weird.

They cater to a lot of small businesses, thats definitly helping them.

But besides that? They have very little to do what Google, Microsoft, Apple and co are actually doing.

Reply on news.ycombinator.com