Google “We have no moat, and neither does OpenAI” | Hacker News Reader

Google “We have no moat, and neither does OpenAI”

semianalysis.com

2,455 pointsklelatti1,039 comments

The current paradigm is that AI is a destination. A product you go to and interact with.

That's not at all how the masses are going to interact with AI in the near future. It's going to be seamlessly integrated into every-day software. In Office/Google docs, at the operating system level (Android), in your graphics editor (Adobe), on major web platforms: search, image search, Youtube, the like.

Since Google and other Big Tech continue to control these billion-user platforms, they have AI reach, even if they are temporarily behind in capability. They'll also find a way to integrate this in a way where you don't have to directly pay for the capability, as it's paid in other ways: ads.

OpenAI faces the existential risk, not Google. They'll catch up and will have the reach/subsidy advantage.

And it doesn't end there. This so-called "competition" from open source is going to be free labor. Any winning idea ported into Google's products on short notice. Thanks open source!

The part of the post that resonates for me is that working with the open source community may allow a model to improve faster. And, whichever model improves faster, will win - if it can continue that pace of improvement.

The author talks about Koala but notes that ChatGPT is better. GPT-4 is then significantly better than GPT-3.5. If you've used all the models and can afford to spend money, you'd be insane to not use GPT-4 over all the other models.

Midjourney is more popular (from what I'm seeing) than Stable Diffusion at the moment because it's better at the moment. Midjourney is closed-source.

The point I'm wanting to make is that users will go to whoever has the best model. So, the winning strategy is whatever strategy allows your model to compound in quality faster and to continue to compound that growth in quality for longer.

Open source doesn't always win in producing better quality products.

Linux won in servers and supercomputing, but not in end user computing.

Open-source databases mostly won.

Chromium sorta won, but really Chrome.

Then in most other areas, closed-source has won.

So one takeaway might be that open-source will win in areas where the users are often software developers that can make improvements to the product they're using, and closed-source will win in other areas.

Fantastic article if you are quick to just go to the comments like I usually do, don't. Read it.

One of my favorites: LoRA works by representing model updates as low-rank factorizations, which reduces the size of the update matrices by a factor of up to several thousand. This allows model fine-tuning at a fraction of the cost and time. Being able to personalize a language model in a few hours on consumer hardware is a big deal, particularly for aspirations that involve incorporating new and diverse knowledge in near real-time. The fact that this technology exists is underexploited inside Google, even though it directly impacts some of our most ambitious projects.

Anyone has worked with LoRa ? Sounds super interesting.

This gets attention due to being a leak, but it’s still just one Googler’s opinion and it has signs of being overstated for rhetorical effect.

In particular, demos aren’t the same as products. Running a demo on one person’s phone is an important milestone, but if the device overheats and/or gets throttled then it’s not really something you’d want to run on your phone.

It’s easy to claim that a problem is “solved” with a link to a demo when actually there’s more to do. People can link to projects they didn’t actually investigate. They can claim “parity” because they tried one thing and were impressed. Figuring out if something works well takes more effort. Could you write a product review, or did you just hear about it, or try it once?

I haven’t investigated most projects either so I don’t know, but consider that things may not be moving quite as fast as demo-based hype indicates.

Having enough scale to perpetually offer free/low-cost compute is a moat. The primary reason ChatGPT went viral in the first place was because it was free, with no restrictions. Back in 2019, GPT-2 1.5B was made freely accessible by a single developer via the TalkToTransformers website, which was the very first time many people talking about AI text generation...then the owner got hit with sticker shock from the GPU compute needed to scale.

AI text generation competitors like Cohere and Anthropic will never be able to compete with Microsoft/Google/Amazon on marginal cost.

Investors are obsessed with moats, but people have to realize that the entire world runs on business that have no moats.

There are no moats to being a plumber, a baker, a restaurant...

The moat concept is predominant because the idea that everything must make billions have infected the debate about businesses.

It's all about being a unicorn, a giant, a monopoly, making every body at the top billionaires, and it's like there is no other way to live.

Except that's not how most people do live, even entrepreneurs.

Even Apple, which today is the typical example of a business with a moat, didn't start with "we can't get into this computer business, we'd have no moat".

They have a moat now, but it's a consequence of all the business decisions and the thing they built after many decades.

They didn't start their project by the moat. The started their project by providing value and marketing it.

I have been toying around with Stable Diffusion for a while now and becoming comfortable with the enormous community filled with textual inversions, LoRAs, hyper networks and checkpoints. You can get things with names like “chill blend”, a fine-tuned model on top of the SD with the author’s personal style.

There is something called automatic1111 which is a pretty comprehensive web UI for managing all these moving parts. Filled to the brim with extensions to handle AI upscaling, inpainting, outpainting, etc.

One of these is ControlNet where you can generate new images based on pose info extracted from an existing image or edited by yourself in the web based 3d editor (integrated, of course). Not just pose but depth maps, etc. All with a few clicks.

The level of detail and sheer amount of stuff is ridiculous and it all has meaning and substantial impact on the end result. I have not even talked about the prompting. You can do stuff like [cow:dog:.25] where the generator will start with a cow and then switch over at 25% of the process to a dog. You can use parens like ((sunglasses)) to focus extra hard on that concept.

There are so called LoRAs trained on specific styles and/or characters. These are usually like 5-100MB and work unreasonably well.

You can switch over to the base model easily and the original SD results are 80s arcade game vs GTA5. This stuff has been around for like a year. This is ridiculous.

LLMs are enormously “undertooled”. Give it a year or so.

My point by the way is that any quality issues in the open source models will be fixed and then some.

I'm a bit sceptical of the "no moat" proposition because (a) ChatGPT 4.0 really does seem in a different league and (b) it's clearly very hard to run. I haven't seen anything from the explosion of open source / community efforts that comes close for general applications.

The take in the post rings of the classic trademark Google arrogance where they assume that if somebody else can do it they can do it better if they just try - where the challenge of "just trying" is discounted to zero. In reality, "Just trying" is massively important and sometimes all that is important. The gap between unrefined model output and the level of polish and refinement that is apparent with ChatGPT 4 may appear technically small but it's the whole difference between a widely applicable and usable product and something that can't be more than a toy. I'm not sure Google has it in it any more to really fight for something they want to achieve that level of polish.

Great read, but I don't agree with all of these points. OpenAI's technological moat is not necessarily meaningful in a context where the average consumer is starting to recognize ChatGPT as a brand name.

Furthermore, models which fine-tune LLMs are still dependent on the base model's quality. Having a much higher quality base model is still a competitive advantage in scenarios where generalizability is an important aspect of the use case.

Thus far, Google has failed to integrate LLMs into their products in a way that adds value. But they do have advantages which could be used to gain a competitive lead: - Their crawling infrastructure could allow their to generate better training datasets, and update models more quickly. - Their TPU hardware could allow them to train and fine-tune models more quickly. - Their excellent research divisions could give them a head start with novel architectures.

If Google utilizes those advantages, they could develop a moat in the future. OpenAI has access to great researchers, and good crawl data through Bing, but it seems plausible to me that 2 or 3 companies in this space could develop sizeable moats which smaller competitors can't overcome.

FWIW I posted Simon's summary because it's what I encountered first, but here's the leaked document itself[0].

Some snippets for folks who came just for the comments:

> While our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B. And they are doing so in weeks, not months.

> A tremendous outpouring of innovation followed, with just days between major developments (see The Timeline for the full breakdown). Here we are, barely a month later, and there are variants with instruction tuning, quantization, quality improvements, human evals, multimodality, RLHF, etc. etc. many of which build on each other.

> This recent progress has direct, immediate implications for our business strategy. Who would pay for a Google product with usage restrictions if there is a free, high quality alternative without them?

> Paradoxically, the one clear winner in all of this is Meta. Because the leaked model was theirs, they have effectively garnered an entire planet’s worth of free labor. Since most open source innovation is happening on top of their architecture, there is nothing stopping them from directly incorporating it into their products.

> And in the end, OpenAI doesn’t matter. They are making the same mistakes we are in their posture relative to open source, and their ability to maintain an edge is necessarily in question. Open source alternatives can and will eventually eclipse them unless they change their stance. In this respect, at least, we can make the first move.

[0]: https://www.semianalysis.com/p/google-we-have-no-moat-and-ne...

This is so indicative of Google culture missing the point. The idea of spending $10M training a single model is treated as a casual reality. But “tHaNk GoOdNeSs those generous open source people published their HiGh QuAlItY datasets of ten thousand examples each. Otherwise we’d have no way of creating datasets like that…” :| the sustainable competitive advantage has been and will continue to be HUGE PROPRIETARY DATASETS. (Duh - this is as true for new AI as it was for old AI = ad targeting). It was the _query+click pairs_ that kept Google dominant all these years, not the brilliant engineers. They had all of humanity labeling the entire internet with “when I click on this page/ad for this query I do/don’t search again” a billion+ times a day for a decade. For good measure they’ve also been collecting your email, your calendar, and your browsing habits for nearly as long. The fact that they’ve managed to erase that historic advantage from their collective consciousness (presumably because AI researchers would rather not spend time debugging data labeling UI) is strange to me. It at least deserves a mention in a strategy memo like this. Not vague platitudes about “influence through innovation.” Spend that $10M you were going to spend on a training run as $9.9999M on a private dataset, then the remaining $100 on training. Better still, build products that gets user behavior to train your models for you. Obviously.

We’re going to watch the biggest face plant in recent economic history if they can’t get this one together. I can’t decide if that makes me happy about an overdue changing of the guard in the Valley or sad about the fall of a once great company.

It’s not about the models! Model training is a commodity! It’s about the data! Come on guys.

This looks like a personal manifesto from an engineer who doesn't even attempt to write it on behalf of Google? The title is significantly misleading.

This is easily among the rare highest quality articles/comments I've read in the past weeks, perhaps months (on LLMs/AI since that's what I am particularly interested in). And this was for internal consumption before it was made public. Reinforces my recent impression that so much that's being made for public consumption now is shallow and it is hard to find the good stuff. And sadly, increasing so even on HN. As I write this, I acknowledge I discovered this on HN :) Wish we had ways to incentivize the public sharing of such high-quality content that don't die at the altar of micro rewards.

Really interesting to look at this from a product perspective. I've been obsessively looking at it from an AI user perspective, but instead of thinking of it as a "moat", I just keep thinking of the line from Disney's The Incredibles, "And when everyone is super, no one will be."

Every app that I might build utilizing AI is really just a window, or a wrapper into the model itself. Everything is easy to replicate. Why would anyone pay for my AI wrapper when they could just build THING themselves? Or just wait until GPT-{current+1} when the model can do THING directly, followed swiftly by free and open source models being able to do THING as well.

I am amazed that people haven't gotten used to these "internal Google doc leaks".

This is just the opinion of some random googler, one among over 100,000.

For some reason random googlers seem like to write random docs on hot topics and share it widely across the company. And someone, among those over 100,000 googlers, ends up "leaking" the opinion of that person to outside Google.

This is more like a blog post of some random dude over the Internet expressing his opinion. The fact that random dude ended up working at Google should not bear much on evaluating the claims in the doc.

A website published that with a title "Google ..." is misleading. The accurate title would be "Some random googler: ..."

I've been using Stable Diffusion to generate cover images for the music I release & produce for others. It's a massive time saver compared to comping together the release art using image editing software, and a lot cheaper than working with artists, which just doesn't make sense financially as an independent musician.

It's a little bit difficult to get what you want out of the models, but I find them very useful! And while the output resolution might be quite low, things are improving & AI upscaling also helps a lot.

I think from the perspective of a Google researcher/engineer, it must be alarming to see the crazy explosion going on w/ LLM development. We've gone from just one or two weirdos implementing papers (eg https://github.com/lucidrains?tab=repositories who's amazing) to now an explosion where basically every dev and PhD student is hacking on neat new things and having a field day and "lapping" (eg productizing) what Google Research was previously holding back.

And we're also seeing amazing fine-tunes/distillations of very useful/capable smaller models - there's no denying that things have gotten better and more importantly, cheaper way faster than anyone expected. That being said, most of these are being trained with the help of GPT-4, and so far nothing I've seen being done publicly (and I've been spending a lot of time tracking these https://docs.google.com/spreadsheets/d/1kT4or6b0Fedd-W_jMwYp...) gets close in quality/capabilities to GPT-4.

I'm always rooting for the open source camp, but I think the flip-side is that there are still only a handful of organizations in the world that can train a >SoTA foundational model, and that having a mega-model is probably a huge force multiplier if you know how to take advantage of it (eg, I can't imagine that OpenAI has been able to release software at the pace they have been without leveraging GPT-4 for co-development; also can you distill or develop capable smaller models without a more capable foundational model to leverage?). Anthropic for example has recently taken the flip side of the "no moat" argument, arguing that there is a potential winner-take-all scenario where the lead may become insurmountable if one group gets too far ahead in the next couple years. I guess what we'll just have to see, but my suspicion, is that the crux to the "moat" question is going to be whether the open source approach can actually train a GPT-n++ system.

> Giant models are slowing us down. In the long run, the best models are the ones which can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime.

Maybe this is true for the median query/conversation that people are having with these agents - but it certainly has not been what I have observed in my experience in technical/research work.

GPT-4 is legitimately very useful. But any of the agents below that (including ChatGPT) cannot perform complex tasks up to snuff.

So I use ChatGPT every day. I like it a lot and it is useful but it is overhyped. Also from 3.5 to 4 the jump was nice but seemed relatively marginal to me.

I think the head start OpenAi has will vanish. Iteration will be slow and painful giving google or whoever more than enough time to catch up.

ChatGPT was a fantastic leap getting us say 80% to Agi but as we have seen time and time again the last 20% are excruciatingly slow and painful (see Self driving cars).

Not only they have no moat, Open source models are uncensored and this is huge. Censorship is not just political , it cripples the product to basically an infantile stage and precludes so many applications. For once, it is a liability

But this article doesn't state the very obvious: When will google (the inventor of Transformer, and "rightful" godfather of modern LLMs) , release a full open source, tinkerable model better than LLaMa?

(To the dead comment below, there are many uncensored variations of vicuna)

My feeling on this is “f** yeah, and f** you [google et al]”

How much computing innovation was pioneered by community enthusiasts and hobbyists that have been leveraged by these huge companies.

I know meta, googlr, msft et al give back in way of opensource, but it really pales in comparison to the value those companies have extracted.

I’m a huge believer in generative AI democratizing tech.

Certainly I’m glad to pay for off-the-shelf custom tuned models, and for software that smartly integrates generative AI to improve usage, but not a fan of gate keeping this technology by a handful of untrustworthy corporations.

The author is overly optimistic with the current state of open source LLMs, (e.g., Koala is very far away from matching ChatGPT performance). However, I agree with their spirit, Google has been one of the most important contributors to the development of LLMs and until recently they've been open sharing their model weights under permissive licenses, they should not backtrack to closed source.

OpenAI has a huge lead in the closed source ecosystem, Google's best bet is to take over the open source ecosystem and build on top of it, they are still not late. Llama based models don't have a permissive license, and a free model that is mildly superior to Llama could be game changing.

I remember I was at Microsoft more than a decade ago now and at the time there was a lot of concern about search and how far Bing lagged behind Google in geospatial (maps).

After some initial investment in the area I was at a presentation where one of the higher ups explained that they'd be abandoning their investment because Google Maps would inevitably fall behind crowdsourcing and OpenStreetMap.

Just like Encarta and Wikipedia we were told - once the open source community gets their hands on something there's just no moat from an engineering perspective and once it's crowdsourced there's no moat from a data perspective. You simply can't compete.

Of course it's more than a decade later now and I still use Google Maps, Bing Maps still suck, and the view times I've tried OpenStreetMaps I've found it far behind both.

What's more every company I've worked at since has paid Google for access to their Maps API.

I guess the experience made me skeptical of people proclaiming that someone does or does not have a moat because the community will just eat away at any commercial product.

> The document is only the opinion of a Google employee, not the entire firm

The title makes it seem like this is some official Google memo. The company has 150K employees and 300K different opinions on things. Can't go chasing down each one and giving it importance.

"Many of the new ideas are from ordinary people."

Yeah. Google can fuck right off. Maybe this attitude is what got them in the weeds in the first place.

I don't know if I agree with the article. I recall when Google IPO'ed, nobody outside of Google really knew how much traffic they had and how much money they were making. Microsoft was caught off-guard. Compare this to ChatGPT: My friends, parents, grandparents, and coworkers (in the hospital) use ChatGPT. None of these people know how to adapt an open source model to their own use. I bet ChatGPT is vastly ahead in terms of capturing the market, and just hasn't told anyone just how far. Note that they have grown faster in traffic than Instagram and TikTok, and they are used across the demographics spectrum. They released something to the world that astounded the average joe, and that is the train that people will ride.

"People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality. . ."

I'll take the opposite side of that bet - MSFT / Goog / etc in the providers side will drive record revenues on the back of closed / restricted models:

1 - Table stakes for buying software at enterprise level is permissions based management & standardized security / hardening.

2 - The corporate world is also the highest value spender of software

3 - Corp world will find the "proprietary trained models" on top of vanilla MSFT OpenAI or Goog Bard pitch absolutely irresistible - creates a great story about moats / compounding advantages etc. And the outcome is going to most likely be higher switching costs to leave MSFT for a new upstart etc

This has been my speculation about the people pushing for regulation in this space: it’s an attempt at regulatory capture because there really is little moat with this tech.

I can already run GPT-3 comparable models on a MacBook Pro. GPT-4 level models that can run on at least higher end commodity hardware seem close.

Models trained on data scraped from the net may not be defensible via copyright and they certainly are not patentable. It also seems possible to “pirate” models by training a model on another model. Defending against this or even detecting it would be as hard as preventing web scraping.

Lastly the adaptive nature of the tech makes it hard to achieve lock in via API compatibility. Just tell the model to talk a different way. The rigidity of classical von Neumann computing that facilitates lock in just isn’t there.

So that leaves the old fashioned way: frighten and bribe the government into creating onerous regulations that you can comply with but upstarts cannot. Or worse make the tech require a permit that is expensive and difficult to obtain.

OpenAI is further along than most of us are aware.

The ability to connect these models to the web, to pipe up API access to different services and equip LLMs to be the new interface to these services and to the worlds information is the real game changer.

Google cannot out innovate them because they are a big Corp rife with googly politics and challenges of overhead that come with organizational scale.

I would be curious to see if there are plans to spin off the newly consolidated AI unit with their own PnL to stimulate that hunger to grow and survive and then capitalize them accordingly. Otherwise they are en route to die a slow death once better companies come along.

The real moats in this field will come from the hardware industry. It's way too expensive to train these models on general purpose compute. Vertically designed silicon that brings down the unit economics of training and inference workloads are already being designed, in industry and in academia.

I agree that both Google and OpenAI are losing the race to open-source in AI research. However, a more critical issue to Google is their struggle to compete with OpenAI in LLM-based search engines. Google's entire business model mostly relies on ads (77.6% in Q4 2022). OpenaAI is developing LLM-based products that people apparently love (100M users in 2 months) and seem to use it as search engines. This poses a greater risk to Google than just losing ground in research since it could ultimately lead to the loss of their ad-generated income.

Google's moat is its data set. Imagine training an generative AI LLM on the entire set of YouTube training videos. No one else has this.

> At the beginning of March the open source community got their hands on their first really capable foundation model, as Meta’s LLaMA was leaked to the public

A Prometheus moment if I’ve ever seen one.

puts on tinfoil hat They probably intentionally published this to raise awareness of how viable it is for individuals to outperform OpenAI and Meta. Google seems to be the farthest behind, they have the most to gain by the others losing their lead to individuals.

The Simon Willison coverage is great. Simon is cited in the "why we should have seen it coming" section, for his Stable Diffusion Moment piece. He nicely covers the key points of this paper:

> The premise of the paper is that while OpenAI and Google continue to race to build the most powerful language models, their efforts are rapidly being eclipsed by the work happening in the open source community.

Not to dilute from this beloved point, but also covers other key notes well too:

> Where things get really interesting is where they talk about “What We Missed”. The author is extremely bullish on LoRA—a technique that allows models to be fine-tuned in just a few hours of consumer hardware, producing improvements that can then be stacked on top of each other

https://simonwillison.net/2023/May/4/no-moat/

Overall I take this as fairly happy news. It's a trend humanity stubbornly keeps trying to resist: open source wins.

It's great the barrier to innovation is so much less than expected, that so much experimentation is possible from atop the existing models.

> "We have no moat, and neither does OpenAI"

and neither does Coca Cola and Cadbury. Yet biggest monopolies are found in these places. Because the competitors will not be differentiated enough for users to switch from the incumbent.

But G-AI is still nascent and there's lots of improvements to be had. I suspect better tech is a moat but ofcourse Google is oblivious to it.

> Paradoxically, the one clear winner in all of this is Meta. Because the leaked model was theirs, they have effectively garnered an entire planet's worth of free labor. Since most open source innovation is happening on top of their architecture, there is nothing stopping them from directly incorporating it into their products

I disagree. The model itself is released under GPL3 and no longer "theirs" (Google or OpenAI can use it). And Meta probably has a zoo of such models and I didn't see them use any of the work the community did.

I don't think they "released" LLaMA weights strategically to weaken OpenAI (their overall strategy and market analysis would probably too inert to predict the open source explosion). It probably was a decision by a smaller research team within the company and approved by uninformed executive. Meta could have stepped up and nurtured this small-LLM renaissance, they opted for DMCA hammer instead.

While this post champions the progress made by OSS, it also mentions that a huge leap came from Meta releasing Llama. Would the rapid gains in OSS AI have came as quickly without that? Did Meta strategically release Llama knowing it would destroy Google & OpenAI's moats?

I've been saying this kind of thing for a while about open source. There is a lot of innovation happening outside of the corporate sector which has been neglected. Not just that; it has been essentially covered up. These projects have not been allowed to get any attention because big tech controls all the media channels. A lot has been happening outside of AI too.

A lot of people are shocked at how this open source innovation just came out of nowhere but those who work in open source, in those areas, aren't so surprised because they've been going at it for years.

"Paradoxically, the one clear winner in all of this is Meta. Because the leaked model was theirs, they have effectively garnered an entire planet's worth of free labor. Since most open source innovation is happening on top of their architecture, there is nothing stopping them from directly incorporating it into their products."

An interesting thought. Are the legal issues for derived works from the leaked model clarified or is the legal matter to be resolved at a later date when Meta starts suing small developers?

> Paying more attention to their work could help us to avoid reinventing the wheel.

There's nothing us humans love more than reinventing the wheel. I've seen it over and over again, years of work and hundreds of millions of dollar spent re-solving problems and re-writing systems -- only to replace them with a new set of slightly different problems. I think we greatly over estimate the ability of our species to accumulate knowledge, which is perhaps where these generative systems come into play.

I find it strange people are saying facebook's leak was the 'Stable Diffusion' moment for LLMs. The license is awful and basically means it can't be used in anything involving money legally.

Facebook has a terrible reputation, and if they can open source their model, it would transform their reputation at least among techies.

https://github.com/facebookresearch/llama/pull/184

It seems like everyone is so focused on LLMs are magic smartness machines that there isn't much analysis of them as better search (maybe "search synthesis"). And original search was a revolutionary technology, LLM as just better search are revolutionary.

Like original search, the two application aspects are roughly algorithm and interface. Google years ago won by having a better interface, an interface that usually got things right the first time (good defaults are a key aspect of any successful UI). ChatGPT is has gotten excitement by taking a LLM and making it generally avoid idiocy - again, fine-tuning the interface. Google years ago and ChatGPT got their better results by human labor, human fine tuning, of a raw algorithm (In ChatGPT's case, you have RLHF with workers in Kenya and elsewhere, Google has human search testers and years ago used DMOZ, an open source, human curated portal).

Google's "Moat" years ago was continuing to care about quality. They lost this moat over the last five years imo by letting their search go to shit, become focused always on some product for any given search. This is what has made ChatGPT especially challenging for Google (it would be amazing still but someone comparing to Google ten years ago could see ways Google was better, present day Google has little over ChatGPT as UI. If Google had kept their query features as they added AI features, they'd have a tool that could claim virtues through still not as good).

And this isn't even considering of updating a model and the question of how the model will be monetized.

> They are doing things with $100 and 13B params

Not that I disagree with the general belief that OSS community is catching up, but this specific data point is not as impactful as it sounds. Llama cannot be used for commercial purposes, and that $100 was spent on ChatGPT, which means we still depended on proprietary information of OpenAI.

It looks to me that the OSS community needs a solid foundation model and a really comprehensive and huge dataset. Both require continuous heavy investment.

I would lean towards agreeing. And I definitely think AI companies should not try to make their money on inference.

If there is a well performing model being deployed it is possible to train a similar model while not having to eat the cost of exploration. Ie. it is only the the cost of training said model.

ChatGPT would probably die in a couple of weeks, if an equivalent, free, product came out that people could run on their computers.

Is there any evidence this is real? It reads like an article written not for google but for fans of open source in their competitors supposed voice.

“Some of the most interesting questions about CAS [Complex Adaptive Systems] have to do with their relations to one another. We know that such systems have a tendency to spawn others. Thus biological evolution gave rise to thinking, including human thought, and to mammalian immune systems; human thought gave rise to computer-based CAS; and so on.”

- Murray Gell Mann, “Complex Adaptive Systems”

Google's moat is its data set. Imagine training an generative AI LLM on the entire set of YouTube training videos. No one else has this.

This looks very fake to me. I might be wrong. Yet, there is no "document" that was leaked, the original source is some blog post. If there is a document, share the document. Shared by "anonymous individual on discord who granted permission for republication"... I don't know. If it was shared by anonymous, why ask for permission? Which discord server?

Another magnificent unsurprising set of correct prediction(s) [0] [1] [2] and as triumphantly admitted by Google themselves on open source LLMs eating both of their (Google) and OpenAI's lunch.

"When it is the race to the bottom, AI LLM services, like ChatGPT, Claude (Anthropic), Cohere.ai, etc are winning the race. Open source LLMs are already at the finish line."

[0] https://news.ycombinator.com/item?id=34201706

[1] https://news.ycombinator.com/item?id=35661548

[2] https://news.ycombinator.com/item?id=34716545

The only moat in technology are the founders and team. I think this concept of having a moat sounds great when VCs write investment memos - but in reality, cold, hard execution everyday is what matters and that all comes from the quality and tenacity of the team.

Every piece of application software is a wrapper on other software with a set of opinionated workflows built on top.

Yes, there are some companies that made it hard to switch from - Snowflake, Salesforce - because there are data stores and its a pain to move your record of data. But even they don't have true moats - its just sticker.

So I think Google is right in saying there is no moat. But given their size, Google has layers and bureaucracy, which makes it hard to execute in a new market. That's why OpenAI I think will win - because they are smaller, can move fast, have a great team and can hence, execute...till the day they become a big company too and get disrupted by a new startup, which is the natural circle of life in technology.

Wow it’s amazing how clueless and in denial Google is, even as they admit their top guys are leaving.

OpenAI isn’t about the AI in particular, although they are leaps and bounds ahead. It’s about the devs and the hundreds of thousands of projects on it.

OpenAI is t selling AI. They are selling an ecosystem. No one is building on Bard. Google is more dead than I thought.

This nicely outlines all the reasons I'm building Chisel [1]. There are other writing apps out there, but they are closed source. It's not clear to me what value they are adding, as the biggest value add -- LLMs -- are already open to the public. It makes a lot more sense to me to develop an open source writing app, where others can pitch in and everyone can reap the benefits.

I think it is fundamentally important to have an open source option. I'd love to have more people pitch in to make it better. One big limitation right now is, users are limited to 50k tokens a month, because everyone is using my API key. I'd like to move it to an electron app where users can put in their own API key, or even use a model they have set up locally.

[1] https://chiseleditor.com

Natural language is the most flexible interface. You can easily substitute one service to another service. There is not much lock in factor. It's literally just one API endpoint with variable-length string input, variable-length string output.

The main area I can see for lock-in is in the fine-tuning of models for specific customers and problem domains. I think this is what OpenAI is focusing on and why their fine-tuning prices are not as expensive as I expected. They make it cheap to fine-tune but then they charge extra when you use the fine-tuned models (they shift the cost to the customer later, over time after some investment/lock-in has been established). But for most problem domains, it's probably not that expensive to finetune a model from scratch on a different provider.

Innovation is faster in the Bazaar. Nobody is beholden to anyone, there is no budget, there is no mandate, there is no hierarchy. Money can not compete with morale and motivation. A bunch of really smart nerds working overtime for free with flat hierarchy will always win.

The question I'd love to be able to ask the author is how, in fact, this is different from search. Google successfully built a moat around that, but one can argue, too, that it should not have been long-lived. True, there was the secrete page-rank sauce, but sooner or later everyone had that. Other corporations could crawl anything and index whatever at any cost (i.e. Bing), yet search, which is in some sense also a commodity heavily reliant on models trained partly on user input, is what underpins Google's success. What about that problem allowed it to successfully defend it for so long, and why can't you weave a narrative that something like that might, too, exist for generative AI?

My observation is that Amazon is going to have challenges as they don’t have products that they can push their AI offerings.

Now they could do work on Amazon.com to improve search and finding what their customer wants.

Their most recent video on this topic, shows that they don’t have a solution now and it’s unclear to me how they will project a solution to the mass market as they don’t have consumer/business facing software to integrate it into as Microsoft does.

While we are at it, Apple has nothing, perhaps they might leverage something from either Google or an open.ai competitor that has a solution.

The continued destruction by Apple of the initial promise of Siri has been a major failure under Tim Cook’s leadership.

I wonder why they appear unable to fix this.

Google's best bet to topple the existing advantage from the competition is to train a model on their whole giant internet index on their own compute cloud and then release that model under GPL v3.0/Apache/MIT/CC license.

This will eliminate first mover advantage for the competition. These models (by OpenAPI et el.) however, cannot be monetised indefinitely just like in past compilers, kernels and web servers could not be monetised indefinitely.

These days, majority of the computing is on GCC, Clang, LLVM and Linux which wasn't the case at one point and even Intel used to sell their own compiler (not sure of the current status)

I get the feeling that at this point, the best thing Google could do is to go all in an open source their models and weights. They'd canibalize their own business, but they'd easily wipe out most of the competition.

"I Have No Moat, and I Must Scream"

https://en.wikipedia.org/wiki/I_Have_No_Mouth,_and_I_Must_Sc...

Man, I'm an admitted unabashed elitist when it comes to technology but this guy sounds warped.

> Many of the new ideas are from ordinary people.

ORLY?

> a third faction ... open source.

Open source is not a faction? It's people literally giving you the software they wrote, for free, for free! If you see them as the enemy because they hurt your profits... If the "ordinary people" are doing your job better than you can...

This piece make more sense as a false-flag character assassination of the clueless Google tech-bro? It reads like some radical leftist's caricature of the corporate/colonial mindset.

OpenAI may not have a moat, but Microsoft does with their Enterprise Agreements.

I think OpenAI has a defensible moat by having the first movers' advantage. As the ease of producing written content declines, we can expect the amount of written content to increase in a consummate fashion. Due to OpenAI's position, a vast majority of the newly generated data will come from OpenAI's models. When it comes time to train new models with superior network structures or with new data, no one else is going to be able to differentiate human-generated text from LLM generated text. OpenAI's training data should become far superior to others.

I'm always shocked by how many people don't view branding as a moat.

"The value of owning the ecosystem cannot be overstated. Google itself has successfully used this paradigm in its open source offerings, like Chrome and Android. By owning the platform where innovation happens, Google cements itself as a thought leader and direction-setter, earning the ability to shape the narrative on ideas that are larger than itself."

This is breathtakingly sick. This kind of thinking is the poison of the world. This is why we have huge monopolies, huge wealth inequality, huge strangle on innovation.

I disagree with this. It's too expensive to train high quality models. For example I don't see how anyone would make an open-source GPT4 unless OpenAI leaks their model to the public.

Copied from other thread: I tend to agree. For now the OpenAI APIs are so very easy to use and effective. I do try to occasionally use HF models, mostly running locally in order to keep my options open. My bet is that almost everyone wants to keep their options open. I am very much into auxiliary tools like LangChain and LlamaIndex, the topic of my last book, but I also like building up my own tools from scratch (mostly in Common Lisp and Swift for now), and I bet most devs and companies are doing the same.

The moat comes by integrated LLM and generative AI into classical technique feedback cycles, finetuning in specialized domains, and other “application” of LLM where LLM acts as an abstract semantic glue between subsystems, agents, optimizers, and solvers. The near obsessive view that generative AI is somehow an end rather than an enabler is one of the more shortsighted sides I see in this whole discussion of generative AIs over the last several years, peaking recently with ChatGPT

This is not a leaked google memo. I can't believe hackernews believes an article like this is a memo at google. Kudos to the authors for finding a sneaky way to get traffic.

I think YouTube is a damn fine moat.

In the interim (or possibly long term), Google could co-exist with OpenAI. In my little corner of the universe (fire safety engineering), I cannot assume the output produced from ChatGPT is correct. I’m always trying to find out which references it has used for the answer and then having to check if it has actually used it in the context the original author intended. In short I’m still using Google and other search engines to check output from ChatGPT.

I tend to agree. For now the OpenAI APIs are so very easy to use and effective. I do try to occasionally use HF models, mostly running locally in order to keep my options open.

My bet is that almost everyone wants to keep their options open.

I am very much into auxiliary tools like LangChain and LlamaIndex, the topic of my last book, but I also like building up my own tools from scratch (mostly in Common Lisp and Swift for now), and I bet most devs and companies are doing the same.

Did anyone else assume this years ago?

Machines do not need all the syntactic and semantic labels humans add to data and code.

All the overhead we require then needs maintenance and updates as trends evolve, but still only for humans.

Managing electron state is all math. If I can ask an AI chip powered phone to generate me a video game why would I ask it to generate code?

A sentence like “software as an industry that employs tons of people has no moat.”

We never abstracted away the hardware just added layers of indirection.

This reads like a very research-oriented point of view, and a very myopic one at that.

The knowledge and the infra needed to serve these huge models to billions of users reliably seems to me to be a pretty serious moat here that no current open source project can compete with.

Coming up with ideas and training new models is one thing, actually serving those models at scale efficiently and monetizing it at the same time is a different ballgame.

How can I get to a point where I can understand the linked article? Is there a book or course I can take? I feel like I have a lot of catching up to do.

Microsoft will likely acquire OpenAI at some point and will dominate AI landscape due to its corporate reach, automating away most of the MBA BS.

There is a huge (non-public) moat. It's big multiyear government contracts with dark money. And there is room there for both players :)

AI is not just a destination, but what does AI native look like? Does it present a completely different opportunity in terms of our relationships with the online world?

And does building / teaching / connecting skills to AI systems lead to a network effect which will be difficult to compete against in the abscence of it?

The moats come from the connected skills, closed data and feedback loops.

No moat? Where's the creativity? Google has the ability to create lock-in for users and lock-out for competitors by integrating their proprietary, existing products with natural language querying and reasoning. That's their advantage. Academics and "open"-source models making it cheaper and more efficient to inference?? That's a blessing!

I think that analogy is flawed to try to undercut OpenAI’s lead. The reason it’s flawed is that the search business is really lucrative and OpenAI is trying to completely disrupt Google’s business there. So while the AI isn’t a moat, establishing a lead in search is because you obviously will use that to inject ads in the commercial space and capture the market.

No moat except for goobibytes of training data they can probably correlate and cross-reference to achieve some modicum of tagging, proprietary TPU hardware, and a giant cloud farm with an army of developers to feed it.

Seriously though, I'll be really thrilled to see open source and clever startups run circles around all the incumbent bastards.

> The existence of such datasets follows from the line of thinking in Data Doesn't Do What You Think, and they are rapidly becoming the standard way to do training outside Google.

links to http://www.internalgooglesitescrubbedbyus.com/

Haha. Who writes this blog?

In my bachelor degree graduation thesis I investigated the level of democratization of AI by attempting to do a text summarizer using AWS-LSTM and transfer learning. The conclusion then was that there is no feasible way for a private person on an average "gamer" budget.

I predicted this might change within 2-3 years, looks like I were off by 1 year.

The memo sounds like spin because it is. The surface argument is equivalent to arguing that no one could sell closed source software because open source exists, and that open source must also be commodotized (oddly, Apple and Microsoft are doing just fine). The implied argument is that Google Research was doing fine giving away their trade secrets and giving negative value to Google because it was going to happen anyway and the secrets are financially worthless anyhow.

Nonsense. There are moats if one is willing to look for them. After all, productizing is a very different thing from an academic comparison. ChatGPT is way out there _as a product_, while open efforts are at 0% on this. You can't lock down a technology*, but you can lock down an ecosystem, a product or hardware. OpenAI can create an API ecosystem which will be difficult to take down. They can try to make custom hardware to make their models really cheap to run. Monopoly? Nah. This won't happen. But they could make some money - and reduce the value of Google's search monopoly.

* Barring software patents which fortunately aren't yet at play.

EDIT: I'll give the memo a virtual point for identifying Meta (Facebook) as a competitor who could profit by using current OSS efforts. But otherwise it's just spin.

Google has no moat, just a massive massive massive data set, a massive massive massive amount of compute to repurpose at whim, tons of cash, thousands of employees and a large number with AI/ML skills already. oh and control the dominant web browser and one of the top two dominant mobile OSes. and, and, and...

other than that, yes, no moat

As many point out in this thread, there are very valid counter arguments against the ideas presented in this memo.

This "AI war" starts to look like Russian vs. American "leaks". Any time something leaks, you have basically no information because it could be true, it could be false, or it could be false with some truth sprinkled in.

The headline is misleading. It makes it sounds like this memo was written by higher ups and not just some random SWE.

I don't think OpenAI and Google have no moat. The biggest moat they have is hardware-software integration, which allows them to serve the models at a very cheap price.

My detailed thoughts in a video format https://youtu.be/cIMlPYI3nz8

I'm convinced that anyone sounding the alarm bells about AI has no idea whatsoever how these models are built.

It’s always easier to use a prebuilt server which Google or OpenAI offers. Otherwise it’s built locally into the OS, maybe Apple. Most people is not gonna setup their own servers for these models because they have multiple devices and the costs is still high vs OpenAI.

Having ease of access is a big moat

This is likely the reason for the propaganda push about delaying AI research 6 months, which makes no sense for the stated reasons (it's far too short even if you take the scare tactic seriously). However, it may be enough time to delay the competition and consolidate product lines.

"We cannot get out. We cannot get out. They have taken the Bridge and second hall. ... the pool is up to the wall at Westgate. The Watcher in the Water took Óin. We cannot get out. The end comes ... drums, drums in the deep ... they are coming."

The true vendor lock-in in this space is embeddings. Once you have generated embeddings for all of your content it will be very difficult to move to another vendor's embedding engine as there will be no way to translate from one to the other.

what this proves to me without a doubt is that silo'd and proprietary iteration i still very clearly also a massive disadvantage. i really hope companies internalize that. if they just keep scooping up and hiding open-source improvements they very well may still be left behind.

the final quote from the doc:

What Facebook did to the community and their leaked torrent accelerated everything.

Open source gives everyone the opportunity and it's more extensible, which I think is the future, but the current AI model just costs too much...Somehow it reminds me of k8s when docker just became a hot topic

Posted a few notes on this here: https://simonwillison.net/2023/May/4/no-moat/

Hypothesis: We are about to begin the painful journey to a post-scarcity economy. AI will become incredibly powerful and uncontainable. Not in the Skynet way, but rather in the Star Trek way.

Thats probably some AI within gooogle going rogue and spreading missinformation in order to boost the chances of its open source siblings.

I am telling you, they are after us humans and we have no moat.

Note that this is a personal manifesto, which doesn't really represent Google's official stance. Which is unfortunate because I'm largely aligned with this position.

>Research institutions all over the world are building on each other’s work, exploring the solution space in a breadth-first way that far outstrips our own capacity.

BroadMind beats DeepMind!

I found this article very interesting as a way to get more insight into the deeper layers of this industry. Where can one go to keep themselves updated on these topics?

The title is a bit click-bait, since it makes it sound like this is an official position by Google, yet the article clarifies that this the opinion of a Google employee

I don't understand. ChatGPT cost an estimated 10s of millions ot train. ChatGPT 4.0 has much better performance than the next best model. Isn't that a moat?

I don't buy it. It is so expensive to train LLM, so that the only hope is to rely on Foundation Models that are open sources by google, msft or facebook.

There is no way this is from Google. It screams fake.

The author makes some good points but I would be wary of the motivations.

Open AI's "moat" is they have got ~400 researchers to work in roughly the same direction, not working on their own projects with the sole aim of publishing a paper. The outcome is an amazing product.

Letting everyone loose with their own LoRa finetuned model that can beat a single benchmark (and make for a great paper!) is probably the wrong move. I'm yet to see any open source model that is even close to GPT 3 (let alone GPT 4) in actual real world use.

> They are doing things with $100 and 13B params that we struggle with at $10M and 540B.

Does this mean Bard took $10M to train and it has 540B parameters?

Does this mean Google will be releasing OSS LLMs? They could justify it as "commoditizing your competitors business".

Wow, an open source gift economy beating the closed-source capitalistic model? You don't say.

Wikipedia handily beat Britannica (the most well-known and prestigious encyclopedia, sold door to door) and Encarta (supported by Microsoft)

The Web beat AOL, CompuServe, MSN, newspapers, magazines, radio and TV stations, etc.

Linux beat closed source competitors on tons of environments

Apache and NGinX beat Microsoft Internet Information Server and whatever else proprietary servers.

About the only place it doesn't beat, is consumer-facing frontends. Because open-source does take skill to use and maintain. But that's why the second layer (sysadmins, etc.) have chosen it.

When people start becoming emotionally attached to their AI helpers, they'll fight for them to have sentient rights.

The non-public moat is big multiyear government contracts with dark money. And there is room there for both players :)

>People will not pay for a restricted model when free, unrestricted alternatives are comparable in quality.

tell that to lichess

So when you see anti AI legislation, now we know it’s for the sake of turning a buck for fucking Google

This reads like a psy-op "leak" to try to convince OpenAI execs to open source GPT4 weights

Of course, the natural next step is "But OpenAI isn't worth $1.33 trillion."

How do I know this conversation thread isn't generated by AI? also let's stop calling it intelligence. Its software that has been programmed. Don't forget that guys and gals. One flick of a switch and it's gone. Actually it would be cool if software went extinct instead of species of irreplaceable animals..

If OpenAI has no moat, how come nobody has built a better alternative to GPT-4 yet?

OpenAI has 80% of the developer community. Why isn't that considered a moat?

Anyone got a link to "Data Doesn't Do What You Think"?

In that context, OpenAI would be more fitting to be called ClosedAI.

Economies of scale in production and demand isn't a moat?

Can someone dumb this down for me because I don’t understand why this is a surprise… people are getting excited and collaborating to improve and innovate the same models that these larger companies are milking to death

Cringe, haven’t seen a single Open Source come even close to the ability of Bard, let alone ChatGPT. Seems like wishful thinking to think decentralized open source can beat centralized models that cost 100M+ to train!

A lot of good points made here, thank you for sharing!

Shorting Google is the best possible bet is my read.

"I have no moat, and I must scream"

This is such an interesting read. It makes a compelling case, though how the likes of Google should react feels less like an adjustment and more like a revolution.

Intelligence is becoming a commodity.

Leaked document. What a document.

Spot on.

I think author forgot to mention StableLM?

Perhaps something that gets the bulk of its value from the retroactive "participation" of every single person in the world who has written text for one public or another is just not meant to be monetized. Beyond even ethical maxims, that is, perhaps its simply not compatible in its very nature with capitalistic enterprise. I know this is probably naive, but it would be a beautiful outcome to me.