[Question] How common is it for one entity to have a 3+ year technological lead on its nearest competitor?

[Daniel Kokotajlo]

I’m writing a follow-up to my blog post on soft takeoff and DSA, and I am looking for good examples of tech companies or academic research projects that are ~3+ years ahead of their nearest competitors in the technology(ies) they are focusing on.

Exception: I’m not that interested in projects that are pursuing some niche technology, such that no one else wants to compete with them. Also: I’m especially interested in examples that are analogous to AGI in some way, e.g. because they deal with present-day AI or because they have a feedback loop effect.

Even better would be someone with expertise on the area being able to answer the title question directly. Best of all would be some solid statistics on the matter. Thanks in advance!

[Rob Bensinger]

I want to see more attempts to answer this question. Also related to another post I nominated: https://​​www.lesswrong.com/​​posts/​​PKy8NuNPknenkDY74/​​soft-takeoff-can-still-lead-to-decisive-strategic-advantage

[Jay Molstad]

In practice, it is often hard to tell the difference between “far ahead of the competition” and “off on a weird tangent”. You could make the case that NASA is 50 years ahead of the competition in manned lunar space travel, and at the time being ahead of the Soviets was a huge part of the appeal. But over time manned space travel was all but abandoned because there didn’t seem to be enough value in it. The USAF is in a similar position with stealth aircraft; competing countries never heavily invested in stealth and radar improvements have made it increasingly obsolete.

If you’re three years ahead of intelligent, well-funded competition, they may be waiting to see if the game is worth the ante.

[Andrew Rogers]

An example of a 10+(!) year technology lead is computational discrete topology. Every large-scale geospatial, graph, et al analysis system is based on it — you can’t build one without it — but there is virtually no literature on how it works and a practical expression of the theory is robustly non-obvious. The same few people continue research and design every kernel for companies/​governments. AGI and autonomous systems specifically drive much demand for this tech currently, since it is needed to reason about relationships/​behaviors in space-time at scale.

There is no company behind this tech currently but I’ve heard rumors of one being created. It could have a strong feedback loop, not just due to tech exclusivity but because a platform-level implementation would effectively provide a consensus model of physical reality for machines.

Tangentially, I am aware of AGI research programs working from first principles that have made impressive theoretical CS advances while completely under the radar. It is difficult to determine if any have 3+ year leads on any other program though since that assessment implies global visibility.

Distinguishing between a technological lead and ineffective competition is also important. An example is database engine technology. Some proprietary databases are orders of magnitude more efficient/​scalable than any open source comparable, which looks qualitative, but is widely recognized as a product of design quality rather than any technological lead. (see also: Google’s data infrastructure)

[George3d6]

Distinguishing between a technological lead and ineffective competition is also important. An example is database engine technology. Some proprietary databases are orders of magnitude more efficient/​scalable than any open source comparable, which looks qualitative, but is widely recognized as a product of design quality rather than any technological lead. (see also: Google’s data infrastructure)

Seems untrue to me, and I’ve benchmarked dozens of databases for dozens of problems.

In the column-store space (optimized for aggregate analytics… distributed execution of aggrgated queries, quick filtering based on ordering and data compression) Clickhouse is the best there is in my experience… I made that point 4 years ago, but now you can find plenty of other benchmarks for it. It’s used by many large scale search engines and advertisers except google, and among others, by CERN.

In wide column storage space, and more broadly in the “heavy filtering, large amounts of data space” cassandra (, facebook) and now Scylla seem to lead. I’ve never had to put dozens of petabytes in a database, but the few people that do need this seem to agree.

In the transactional space I haven’t seen anyone bring a significant gain over postgres and mariadb yet.

Kv store and in memory caching you have aerospike, rocksdb and stuff that’s based on tikv more recently… All slightly different trade-offs, all open source. I’m not even aware of proprietary products here to be honest.

Those 4 combined cover most use cases a db has.

So, not saying I’m convinced I’m correct, but could you provide some examples to back up your claims ? Name some names, or, ideally, provide some uecases/​domain where one could find benchmarks that demonstrate a proprietary database gas the upper hand.

[lincolnquirk]

Google immediately jumps to mind. The search result quality combined with the infrastructural investment required to execute on copying Google seems like it would take even an entity with no budget constraints more than 3 years, and that’s just search; Google also has maps, email, etc. Does your question assume any budget constraints? (I’ve been using DuckDuckGo as my default search engine for a few weeks and the results are obviously substantially worse than Google. And DDG has been trying pretty hard for over a decade, but with less than unlimited resources but still a lot.)

[ChristianKl]

DuckDuckGo isn’t a big player but there are other companies with massive investment into search like Baidu and Microsoft.

There are some articles about Bing winning in blind tests against Google like https://​​www.neowin.net/​​news/​​microsoft-new-blind-test-shows-users-still-prefer-bings-results-over-googles/​​ .

[Chris Hibbert]

Xerox PARC in the 70s and early 80s had a head start on inventing the personal computer revolution. PARC was well-funded, and they spent a fair amount of money building computers so each researcher had one on their desk. A little later, PCs were made in such quantity that it was no longer possible to stay ahead and have better equipment than businesses and consumers could buy. At PARC, they invented ethernet, the desktop metaphor, WYSIWYG editing, and much else. they didn’t invent the mouse, but they built platforms and environments that made really great use of it.

[lukeprog]

I’m very interested in this question, thanks for looking into it!

[Daniel Kokotajlo]

Yeah, me too. Well, I won’t exactly have done a full lit review by the time the blog post comes out… my post is mostly about other things. So don’t get your hopes up too high. A good idea for future work though… maybe we can put it on AI Impacts’ todo list.

[Jay Molstad]

It took about 4 years from America going public with the existence of nuclear weapons (Hiroshima) to the first Soviet nuke. If a technology is hugely impactful (so that money is no object) and proof of concept is public knowledge, three years is a really long time.

[ChristianKl]

Is there a reason you only want tech companies or academic research projects and not government sources like the NSA?

[Daniel Kokotajlo]

Yes, glad you asked. The reason is that in the comments of my original post, it turned out that Paul Christiano agrees that a big world government might be able to pull ahead of the rest of the world and achieve DSA even in a soft takeoff scenario. Since even Paul agrees on that point, I figured I’d make a note and then move on to talk about the cases we still disagree on: sub-governmental entities like corporations and academic research projects.

But I mean… if you have good examples of government sources I’d be happy to hear them!

[Pattern]

DSA?

[Vaniver]

Decisive Strategic Advantage.

[FactorialCode]

Has anyone ever mananged to replicate Google’s wavenet results? If not, that’s an AI example with a 3+ year tech advantage right there.

[shminux]

Almost anything Elon Musk touches: online payments, reusable rockets, electric (and some day soon autonomous) vehicles, solar panels, underground tunnels.

[PeterMcCluskey]

What’s the evidence for a technological lead in solar panels? That industry has looked pretty competitive recently.

PayPal’s lead in online payments never looked like it had much to do with technology.

[Daniel Kokotajlo]

I agree for rockets and electric vehicles (and online payments?), but I think the jury is still out on the rest. Give it a few more years and we’ll see whether those efforts bear fruit.

[Steven Byrnes]

Are you interested in hardware, software, or both? Hardware tends to have slower development cycles. Also hardware is more easily patentable, which complicates your question (if it takes a decade for the patent to expire, does that count as a decade-long technological lead?)

It’s entirely possible that the first real AGI algorithm will require (in practice) a custom ASIC to run it, maybe even using non-standard cleanroom fabrication processes or materials. OTOH, it’s also entirely possible that it will run on a normal GPU or FPGA. Hard to say, but important to think about...

[Daniel Kokotajlo]

I’m interested in both and even neither, but of course the closer the analogy to AGI the better.

Yeah, that’s a good fork of possibilities to think about. As for the issue of patents, insofar as someone has a lead because no one else is trying because patents, then that’s a disanalogy between their case and AGI because presumably patent laws won’t be powerful enough to prevent multiple projects from racing hard for AGI.