A guy from BP told me in 2013 that they forked iPython, replaced all iPython references with Palantir and tried to sell it to them for $500K p.a.
For me; back in the day (2010) they were less secretive about their technology which was essentially an ontological reasoner. This was pre the Big Data hype boom - and AFAIK Palantir has never been about Big Data. Ontological reasoners have problems that prevent them from scaling or generalizing so they generally fail. Due to a long long history of failing ontological systems have a very bad name. But they look good for guided demos and has a ton of academic backing so it's easy to sell - as long as you call it something else - which is what they did. So if you want to use ontologies a better open source alternative software is Protege. But for the problems Palantir targets I'd recommend using standard machine learning technology where all the good stuff is open sourced.
As an aside, Peter Thiel also helped found Quid. A start-up that ripped off the Gephi layout engine and charges people $20K p.a. a seat. They've since rebuilt it but like Palantir it's still not solving people's problem and they've evolved into a consulting firm.
That's especially hilarious given that approach's failures are what led to investment in machine learning in the first place. Such approaches tend to assume precise information, variables, and rules about the world. Most problems Palantir wants to address... the hard ones... are imprecise with hidden variables/relationships. The machine learning techniques did very well on those kind of mess problems. So, research shifted.
If Palantir is using ontologies for that stuff, then that would certainly be a sign for buyers to run. I still encourage academics to look into such approaches with probabilistic, simple methods in case any advances come up. Fuzzy logic was main one in my day. Just stumbled on a claim today a drone AI did human-level performance using that. Some corroboration for R&D in underdog solutions but not production apps. Haha.
Huh, so they are open about it now. They definitely were not when I talked to them. I guess enough time has passed that people have forgotten the hard won lessons of the past.
I worked on scaling and generalizing ontologies at university and had already switched to working with Big Data / ML at a big company when Palantir tried to recruit me. I talked to some of their senior engineers about their tech and made the point that their tech sounded just like ontologies. I tried to get them to admit what it was so I could be sure I was having an honest conversation with them. They flatly denied it and made it out like the whole thing was their great new idea. I was unimpressed.
I was still interested in working for them. Access to hard interesting problems can be hard to come by. In the end I couldn't take their legendary arrogance and insecurities - to me these are bright red flags of a toxic corporate culture. And they low balled me. I would have temporarily put up with the toxic culture for large piles of money.
"I was still interested in working for them. Access to hard interesting problems can be hard to come by. In the end I couldn't take their legendary arrogance and insecurities - to me these are bright red flags of a toxic corporate culture. And they low balled me. I would have temporarily put up with the toxic culture for large piles of money."
Smart decision. Far as ontologies, the Cyc project to create common sense in machines was my favorite at the time. Used ontological, knowledge base if my broken memory is accurate. I was and still am firmly convinced that finding an architecture suitable to solve that problem is a pre-requisite for the AI's we really want. Deep-learning is approximating it but closer to how brain does vision than common sense. Minsky noted at one point he could count number of researchers doing common sense on a single hand or so. That's a hard problem if you want one. Also unbelievably hard to get funded. (sighs)
Thanks. There is some interesting work being done with Deep Learning mixed with Bayesian Models and external databases of facts. The training apparently ungodly slow though. My work these days is visual so CNN works well enough. I'm not an authoritative expert in this field though so I'm taking 6 months off to study it intensely, after which I hope to be able to make some meaningful contributions.
There is nothing wrong with the ontologies approach. Palantir is not aiming for a fully automated approach. Palantir's core product is a an entity graph mostly used and manipulated by humans for analysis. Basically their software is only meant to augment human analysis.
That being said, from my conversations with them, they also have a traditional machine learning team for whenever that approach is needed for a product. But their core product is meant to only help analysis that is mostly done by humans.
There is usually something wrong with the ontologies approach as it rarely works. There is roughly two decades of evidence for this for anyone who cares to look. Five decades if you loosen the definition to include the family of logic and constraint programming - see AI Winter. There is nothing new about these ideas. It always looks and feels like it's going to work which is why humanity has persisted with it for so long and will likely continue to persist for some time to come.
There is a whole generation of better techniques that have come out of machine learning that totally eclipses ontologies and I know Palantir isn't using them. Their corporate culture isn't set up for fostering that kind of applied research.
No-one is advocating for a fully automated approaches. I don't know where that notion came from.
In my view is that Palantir is a consulting company that is pretending to be a tooling company. And their consultants are not worth the money they charge. Just one of many Silicon Valley based frauds.
Is this ontologies within the field of AI not working, or more generally?
Do you have references to any specific discussions on this?
Curious as I'm doing some work of my own (well outside AI) in which developing ontologies strikes me as useful, though I'd prefer not falling into any well-worn traps.
(My use is largely comping up with useful descriptive models of otherwise hairy concepts.)
What bazqux2 said is accurate. I'll go further to say that the kinds of work Palantir is involved in is mostly probabilistic. Especially intelligence work. So, use of models requiring certainty or straight logic in areas rife with uncertainty & degrees of truth seems set up to fail outside easy inferences. One can encode the logical stuff in probability models but harder to do reverse. Hence, their underlying tech should be probabilistic, fuzzy logic, or something similar for best results instead of just some results.
Far as ontologies in general, they have a mixed, track record. They take a lot of work to create. Then, they have to be mapped to real world inputs and outputs. One way they got applied is so called business rules engines or business process management. It's like a subset of ontology approaches of past. Here's a company that uses the real thing for enterprise software with Mercury language for execution part:
So, there's definitely companies using it for long periods of time for real-world, use cases. Palantir just seemed to be mixing it with hype and secrecy to maximize their sale price later. ;)
I meant generally they are generally not useful. Sometimes they are. It depends on the purpose and what you want to build and who it's for.
Given that you're building a descriptive model it would depend if you're working with facts or with probabilities. If it's facts then Ontologies should work fine, for probabilities I'd recommend Bayesian techniques.
The input for these are usually small. From the sounds of it you're generating the input yourself so you should be safe.
Again, Palantir is not an AI company. They are a data visualization and analytics company. So all your perfectly fine points about ontologies and AI winter are not relevant.
As far as I know, per annum is actually a part of the English language. I typically have seen it used in the context of pricing or finance which seems to be verified. [1] I don't disagree with you about p.a. less clear than /yr though.
I'd argue "per annum" or p.a. isn't a foreign term and just standard English. English is full of Latin words, transparent, opaque, foreign, those are all words of Latin origin.
> I'd argue "per annum" or p.a. isn't a foreign term and just standard English.
Did you notice that a fluent English speaker had to ask what "p.a." meant? Do you think that would have happened if it had been written "per year" instead?
Per annum, per centum, per mille, per capita, et cetera, et cetera.
These are Latin phrases, borrowed especially in British English as Great Britain was occupied by Latin speakers for nearly 400 years -- 43 CE through 410 CE. Latin continued to be the language of diplomacy, religion, philosophy, and science through the 18th and 19th century.
It is, for all practical intents, proper British English.
> borrowed especially in British English as Great Britain was occupied by Latin speakers for nearly 400 years -- 43 CE through 410 CE.
Can you explain how Latin loanwords were loaned into English during this period? In your explanation, please make use of the facts that (1) there were no English speakers in Great Britain before 410 CE, and moreover (2) there was, by definition, no such language as English until Anglo-Saxon migrations into Great Britain (around 450 CE) established a distinct West Germanic linguistic community on the island.
I think this is the most surprising thing, from my understanding, Palantir's user-facing tech is an ancient c.2007 era Java web start tool backed by a fairly ho-hum enterprise ontology system, some free-text search and fairly basic geospatial map server.
It sounds like a 3-6 month, 3 person project to replicate with modern tech. I think all of the pieces of a Palantir-like system exist with open/free alternatives, but nobody has just bothered to write the glue code to make it happen.
This makes me wonder at the ultimate utility of the particular shape of a Palantir system if nobody else is bothering to do it in quite the same way.
It's probably some combination of PostgreSQL (plus PostGIS) + Elastic Search + Neo4j for the storage tech, pick-your-web-framework for the server-side, D3.js + some mapping library for the front-end and have all the major pieces. A few months of glue code and CRUD writing and it would be done. I'd definitely welcome a quick-to-deploy open/free alternative.
The real money is in the integration, ETL, ontology consulting service bit and so anybody could really build a company around that stuff.
That matches my understanding of their tech stack as well. I also understand why people don't do an open sourced version - it doesn't pay nearly as well as having a job.
What I find interesting about this is that in effect Palantir is acting as a consulting company while pretending to be a software tooling company. This allows them to claim a higher earnings multiple to inflate value and extract more money out of VCs and offer a lower percentage of equity to employees. This is a very old trick. The problem is that consulting companies are much harder to scale than software companies and the inevitable disappointment will lead to a loss of equity.
There is good money in consulting (I am one) but it's hard to build a large consulting company when the 'tooling' companies can poach your talent away with cheap VC money and fairy tales about future piles of cash. It spoils the market. VC powered tooling companies masquerading as consulting companies are a real problem right now.
What's pretty interesting is that my understanding of their licensing scheme is that it is not cheap per seat per year. Something on the order of $10-12k. There was a pricing sheet that floated around a few months ago that was released through a FOIA request. So it sounds like they covered their initial development R&D costs through licensing, led with that tool into consulting.
My understanding is that many of the tooling companies who became consulting firms mostly jettisoned the tools along the way. Palantir has just let theirs stagnate.
You should check out Ab Inito, an even more secretive ETL company that predates Palantir and is also milking the DoD (it's an age old racket). They're $90K per seat per year and it's basically a crappy version of Jenkins. Apparently they have staff that search the internet and sue anyone who mentions them. Not even Palantir does that. So let's see how long this posts lasts :)
Wikipedia:
"Ab Initio maintains a high level of secrecy regarding their products. Some people working with their product, even those who work for organizations who use Ab Initio, operate under a non-disclosure agreement which prevents them from revealing Ab Initio technical information to the public."
> The real money is in the integration, ETL, ontology consulting service bit and so anybody could really build a company around that stuff.
Yes, and that stuff takes years to build. You're missing the whole point of Palantir when you leave this as essentially a footnote.
As a sidenote, your info is outdated. At least for the user-facing portion (which I saw - I obviously didn't get to see the graph/backend internals), it's all web-app-based now. At least from a superficial perspective, the UI seemed pretty slick.
> Yes, and that stuff takes years to build. You're missing the whole point of Palantir when you leave this as essentially a footnote.
Well, I think that's their intention. Pretend to be a software tool maker, but really be a services group. However, that radically changes the investment/return story. Because software, once you make a tool, you can sell it a hundred billion times for pretty much no additional cost and make pure profit after the first few thousand seats pays off the initial investment. But humans (services) don't return investment like this and most services work is very one-off by nature.
> As a sidenote, your info is outdated. At least for the user-facing portion (which I saw - I obviously didn't get to see the graph/backend internals), it's all web-app-based now.
This is important either way:
- if they're still pushing out the JWS client they used to have available as a demo for most of a decade (Op Tradestop), then the VC money isn't funding technology, it's funding sales and marketing expansion of the core services business
- if they're using web tech, there's plenty of very good, very mature web tech that's open/free for anybody to use, killing off their "secret sauce" sales lead-in and their customer stickiness. Here's their only known publicly facing web tech [1] I'll let readers decide if this is the output of a $20billion valuation company or something a couple guys over a weekend could cook up.
- if their back-end is just big-data scalable whatever (and their technologies page seems to indicate it is [2]), then they aren't offering any value to their customers there either
They have core tools, but obviously the majority of their code is bridging the core tools and their client's services.
Are you kidding me? If that extremely-slick dashboard (I hadn't seen it before) is representative of the quality of their UI for all their products (I actually doubt it is), as an enterprise/government company I am shocked they have not yet annihilated their competition. If it is representative, we now know why they have to spend money on lobbying - the only reason their competitors are alive is because of entrenched influence.
What's hilarious is you've broken down individual components of Palantir's offering and made the case that each individual component can be replaced by open-source or by a competitor. But that's literally missing the forest for the trees. It's the whole integrated package that matters. Nobody (or very few) is doing the whole package as well as they are.
> Nobody (or very few) is doing the whole package as well as they are.
If you read what I've said in this thread, I actually completely agree with you.
That map dashboard is like hundreds of similar dashboards I've seen elsewhere. I literally just saw a guy put something like that together in a couple weeks for an internal project where I work using mapbox and some d3.js bits. It even had a time slicer like this one and clickable map features.
So why isn't everybody as visible? I suspect getting a billion+/year in VC helps. But they've also managed to unify several key user-facing tools into a nice package and generalized it enough to support lots of different problem sets. Something that still seems beyond most companies.
From what I recall of their old public demo (Op Tradestop), their ontology data-input engine was pretty nice (if labor intensive) and it allowed for some easy to use queries against the enterpise graph. You can get pretty far locally with Visio, omnigraphle or yED, but they've managed to centralize the information input and retrieval in a nice way that seems easy enough to do by anybody who's seen it.
It might turn out that at scale it just doesn't provide enough value, or the labor intensive data entry parts make it not work well. Who knows?
Their competition isn't dashboards it's modelling.
Customers want dashboards and Palantir provides excellent dashboard tooling. Dashboards make people feel smart and feel like they are learning something. But dashboards are not as useful as people think and usually fail to return enough value to cover the cost. As Palantir is so expensive the bar is higher and often not met - hence the losing of customers.
Customers need data models but they don't know it yet. The charts used with models are usually not interesting if they exist at all. If you did show charts of the data models to the customer it often makes them feel dumb and out of control - few people like that. They're also dependent on you to interpret the models and customers don't like that either.
So given the choice of comforting lie or uncomfortable truth the vast majority will choose the lie. So if you're in the business of selling comforting lies don't be surprised when they fail to work.
A guy from BP told me in 2013 that they forked iPython, replaced all iPython references with Palantir and tried to sell it to them for $500K p.a.
For me; back in the day (2010) they were less secretive about their technology which was essentially an ontological reasoner. This was pre the Big Data hype boom - and AFAIK Palantir has never been about Big Data. Ontological reasoners have problems that prevent them from scaling or generalizing so they generally fail. Due to a long long history of failing ontological systems have a very bad name. But they look good for guided demos and has a ton of academic backing so it's easy to sell - as long as you call it something else - which is what they did. So if you want to use ontologies a better open source alternative software is Protege. But for the problems Palantir targets I'd recommend using standard machine learning technology where all the good stuff is open sourced.
As an aside, Peter Thiel also helped found Quid. A start-up that ripped off the Gephi layout engine and charges people $20K p.a. a seat. They've since rebuilt it but like Palantir it's still not solving people's problem and they've evolved into a consulting firm.