This is fun. I'd like to see the same idea but oriented for richer tokens instead of simpler tokens. If you want to spend less tokens, then spend the 'good' ones. So, instead of saying 'make good' you could say 'improve idiomatically' or something. Depends on one's needs. I try to imagine every single token as an opportunity to bend/expand/limit the geometries I have access to. Language is a beautiful modulator to apply to reality, so I'll wager applying it with pedantic finesse will bring finer outputs than brutish humphs of cavemen. But let's see the benchmarks!
I'm reminded by the caveman skill of the clipped writing style used in telegrams, and your post further reminded me of "standard" books of telegram abbreviations. Take a look at [0]; could we train models to use this kind of code and then decode it in the browser? These are "rich" tokens (they succinctly carry a lot of information).
I would point out that the default BPE tokenization vocabulary used by many models (cl100k_base) is already a pretty powerful shorthand. It has a lot of short tokens, sure. But then:
Token ID 73700 is the literal entire (space-prefixed) word " strawberry". (Which neatly explains the "strawberry problem.")
Token ID 27128 is " cryptocurrency". (And 41698 is " disappointment".)
Token ID 44078 is " UnsupportedOperationException"!
Token ID 58040 is 128 spaces in a row (and is the longest token in the vocabulary.)
You'd be surprised how well this vocabulary can compress English prose — especially prose interspersed with code!
For a while I was missing the ability one uses all the time in stable diffusion prompts of using parentheses and floats to emphasize weight to different parts of the prompt. The more I thought about how it would work in an LLM though, the more I realized it's just reinventing code syntax and you could just give a code snippet to the LLM prompt.
Hmm... this sounds a lot like the old RISC vs CISC argument all over again. RISC won because simplicity scales better and you can always define complex instructions in terms of simple ones. So while I would relish experiencing the timeline in which our computerized chums bootstrap into sentience through the judicious application of carefully selected and highly nuanced words, it's playing out the other way: LLMs doing a lot of 'thinking' using a small curated set of simple and orthogonal concepts.
RISC good. CISC bad. But CISC tribe sneaky — hide RISC inside. Look CISC outside, think RISC inside. Trick work long time.
Then ARM come. ARM very RISC. ARM go in phone. ARM go in tablet. ARM go everywhere. Apple make ARM chip, beat x86 with big club. Many impressed.
Now ARM take server too. x86 tribe scared.
RISC-V new baby RISC. Free for all. Many tribe use. Watch this one.
RISC win brain fight. x86 survive by lying. ARM win world.
Knowing the nature of a test ahead of time, building out your capabilities and tooling before entering the exam hall when your peers don't have that advantage, makes you a cheater.
Lots of people doing the same with extra steps (generating synthetic data from test questions with the LLM then training on it)
I wish we'd move past public test sets for LLM benchmarks: publish a plain english explanation of the tasks, allow questions and clarifications, and but never release a single question from the test set verbatim.
It made sense back when models needed to be finetuned on the task to even reliably answer. If we're saying this is the path to AGI we should be able to rely on the generalization of the model to get it right.
If a model was trained on <|begin_text|> <|end_text|> and you change the tokens passed to <|start_text|> <|end_text|>, it loses several 'IQ points' if it can even answer back at all anymore.
Synthetic data is fine. Synthetic data on very similar questions generated based on the description is typically fine. But once the shape of what you're training on gets too close to the actual holdout questions, you're getting an uplift that's not realistic for unseen tasks.
There is indeed a painful dissonance here. I like this new world, but feel sorrow for the loss of something. I try to remember how empowering AI is. It is already allowing millions of people to finally use the devices they've been sitting in front of all these years. No longer do they have to feel constrained by software creators who have made choices for them. Now it is their tool through-and-through, and they can construct software on-the-fly to match their needs precisely. They have been buying computers with both hands tied behind their backs. Now they are in control.
I disagree. There's definitely _some_ who will use these tools to build systems for themselves. But do you think the chef who's been pulling insane hours in the restaurant wants to come home and build his own software? Or the teacher who just had to deal with an annoying classroom all day?
People want software that just works, they'll pay for it, they don't want to use their computers to build their own software. That idea is just software and computer geeks (said affectionately) projecting their own desires on a larger community.
Does it have to be mutually exclusive? On-the-fly software does not destroy software. Gatekeeping software creation does not mean shoving the existing creators out, it just means creating a larger space that others can occupy, like when 'real' programmers had to slowly permit 'script kiddies' into their spaces. All feels a bit 'old guard' vs 'new guard'.
Not mutually exclusive, but I thought your initial post painted an overly rosy picture with the sentence "[..] allowing millions of people to finally use the devices they've been sitting in front of all these years".
I don't think it's happening at this scale. I'll admit I have no real data to back that up, it's just a hunch really. But I find it hard to believe that those people whom previously weren't interested in building software are now suddenly interested to build stuff with an LLM. I'm sure _some_ people are doing this, and then they either hit roadblocks and quit or stick with it an learn actual software engineering.
Looking at my non-tech bubble of friends and family, I don't see anyone actually doing that. I think it's a vocal minority that is doing this. That's just anecdata of course.
I think using GPT et.al. to create a bespoke tool to do what you need is giving the average home user too much credit. What I see more of is just using the prompt in the place of software to create an outcome. "Transcribe this recording", "give me a synopsis of the Godfather films", "How can I wow my girlfriend?". The fraction of home users who are using this to create software is likely highly limited to people with no skills trying to make apps to sell, which is not a tool to help them with something else. Even the software devs I know are using tools made for them, not making their own Claude Code or Cursor.
Right now, the greenfield is in how you use these tools. Making a bespoke specialized tool for yourself, or automating onboarding or CICD setups with simple commands or building bridges between "gatekept" existing software and agents are ripe for growth.
I get that we should see this as a good thing, but I see it as entering the last act of a play. Thousands of people are doing these things and coming up with uses for the tools around the clock. Novel uses for the technology will all be exhausted in the next couple of years and there will be less room for innovation than there was before LLMs.
We’re not there yet, but that chef or that teacher definitely would want an AI voice assistant as good as the computer in Star Trek. Maybe to achieve that, a language model builds software entirely autonomously and runs it to carry out the user’s command. Or maybe they want the computer to build them software that they can then use to do their own work more efficiently.
> We’re not there yet, but that chef or that teacher definitely would want an AI voice assistant as good as the computer in Star Trek.
Since you brought up Star Trek, a good analogue for AI would be the holodeck. Given the appropriate prompts, it produces amazing scenery and even immersive fantasy narratives.
But occasionally, it goes haywire, the safeties no longer work, and the characters from your fictional adventure try to kill you.
That really is a great thing. I do wonder at the segment of the population that from the 70s to today that sculpted their brain to think like a von Neumann machine. What will be lost when the last of us passes. It will likely be viewed as an oddity by future generations and people will try to replicate it as a hobby. But many of us began shortly after learning a primary human language, and that degree of specialization isn't something a hobby can reproduce.
I don't even know what this means. Were programming languages and compilers unavailable? I think this overstates the predicament that users were in quite a lot. In fact, I don't think any users even asked or wanted this really other than maybe very simple musings of wanting to say "computer, enhance" or something like that. I'm not sure how vibe-coding has now given the average user something that they know will not have their hands tied behind their backs. Most users will never vibecode anything and most of the non-technical people that will, will try it once or twice as a novelty or an attempt to solve a problem and then give up.
Now, what you might end up doing is removing the need for them to even have a computer to begin with. Someone without a job doesn't need business productivity software that much.
Can anyone familiar with the technology help disillusion naive people like me as to why on earth palantir needs to exist? It feels like a big pile of nothing. But tbf that's how I feel about Salesforce and Jira too. Big fat database schemas with big fat CRUD atop and layers of snazzy sparklines to make PMs and clients feel nurtured and fuzzy that they've done something material.
Like how Tableau is a great UI for grammar of graphics, Palantir is a great UI for ontological expert systems. Technically you could do everything without it but organizations and especially government typically don’t cultivate that level of expertise in their staff.
In my view expert systems typically failed because the organizations would degrade bureaucratically faster than any expert system could accommodate. With AI there isn’t a pre-requisite need for organizational expertise so the tooling will still work in largely dysfunctional orgs which is a property that did not previously exist. With the help of AI people who don’t understand ontologies can still successfully build one.
Separately it is my opinion that Palantir is a CIA cut-out for the Peter Thiel faction. So paying Palantir is like paying tribute to that particular faction. Similar to how other large military purchases are less about the military hardware and more of a client state subscription to ‘align interests’ such that the US is more likely to act in the donor countries interest.
> Similar to how other large military purchases are less about the military hardware and more of a client state subscription to ‘align interests’ such that the US is more likely to act in the donor countries interest.
I have a feeling this is no longer a viable model. If "subscribers" get threatened every other day, they will be looking for alternatives.
So long as not subscribing is worse than subscribing countries will still do it. Even if it not in the interest of the country the decision makers can and do still get kickbacks / speaking engagements.
It’s interesting to read of the ineffectiveness of influence the gulf states thought they had, though I think that speaks more to the relative cost effectiveness of tributes versus blackmail. These states don’t have the security apparatus to both blackmail US politicians and prevent others from blackmailing those same politicians. This second part is essential as it is what maintains the relative advantage.
I do think they will be less enthusiastic subscribers in the future, and perhaps even shop around for more cost effective approaches. Modi in India is intentionally creating an Indian diaspora as one example and I believe he is bribing politicians to help make this happen.
> read of the ineffectiveness of influence the gulf states thought they had
The primary players in the Gulf - Saudi and the UAE - have been aligned with the ongoing Iran strikes.
KSA's Mohammad Bin Salman has been lobbying Trump to strike Iran [0], just like his predecessor King Abdullah was doing [1]. Similarly, the UAE has an ongoing land dispute with Iran [2].
2. The operationalization of the Iran-Central Asia-China railway in 2025 [0], which allows China to bypass Malacca
3. Iran's relative weakness following the collapse of the Assad regime, the death of much od Hezbollah's leadership, and the Houthis comparative weakness
4. Continued anger amongst policymakers in the Gulf, Israel, and the US that Iran-backed Hamas launched the 10/7 attack barely 3 weeks after the US+EU launched the IMEC project and were about to loop Saudi Arabia into the Abraham Accords [1]
I was hoping to hear the case made as to why Israel was not the primary reason but instead you seem to have chosen to elide it altogether. It seems to be a conspicuous omission especially when both the US and Israeli admin have repeatedly made the case that Israel was the primary reason.
Primary reason is because Israel and American zionists (mostly evangelical christians) lobby for it. The KSA and friends also lobbying for it is just icing on the cake for American politicians.
China also views the EU as a junior partner [0], is running an ongoing disinfo campaign against the industrial exports of an EU member state [1], and has doubled down on it's support for Russia [2] in Ukraine in return for Russia backing China's claim on Taiwan [3].
And the EU is uninterested in building domestic capacity for most critical technologies.
Heck, last week [4] the EU excluded AI, Quantum, Semiconductors, and other technologies from the Industrial Accelerator Act (aka the "Made in EU" act) in order to concentrate on automotive and "net-zero" technologies.
Given that Chinese technology imports are already under the radar in the EU due to the Ukraine war, this is basically the EU creating a carveout for the US.
Even the major European Telecom and Space companies like Eutelsat, Deutsche Telekom, and Telefónica bluntly stated that they view the EU's digital sovereignity strategy as dead in the water [5] in it's current form.
Edit: can't reply
> They/we will go to domestic producers as much as possible, then China, then US, then rest of the world in that order. At least that would make a rational approach since (for now) unique things like f-35 can become an expensive paperweight on a whim of a lonely sick man. You can't build any sort of defense strategy on that, can you
But as I clearly showed, the EU is doing otherwise.
And the EU cannot work with China as long as China backs Russia and undermines European industrial exports.
All the rhetoric about digital sovereignity and domestic capacity has been just that - rhetoric.
They/we will go to domestic producers as much as possible, then China, then US, then rest of the world in that order. At least that would make a rational approach since (for now) unique things like f-35 can become an expensive paperweight on a whim of a lonely sick man. You can't build any sort of defense strategy on that, can you.
> And the EU cannot work with China as long as China backs Russia and undermines European industrial exports.
I mean, that is not that huge a difference compared to the USA (lifting sanctions against Russia, no tariffs there either, but plenty tariffs for "allies"; threatening NATO members in several ways; taking over Russia's "peace" plans for Ukraine 1:1 and putting the pressure solely on Ukraine; (I could go on for pages)).
I am not sure Americans really understand how much trust is already gone.
> that is not that huge a difference compared to the USA
It is for the EU.
The EU dislikes the current deprioritization of the Ukraine Conflict by the US, but also recognizes that the PRC is directly providing material support and subsidizing Russia's military industrial complex [0]. That is the red line for much of the EU.
Similarly, for the PRC it's continued support of Russia in their war in Ukraine is also a non-negotiatable [1], and the CCP's foreign mouthpieces continue to reiterate that "the mainstay of EU foreign policy — supporting Ukraine in a conflict to defeat Russia — has turned into a quagmire of sunk costs with little hope of success" [2].
> I am not sure Americans really understand how much trust is already gone
We know. And we don't care.
As long as the EU views Ukraine's territorial integrity as non-negotiable and a large portion of EU states view Russia as the primary national security threat, the US will remain the less bad option than the PRC or Russia.
Both the US and China are aligned in that we view the EU as a junior party that can be pressured [3].
If the EU views Russia as a threat, it will have to accept American vassalage becuase the PRC will continue to back Russia [1].
If the EU views America as a threat, it will have to accept Chinese vassalage, give up Ukraine, and accept Russia as the primary European military power.
Based on the carveouts within the Industrial Acceleration Act, the EU has chosen American vassalage.
Very bold words. I am not even convinced the USA will stay relevant on the world stage, in the long run. Cutting ties hurts, but the process is underway. Also, "vassalage" is a bold word, if the US cannot make the EU give up Greenland or come running to help them in the Strait of Hormuz (there are also other examples). It is almost as if European politicians are playing it smart.
And my question is - are you fine sacrificing Ukraine in return for a Russian and Chinese military umbrella? This is the hard requirement for China to engage with the EU [0].
The answer in Poland, the Baltics, Czechia, and Finland is NO and that Russia is worse and that Ukraine must be supported, and will back the US no matter how transactional we become.
The answer in Hungary, Slovakia, and Belgium [1] is YES and that sacrificing Ukraine for Russia is acceptable.
> if the Chinese support for Russia can be broken, by economical incentive...
China is not interested in breaking with Russia.
Russia helps China put pressure on Japan [0], helps China put pressure on South Korea [1], allows China to expand it's influence in Central Asia [2], acts as a backchannel for China-India diplomatic normalization [3], gives China the ability to access ONG without dealing with Hormuz or Malaccas [4], and allows China to run the Chongqing-Xinjiang-Europe railway [5] which continues to supply Europe with no sanctions despite the ongoing war in Ukraine.
On the other hand, the EU is tariffing Chinese goods [6]; signing FTAs with Chinese rivals like India [7], Japan [8], and South Korea [9]; and signing defense pacts with Japan [10], South Korea [11], and India [12] while allowing them to participate in ReArm Europe 2030.
Additionally, China-EU trade only represents a little over 10% of all Chinese trade [13], and is easily replaceable with expanded trade with ASEAN, Japan, South Korea, and India.
China views Russia the same way America views the EU - a weak junior partner who can be bullied. The US is somewhat trying to pull Russia to our side, and China is somewhat trying to pull the EU to their side, but the reality is both the US and China view the EU and Russia as junior partners.
> the Chinese support for Russia can be broken, by ... threat
What threat can the EU give to China? Chinese foreign policy already views the EU as sanctimonious [14], weak [15], and declining [16].
> over short or long the EU needs to build its own military to a strength it can at least work as a strong deterrence for aggressors
Yep.
But that will takes decades, which is why the US and China can both bully the EU with complete impunity today.
Heck, both China [17] and the US under Trump [18] are supporting Viktor Orban because he is a great Trojan horse.
Whenever either the US or China feels the EU is leaning towards one at the expense of the other, they then start breaking EU institutions as a result.
You have a very static view there. In my estimation the US is on the way down, at least economically/financially. Their internal stability is already somewhat broken. It will be hard to continue to project power without real allies and the internal issues they have and will have.
So, if the EU is so much inferior, why did they not buckle in the Greenland issue, but Trump was called back by his puppeteers? Why can they say "no" to supporting the US and Israel against Iran? And if they wanted the EU leaders could go further and match tariffs one by one and nothing serious would happen. The picture you are painting does not account for the facts. The relationship is not between equals but lord and vassal is also not a good fit.
I am not sure about the trade figures in your link [13]. It does not open for me. I seem to recall a significantly higher export volume going to Europe. But anyway, China is going to have their own internal issues with an aging populace, an end to strong economical growth and ever-growing social inequality. They are also too rational (compared to the US) to disrupt good business by mutual bullying (at least overtly and systematically).
What threat can the EU give to China? Chinese foreign policy already views the EU as sanctimonious [14], weak [15], and declining [16].
15 is an opinion piece written by a failed politician from Kyrgyzstan for China Daily and 16 is another opinion piece written by a right-wing politician from Slovakia. Neither represent Chinese opinions. 14 doesn't open for me.
The message matters less than the messenger - China Daily is the English language newspaper of the CCP's Propaganda Department and the Global Times is the English language newspaper of the CCP's Central Committee.
The fact that the mouthpieces of two of the CCP's most important committees are constantly publishing content that is dismissive of the EU highlights how China's leadership actually views Europe.
Europeans really need to get it in their head that both the US and China look at the EU dismissively and as a junior partner. Neither the US nor China is interested in a relationship of equals with the EU.
The UK NHS is one of the biggest employers in the world. It absolutely could choose to hire and cultivate that level of expertise but then how would senior management retire into Palantir sinecures?
(It actually has quite a few expert staff who are not delighted with the tools they have been given but they don't have the lobbying power of Palantir and the cluster of consulting firms around it)
Plenty of companies don't "need to exist". A company exists because someone decided to start it (usually to make some money) and lasts until someone decides to end it (usually when it stops making money).
If you're asking why Palantir (and Salesforce, Jira, etc) continue to make money despite not having any novel or complex technologies, my experience has been that these are not prerequisites for solving the vast majority of business problems. Usually network effects, customer relationships, brand identity, user interface, inertia, etc are all more important than the technology.
It is not always easy for a technologist to admit, but companies whose ongoing success is primarily due to some sort of (non-UX) technological superiority are the exception rather than the rule.
> This discounts the value of user experience, which people will pay a premium for.
The people making purchasing decisions at this level aren't the ones using it and don't care one whit about UX.
That isn't to say that it isn't valuable, but it's basically a non-factor. The technology itself is a non-factor. Everything is about connections, buzz words and pretty slide decks.
They literally do, since the people making purchasing decisions are usually the ones that ranked up through a system they used and know the intricacies of, including all the pain points.
As someone who used to teach UX grad courses, I'm happy you feel that way!
But I'm unsure why you feel that my response pointing out that a product's user interface is typically a more important factor in success than the product's underlying technologies was discounting the value of user experience?
> Good design IS technological superiority.
Hmm, I was attempting to respond to someone who wrote "It feels like a big pile of nothing... Big fat database schemas with big fat CRUD atop and layers of snazzy sparklines" which seemed to dramatically undervalue good schemas, CRUD implementations, or sparklines as "nothing". So to contrast those I used "technical superiority" as a catchall for the sort of challenging technical implementations that some developers lionize. Does that make sense? Is there a different term you'd suggest for that? For now I've changed to "(non-UX) technological superiority".
“Palantir is a tech platform that consumes data from their clients in return for providing high level data-driven insights. They assign FDEs (or consultants) to really learn the details of a customers data. Foundry allows them to get single pane view of the data in an org and they actually have both the tech and engineering skills to do the dirty data cleaning jobs.
For an extravagant fee, you give them your data, they clean it for you, and then those same FDEs can tell you interesting things that you should have known, had you actually done proper data architecture in the first place.”
They’re also missing the tidbit that, like any other consultancy, they provide a means for laundering a conclusion that middle management has already come to, confirmation bias be damned. Unsurprising that they’re also useful for parallel construction for LEOs.
The first half is true. They bring in their FDEs to clean and organize your data.
But the difference in what they leave behind is what separates them from classic consultancies and pure tech companies.
They don't leave behind "insights." They leave behind a suite of operational (ie have write capabilities not just dashboards) applications that are "custom" built to actually solve those insights. I put custom in quotes because while the applications are usually bespoke to your company, they are built in Palantir's app-building product Workshop, which significantly lowers the cost of building these custom apps.
So in the end, your company's processes are improved because your employees are using the apps that the FDE's built.
This is distinct from traditional consultancies because those will only leave behind the insights. Also distinct from most SaaS because those have a one-size-fits all approach, so you wind up having to change your company to fit the design of the application, where as Palantir builds its applications to fit your company.
> Contrary to some media reports, we are not a surveillance company. We do not sell personal data of any kind. We don’t provide data-mining as a service.
The commercial product, Foundry, is very well documented and an extensive Data Platform that allows to build data pipelines (similar to Databricks) and build low code / no code applications on top. If you master it, its incredibly powerful but complex
It's not rocket science. Those particular database schemas, together with those particular CRUD layers, do something useful, and neither building nor maintaining those applications is part of the core business for most companies, so buying prebuilt from somebody else, and letting them maintain it for you, makes perfect business sense.
I don't know how you think a b2b company could run sales without a CRM like Salesforce.
To give your question a generous interpretation, Salesforce is more valuable than Apptio or your home grown CRM because it already has all the features any sales org needs, and all the fragmented sales and marketing tooling are already integrated with it.
And Sales is a very expensive and also high ROI activity. You don't want your sales team hung up trying to figure out how to get the random CRM to do something. You're not looking to cut costs in this area, you're looking to enhance the overall productivity of the org. Sales tooling overall is very expensive for this reason, any marginal edge is worth a lot.
It's also worth noting that a big value of things like Salesforce is that it lets management check up on what people are doing, because as much as HN doesn't like to admit it, people are often not very careful or diligent, and you need to perform supervision on the vast majority of people to improve their performance.
Jira is similar, in that eng is very expensive, and its probably better than what these companies were doing beforehand, even if it is suboptimal.
It's true, literally no b2b sales companies existed before Salesforce. We must all continue to pay for Salesforce and support its workflows for now until the endless future, lest b2b sales vanish again.
Because it's hard for the government[1] to build computer systems.
Government salaries are pretty low compared to dev salaries. If the government wants to hire devs and pay them as much as private industry does, they'd have to pay them much more than what their superriors (and their superriors' superriors) make, which would destroy workplace morale. They could raise everyone's salaries, but that's deeply unpopular, as a large part of the population view all high-level government functionaries as crooks by definition.
The way you get around that is by using contractors. Contractors let you hide the cost of software development. Instead of paying $150k to a software developer (which is probably more than the director makes), you pay $10m to a company, not unusual when you also hire companies to build you planes and bridges. How that company allocates that 10m and how much they pay their engineers is no longer your concern, and no longer an embarrassment to your hierarchy and salaries.
However, writing contracts for software is hard, for the same reason waterfall is hard. You just don't really know what the requirements are before the project starts, and in a traditional RFP process, you can't accurately model what requirements are the costliest and should perhaps be reconsidered. This means contracted government projects usually turn into an exercise in checkbox-checking and terrible, unusable UIs which technically fulfill the acceptance criteria, and therefore have to be accepted.
Palantir has somehow managed to actually collaborate with the government, sending forward-deployed engineers to figure out what their actual needs are, and then writing software which fulfills exactly those needs, bringing techniques which modern tech companies have learned along the way. I don't actually know how they managed to circumvent the RFP process well enough to do this.
[1] "The government" here can apply to any government you like, not necessarily the US government.
Palantir’s product is light years ahead of anything any government IT project has ever, and in my opinion can ever, deliver. They’re not even in the same league.
Counter to that I’ve seen a £37m contract for a form on gov uk with absolutely no change in process, just going from a letter received to a online form
GDS is amazing. However, unless we double/triple the GDS salary grades, it'll inevitably be hollowed out. From what I heard, that might've already happened.
Look for yourself, GDS is hiring a "Lead Technical Architect" for £67,126–£91,453 https://gds.blog.gov.uk/jobs/ . FAANG (and Palantir) pays up to triple that. How can GDS compete for talent?
But how many people can you attract, and how quickly can they get the stuff done? There are a lot of sacrifices you have to make working for the gov that not everyone will make.
Horseshit, mate ... basically just pumped up database software aided and abetted by "consultants" parachuted into the client org ... like the industry has been doing since the 80's ...
Edit: I found the following on Glassdoor and, while I don't know the poster personally, it pretty much sums it up:
"If you are in Business Development (BD) - i.e. Delta or Echo - this job will be your life. They deliberately underhire - they claim it's to maintain the culture, but really it's to squeeze every ounce of productivity out of you. You are thrown into chaotic situations with no way out but to "chew glass and excrete product". Don't let the flat heirarchy and encouragement of confrontation / open debate deceive you. Karp has majority founder shares and calls the shots. The company is a dictatorship, not a democracy. Resourcing is a black box. If you are a U.S person without a clearance, you will be bait-and-switched into defense even if you thought you could avoid it. With clearance, you'll end up on something much worse. Trust your gut - the company's leadership are not wise, nuanced philosophers - they are spineless, shifty edgelords with no ethical red lines. As a FDE, you will spend half your time working around stupid limitations in the platform you could not foresee when making grand promises to the customer. Foundry is not a cutting edge product, just like Microsoft Suite is not a cutting edge product. Its just too broad for any other company to easily copy it. Palantir just brought middle-of-the-road Silicon valley tech to old-school government, slapped some AI integration onto it and shrouded it in a veil of mystery to make it seem cool and mysterious and appeal to retail investors."
It's 100% laziness on the side of procurement, aided by some good marketing and a complete lack of guardrails. Exactly the same mindset that has led to every European government now being tied to US big tech.
In government you have to deliver, most of the time the mode of delivery is boring, small, conservative, and disjointed from other government groups because large efforts of work attract big budgets, oversight and doubt.
Consultants are magic, because they come with no baggage and promise the world. They take you hostage with sunk cost fallacy and then after years they deliver something.
At the end you're so tired you think that what they did was beyond your government agency and the cycle continues.
Somebody needs to wrap up open source AI/ML and sell it to governments / defense, and do the integration... (e.g. open source Python face recognition libraries, openCV, YOLO object detection, etc. and more recently LLMs.)
Why does Palantir specifically need to exist? To funnel those juicy government budgets into shareholders' pockets.
Why does anyone bother to use them? Because they have convincing marketing (which may or may not include buttering government palms with, um, "incentives" ...)
Occam's razor: It's a big pile of "list of things being handled by an outside entity so I neither have to think about it, nor hire for them."
If Palantir wasn’t highly effective at aggregating data no one would care about the. They are considered a threat to privacy and freedom because they are a good product
Is Palantir actually that good? Or did all the governments just have enough brain drain they can't think of an alternative?
Like if their product was so good why isn't Amazon using it? Like their case studies all seem to be pre-internet companies that probably never developed a computer competency.
If I bring a themostat back into the past all the peasants are going to think it's black magic. If I show it off as a college project I'm not getting a passing grade.
That's part of it, but not the whole story. If Palantir were a book, explaining how to implement data aggregation systems effectively, people wouldn't be so wary of it. (Critics would still criticise that data aggregation was performed in the first place, of course, but there wouldn't be the additional "and it's Palantir".)
Yep :/ There are just no good heuristics left for quality clothing. It's horrible. One thing I do genuinely have good experience with is Japanese denim. But that's about it.
My approach is that, "you may as well" hammer Claude and get it to brute-force-investigate your codebase; worst case, you learn nothing and get a bunch of false-positive nonsense. Best case, you get new visibility into issues. Of _course_ you should be doing your own in-depth audits, but the plain fact is that people do not have time, or do not care sufficiently. But you can set up a battery of agents to do this work for you. So.. why not?
> I work in software and for single line I write I read hundredths of them.
I'm not sure whether this should humble or confuse me. I am definitely WAY heavier on the write-side of this equation. I love programming. And writing. I love them both so much that I wrote a book about programming. But I don't like reading other peoples' code. Nor reading generally. I can't read faster than I can talk. I envy those who can. So, reading code has always been a pain. That said, I love little clever golf-y code, nuggets of perl or bitwise magic. But whole reams of code? Hundreds upon hundreds of lines? Gosh no. But I respect anyone who has that patience. FWIW I find that one can still gain incredibly rich understanding without having to read too heavily by finding the implied contracts/interfaces and then writing up a bunch of assertions to see if you're right, TDD style.
Most of the software engineers out there do the support, augmenting source code behemoths the least possible way to achieve desired outcome. I believe that more than 90% of software development was support roles as early as 2K or so.
Not that I had an opportunity to write new code, but most of my work through my experience was either to fix bugs or to add new functionality to an existing system with as little code as possible. Both goals mean reuse and understanding of the existing code. For both "reuse" and "understanding" you have to thoroughly read existing code a dozen or so times over.
Tests (in TDD) can show you presence of bugs, not the absence of them. For the absence of bugs one has to thoroughly know problem domain and source code solving the problems.
reply