I think this is a great direction -- for some kinds of users. And this makes me wonder if the 'vs' framing is misleading.
Yes, I think it's a mistake that many organizations are cramming LLMs inside of automated pipelines where the extreme generality/flexibility of the model is at odds with the fact that you're using it for a very specific task that gets repeated over and over, and needs a very specific structured output to be successful. But specifying your task carefully (as well as deciding what counts as your input state representation etc) seems like a form of programming. Something (a person or a model working in a relatively unrestricted way) will need to produce a configuration/specification for this system.
So rather than Jev vs Claude I imagine that using Claude/ChatGPT/whatever interactively to define / refine your Jev config which then runs in prod might be the happy combination?
From what I can see, online media has finally taken over. The horrendous coverage from the 2024 election may have finally been the last straw for many people.
People seem to be focusing more on individual reporters, subscribing directly to good journalists who have gone out on their own. We are also seeing a thing that pleases me ver much: Reporters focusing on the subjects they are experts in and not doing a jack of all trades master of none. Its made me accept that I wont get top tier news in every subject so I have to pick what really matters to me(Tech, Food, entertainment, geopolitics of specific regions of the world etc.).
On the low end though its more bleak. TikTok slop, Youtuber "commentators" etc.
I think the slow down / pause vs acceleration framing is broken b/c it assumes there's basically only one forward direction which is to big and general models. Why should that be the case?
What if instead of building one big scary AI god, we built an ecosystem of extremely effective, efficient and predictable, reliable tools?
Tools that are specialized for particular purposes could potentially be safer, more efficient, more reliable and effective, and even more profitable for their makers. We currently put science, engineering, general question answering, paper-writing, "smart search", and chatting with a fantasy character all into the same token-prediction platform. We struggle with hallucinations when trying to make fact-based decisions using the same tool that our neighbor might be using for creating writing prompts (or at least was trained in part on fanfic).
We pushed people to integrate giant generalist models into their workflows, tie in with tools, etc. And then a coworker can have a bot write slack updates that summarize progress on tickets from your teams project, at the cost of giving an untrustworthy agent access to a bunch of internal systems, and a small risk that it will do something crazy. And when it works, that's worth _something_ but probably our team would pay for that ability at a different price point than the models we use to build products.
If I could do programming with a faster, specialized model that was trained not just to complete program text-tokens but on tuples of program text, compiler IR, program traces, etc, so it had a deep and explicit understanding of how the program would build and execute, and where the model was closely integrated with the language toolchain, I might be more productive and be willing to pay more than for a generalist model. I don't _need_ my coding model to be able to role play, or be able to potentially engineer a super-pathogen, and if the size and latency of my coding model could be lowered by entirely cutting out that possibility, everyone can be better off.
Maybe bioscience applications are super valuable, in which case someone should build them. But does the model that suggests CRISPR edits to a model bacterium need to also know about computer security, and should it have scifi novels in its training data? Does it even need to be capable of producing unconstrained tokens, or should it be limited to producing in some relevant domain-specific language? And would labs be willing to pay more for a model that was specialized, and by construction unable to try to break out of its sandbox and post their experiment protocols to an obscure german wiki?
I do think that for those of us who got into tech in the last couple of decades, the history can be rather obscured, but from the Fairchild semiconductor days, the military was a main customer, right?
I think there's a real question: would the world be a better place had Fairchild and peers e.g. refused to build ICs for missile systems (including the minuteman), for example? Would the same tech have been developed with purely civilian funding? Would missiles built with other tech have caused some awful disaster?
To the extent that the internet descended from ARPANET, and the internet is now both how we access a range of information and also hospitals get shut down by remote attackers, would the world have been better if researchers refused to build it b/c it was from military funding and motivations?
In retrospect it does seem like in the 1950s-70s the military was willing to fund some stuff that was actually not especially close to applications of killing people. I don't know that the kinds of things listed in this research are likely to lead to broad other applications in the future like microchips and networks did. But neither do I think it's straight forward to just condemn this stuff as purely evil and say we shouldn't do it.
> In retrospect it does seem like in the 1950s-70s the military was willing to fund some stuff that was actually not especially close to applications of killing people.
There's no "seem" about it, what you're saying is true, and it was because of changes in federal law. [0]
Thanks for pointing that out; I was not aware of it.
At the same time, when I helped with research in a cognitive psychology lab in school (much closer to the present than to that law), our PI said our funding came partly from a grant from the US Navy. We worked on Bayesian models of human learning and reasoning -- so "direct or apparent relationship to specific military function" has perhaps at times been interpreted somewhat loosely?
Consider that parts of yoga were stripped out and implemented for its effectiveness in reducing overuse injuries.
Teaching individuals quickly and at scale is a core necessity for militaries. Whether not or it gets used, the US Army usually has access to whatever is near the edge there (as in some unit or officer somewhere was promoting the method, and had the sign offs necessary to disseminate it, but culturally it was not fully put in place).
As a practical matter, I'm not sure the Mansfield amendment affected all agencies equally. My impression is that it affected ARPA (later DARPA) most profoundly. (pretty much everything I've read about this was in the context of ARPA and XEROX Parc so my perception might be a little skewed)
I'd speculate the Navy had perhaps more discretion over how it spent its research funding. For example, they funded polywell fusion research for years.
Even before Fairchild Semiconductor was founded, much of the original SF Bay Area tech industry started during and after WWII to develop radar and electronic warfare systems for the military.
Is there a place I can read about this? I am doing research on tech-histories (mainly in Asia) and i would say 100% of all tech coming from there has its origins in radio and radar tech, yes including fridges (you might have in your home right now)
I think this is true, and also a lot of tech, which often isn't as PR as the web/internet tech, gets funded by DOD/DARPA
That being said, I feel like a lot of the web-based tech, device tech etc came out of the liberal/leftist university environments, i.e. BSD unix, CERN, linux ish type projects. Berkeley/Stanford/MIT etc. So the culture of a lot of people who were rank and file and enthusiastic about this stuff from had this mindset.
Good point, but there's a very big difference between dual use tech that gets it's start in the military, vs building weapons systems.
Silicon, GPS, the Internet, nuclear power, most of rocketry and space etc., I think most would believe was worthwhile.
Building cruise missiles, or 'Civilian Monitoring Tech' - that's a different question.
I don't think that DARPA, the internet, or NASA etc. count as 'Military Industrial Complex', but Anduril of course does, and Palantir pretty much does as well.
That's 'new' for the Silicon Valley at least, it has mostly not been building 'defence tech' but rather general tech that has key defence use.
Ok you say rockets and space were "worthwhile" and missiles are "a different question".
But of course the space race proceeded from work on missiles. ICMBs are rockets that go high and fast and then happen to complete less than one full orbit.
And my understanding is that early ICs were for missiles because it was acceptable for the ICs to be very expensive in exchange for them being very light and small. Only a context that was optimizing for a tiny package (b/c extra weight also required more fuel) and was relatively cost insensitive would justify them at first. The point is building weapons sometimes _does_ turn into tech that can be used more generally. Does that justify the risk?
The lines blurred, IMO. Boycotting the 8086 wouldn't hamstring ICBM development. Boycotting the microwave oven wasn't going to prevent radar complexes from being constructed.
Much of the modern pushback seems to come from the new blood companies. Microsoft, Apple, Google, businesses that deliberately blurred the line to obscure which side they stand on. These companies all collude with the US government to some extent, but their desire to hide that cooperation creates a rift in what they say and do. You could have boycott Google and Apple technology to protect Americans[0] from warrantless surveillance. You could have boycott Starlink to thwart Starshield[1] and future bids on the Golden Dome. Information on all of this behavior is classified because it is obviously upsetting and creates the potential for dissent. But hiding the behavior also creates a dam-burst effect once you tell people the truth, including employees. The Snowden files were a legitimate surprise to most people working in FAANG at the time.
It's upsetting to think that American NOBUS abilities[2] hang like a Sword of Damocles over the so-called Free World. When you import American technology, you also import the surveillance and opinions of the American government. It has developed in a new and insidious way that doesn't fully resemble what we saw in WWII or the Cold War, unique to the information age.
Microsoft Apple and Google have not blurred any lines, they are not defence contractors in any way.
They are very large tech companies which will invariably have some relationship with every industry.
Apple had some early contracts in defence, Google's early research grands were National Science Foundation, but that's not defence companies.
'Boeing' blurs the lines definitely - they are 35% defence contractor mostly civilian airline agency and didn't really start as military.
It's important to understand that every company of a certain scale exists a bit differently than others, and in some industries, like aviation, will be considered 'strategic' and they're going to work with 'national interests' in a lot of ways, that may involved some kind of defence contracting.
For Microsoft, the first and largest contract mentioned in the table just below the fold on that page is for an IVAS system. Microsoft had been basing this off their HoloLens system but was a distinct product (they weren't just sending over the normal HoloLens and expecting that they were gonna work in the field). This contract was then transitioned to Anduril. But just saying of Microsoft that "they are not defence contractors in any way" is hard to reconcile with the facts of having a large contract to _develop a new product_ for the military ... even if it turned out to never have been used.
I see what you're trying to say - but MS is not a defence contractor. The amount of business they do selling 'Windows' and 'Azure' to DoD is 100x whatever little tiny bit of co-research they are doing otherwise. It's not relevant to their business, that's just 'a tiny project that happens to cross into some field'.
I understand that technically, it's a project that can be used for defence, but if you say that MS is a 'defence contractor' - then you also have to say:
MS is an 'Investment Bank' - almost 50% of MSFT's Operations are financial. Their financial opts are bigger that most investment banks.
MS is a 'Healthcare Company'
MS is a 'Gaming Company'
MS is an 'Entertainment Company'
MS is XXX, YYY and ZZZ etc.
better: "MS is a Tech Conglomerate that sells Windows (OS), Office, Outlook, Sales and Ops SaaS, and has Bing some other services, with XBox as major subdivision. They have a major financial footprint, and also have research activities in other sectors."
Palantir is partially Defence contractor even if a minority of their business is defence, it's a strategic business line for them, and it's material to their existence.
> I understand that technically, it's a project that can be used for defence
No, it's a project commissioned by and built for defense needs. It's not that someone at the DoD saw how to creatively re-use an existing civilian project.
I mean there's a very clear difference between 'strategic tech' that's supported by defence (GPS) - and weapons systems (Spitfire).
And yes, some weapons systems are just ... 'weapons related' ... and the dual-use comes later, so it's more nuanced. Some of it more obvious (radar), than others (cruise missiles).
But I think it's important to characterize these things.
Because building 'GPS', 'LLMs', 'food preservation', 'vaccines' and 'off-road vehicles' is not 'arbitrarily different' from building 'cruise missiles' or 'F22 fighters'.
They are completely different things.
'Tech' that can be used by the military', and 'weapons' are not 'arbitrary labels'.
HN is the only place in the world where someone would try to conflate the two to make a point, I don't think it's reasonable.
Interestingly AOL, Compuserve and similar ones originally ran without Internet. In fact it took a while until the Internet catched up with the level of interactivity, mostly through Flash. It's absurd that nowadays many websites require a central CDN to stay available.
CompuServe was in many ways its own single-source proprietary internet.
And CDNs are by nature distributed, even if their ownership is not. Just as it's unreasonable to expect a site operator to maintain their own global network backbone, it's unreasonable for every site to deploy global infrastructure for efficient content delivery. Which is not to say that we wouldn't be better off with a more diverse and distributed-in-the-ownership-sense market of CDN providers.
> Self-driving cars will never be safer than human drivers.
You're responding to something that actually brought data to the table to answer the question, though one can quibble with details.
Yet you're saying that self-driving cars will _never_ be safer, which is a much stronger claim, and you don't bring any evidence to back up the assertion.
I do think that currently we're at the stage where Waymo is trying to convince most of the world of the _quality_ of its offering, so for the moment, things are well aligned.
But in 5 years if there are multiple self-driving offerings competing on things like price, door-to-door time, OPEX (e.g. does a camera-only system with no moving parts need less skilled maintenance?), then someone may try to be the cheap but good enough self-driving service that cuts corners -- and we should want some execs to actually go to prison rather than just some cost-of-doing-business fines.
In my neighborhood there is an optometry place with a dark blue lighted sign. At night, it looks fuzzy, and I have to assume they knew this would be the case. People walking by squint at it and inevitably think "maybe I _should_ get my eyes checked!"
I think we gotta get people to see past the false dichotomy: part of their increased cost of living is due to climate change:
- your home insurance is more expensive b/c fire/hurricane/etc risk is worse in your state
- and you probably don't have flood insurance but the risks there are worse too
- your electricity bill is worse b/c you're using AC more b/c it's hotter
- climate change does have healthcare impacts, including expanded geographic range of a bunch of vector-borne illnesses, as well as acute issues during heat events. Even if you're not directly one of these patients, your health insurance costs will increase in part due to these risks.
- replacing key infrastructure which is damaged by extreme weather is expensive
- during extreme weather events, your ability to earn may be disrupted
- and every business you spend at also has increased costs due to the same factors, which they pass on to you
It turns out that multiple generations of emitting without paying attention to the externalities has created literal costs that we _all_ now bear, and these costs will continue to increase.
reply