Hmm, actually lines up for me at least. It was a pretty big news item a few months ago when Salesforce did this drastic reduction in their Customer Service department, and Marc Benioff raved about how great AI was (you might have just missed it):
I’m beginning to doubt very much that will happen. AI/LLMs are already based on 99% of all accessible text in the world (I made that stat up, but I think I’m not far off). Where will the additional intelligence come from that SalesForce needs for the long tail, the nuance, and the tough cases? AI is good at what it’s already good at - I predict we won’t see another order of magnitude improvement with all the current approaches.
Hmm, am no LLM expert, but agree with you that the models themselves for the individual subject domains seem like they're starting to reach their peaks (Writing, solving math, coding, music gen...) and the improvements are becoming a lot less dramatic than couple of years ago.
But, feel like combining LLM's with other AI techniques seems like it could do so much more...
... As mentioned, am no expert, but seems like one of the next major focuses on LLM's is on verification of its answers, and adding to this, giving LLM's a sense for when its result are right or wrong. Yeah, feel like the ability for an LLM to introspect itself so it can gain an understanding of how it got its answer might be of help if knowing if its answer is right (think Anthropic has been working on this for awhile now), as well as scoring the reliability of the information sources.
And, they could also mix in a formal verification step, using some form of proof to prove that its results are right (for those answers that lend themselves to formal verification).
Am sure all this is all currently being tried. So any AI experts out there, feel free to correct me. Thanks!
The idea of formal verification works great for code or math where clear rules exist, but in customer support, there is no formal specification. You can't write a unit test for empathy or for "did we correctly understand that the customer actually wants a refund even though they're asking about settings." This is the Neuro-symbolic AI problem: to verify an LLM answer, you need a rigid ontology of the world (Knowledge Graph or rules), but the real world of customer interaction is chaos that cannot be fully formalized
Ah yes, and actually, Agreed (as mentioned, formal verification is only possible for "those answers that lend themselves to it").
Interesting that you mentioned Knowledge Graphs, haven't heard about these in a long time. Just looked up "Commonsense knowledge" page on wikipedia and seems like they're still being added to. Would you happen to know if they're useful yet and can do any real work? or are good enough to integrate with LLM's?
For me, it’s always having website productivity blockers on all my browsers across all my various devices (and for the most part, not installing news apps on any devices either). Haven’t simplified my digital life, but at least it’s very restricted. Yeah, if even one device doesn’t have one installed, feel like am vulnerable to having hours sucked away.
And actually, still browse the web and watch YouTube, but just on my non-work days.
Interesting point actually. yeah, when spacex was trying to build a reusable rockets, many traditional rocket scientists said that even if you are able to recover stages of the rocket, you still need to refurbish and test a great number of parts, and it just isn’t this panacea for lowering rocket costs (for example, the space shuttle, which was reusable spacecraft, but was super expensive to launch).
When spacex finally got falcon 9 reusability working (and am no expert in this) but from what I read, the pundits were partially right and partially wrong. Yes, refurbishment and testing on the Falcon 9 does cost a lot, but it still brings down the cost significantly (just looked it up, their saying nowadays, the cost savings is something like 70%, which actually is huge). And as importantly, you don’t have to build a new rocket for every launch, and once you get your refurbishment process down like clockwork, you can relaunch them quite often.
So maybe data centers in space won’t be like ones on earth, but they still might be very useful… One idea is that they could become true “space” data centers, that supply powerful computing for satellites near by. This way satellites could get access to much more powerful computing, while still being small themselves (but again, am no expert in this, so maybe this idea also has many holes, for example why not just offload processing to ground based data centers).
Although, I agree that for average people, over instrumentalizing your interactions becomes fake (although, to be honest, most could use a bit more, including myself, to communicate more effectively with those close to us).
Still, agree with others, seems like you're generalizing what is good for the average person is also good for those with personalities that are more at the extremes. Yeah, know a couple of people who just don't understand what people are thinking or feeling, ever. And so they have to learn a system of cues to look for to figure out whether a person is angry or sad or happy... These people need to create systems to make socialization work.
One question for you since your worked at SpaceX. Starship v4 is supposed to be able to bring 200 metric tons to LEO vs 35 metric tons for v2. Do you have any guesses on the finally amount that New Glenn will be able to bring up when it reaches its version/block 4?
Was thinking about that. It is interesting how they got so much working in just two launches compared to SpaceX, which works so incrementally.
Still, am wondering though if SpaceX's highly iterative approach is a better way, because with Blue Origin's more standard approach of getting everything right the first time, you may need to over engineer everything, which seems like it may take longer.
On the flipside, SpaceX's approach might tax the engineers, because they have to deal with launching so often, and maybe if they had done less launches, they might have actually gotten falcon and starship out quicker...
...But, then again maybe at Spacex, the "launch" engineers are really the ones that have to deal with getting the rockets ready for launch, while the core design engineers can focus on building the latest version. And all the launches are used to test out different ideas and gather real life data). Hmm, for my part, am leaning towards the spacex way of doing things.
(maybe SpaceX and Blue Origin engineers could share their thoughts if they're reading this??)
A lot of SpaceX employees went over to Blue Origin over the years, so there also was a lot of knowledge transfer and Blue could capitalize on the iterations of SpaceX.
Very interesting:) And as much as I like the ideas behind starship, think having a strong second launch provider is a good thing. Hope Blue Origin catches up even more
Good point, this is probably the right way to go, to have a factory that is able to build a lot of your rockets quickly and cheaply. Yeah, during development, this would allow for quicker build and launches, to test your vehicles. And afterwards, with a usable rocket, allows for a high number of rockets available for real missions.
agreed, they need to pick more engineer focused people who love building rockets rather than impersonal PR people. Sometimes, the broadcast felt like a standard business seminar.
although, they were doing it with a more complicated vehicle than the falcon 9, so the delay is "somewhat" understandable.
And only "somewhat," because new glenn seemed to take forever compared to starship. It does go to show, maybe the highly iterative approach that spacex takes really is faster (or, it could just be spacex has more highly skilled engineers, but I for one can't tell what the reasons are).
It's not about the delay, they can take as long as they want to build what they want to build. I object to their attempt to use patents to block competitors for decades when they didn't even have a product yet.
ah, yeah, patent trolling is pretty horrible (and Bezos is known for this - one click)...
... although, just to play a little devil's advocate, Bezos doesn't get enough credit for jump starting private spaceflight companies. Blue Origin was started 2 years before SpaceX. Am sure Blue Origin racked up a ton of patents.
yeah, didn't state it clearly. only meant that Blue Origin has actually been at it longer than SpaceX, and probably has around the same amount of patents as them because of it. Yeah, Blue Origin doesn't get as much credit for commercial space flight as spacex, and rightful so, but seems like they still did contribute a great deal (in fact, Blue Origin was the first to complete a vertical takeoff and landing, although it was with a suborbital vehicle).
Iterations are faster than modelling, no different for software where testing in prod with actual users ends up being quicker than in a testing environment.
Iterations in hardware businesses are far more expensive, particularly for early stage (by revenue not age) companies like Blue Origin. Outside of the Vulcan engine sales, joy rides and NASA grants they don't have much inflow and depend on equity infusion.
SpaceX also would find it tough without Starlink revenue to fund iterations for Starship. Similarly the early customer revenue ( plus the generous NASA grants) contributed to iterate on F9 be it Block V or for landing etc.
Beyond money, it also requires the ability to convince customers to be okay with the trade-offs and risks of constantly changing configurations, designs.
It is not that people do not know iterative testing with real artifacts is quicker, but most are limited in their ability to fund it or cannot convince customers, regulators to allow them.
Yeah, it does seem like iterative development with hardware is an extremely cash intensive way of development. And yes, what a genius move to fund a lot of this development with Starlink - it's amazing this seemingly off the cuff idea is such a cash cow, and it seemed at least like they got it up and running relatively quickly. Yeah, regardless how someone feels about Elon these days, Starlink has got to be up there for one of the most brilliant moves by an entrepreneur of all time.
And to come back to you point, yeah, I do see, you need the funds first to be able to support such a cash hungry way of development - which, on a tangent, kind of disappointed me (and a few others online) when Stoke Space decide to build their own 1st stage instead of just focusing on their unique 2nd stage. Like many in the past have mentioned, it seems like they'd be getting to space a lot quicker if they had just designed their 2nd stage to fit on a Falcon 9.
Starlink was not that amazing as a business decision.
If one expects to generate orders of magnitude more supply of a good (launch capacity), then one needs to expect the existing (conservative, long lead-time) market will have insufficient demand.
So one needs to generate additional demand.
So one needs to find a profit-generating business that's limited by mass in space / launches, where each component is inexpensive enough that its loss doesn't bankrupt the company.
Space-based telecommunications falls out pretty obviously from those requirements, given the pre-Starlink landscape (limited, exquisite assets serving the market at high premiums).
In small irony, it's also the exact same possibility space optimization that led to Amazon starting with books: Bezos didn't give a shit about books specifically, but he did like that they were long-tail, indefinitely warehouse-able, and shaped for efficient shipping.
In novel logistics spaces, it's better to find the business that matches capabilities than the other way around, because the company's core competency and value is their novel logistics solution.
It was an obvious market, that was visible years before the project was announced. I don't think any one was surprised, it was not like Apple launching iPod or the iPhone.
What was impressive is at that they solved a lot of hard problems like satellite manufacturing at scale, phased array dishes, or fleet management of thousands of satellites or laser interconnects between satellites, and so on, for basically a side project to increase their primary product demand enough to justify the reuse being a useful feature.
Hmm, don't know, easy to say it was obvious in hindsight. But over the years, Google project Loon and other similar attempts at increasing internet coverage (think Facebook tried too at one point) have not been nearly as successful. Yeah, still not convinced it was obviously going to be successful, but maybe am missing some aspect you're seeing.
Market was obvious, solutions weren’t as you cite there were many tech failures, it was just a logical extension to their business that is not really hindsight.
It was not the same kind of new market entry Apple did with the iPhone or even the iPod , or Amazon doing AWS, which if we claim today as obvious would be hindsight
Well, sure, agree that there is a natural logic to the idea, but to actually go through with something that no one has done before and actually execute it (which as we all from the tech/sci industries here know), and also do it on a large scale and be very successful is an entirely different matter. Yeah, the number of things that need to go right is still pretty high, and at least to me, was extremely impressive. But to each his own.
It was also a great move because they could take more risks launching their own Starlink satellites and prove out the reliability of the Falcon 9 to others. They also are very had to compete with when they build, launch, and deploy the system all in house.
Was thinking this too. It reminds me of how TSMC's fab has a lot more volume than Intel's, because TSMC has outside customers and high-volume is what is needed to perfect a chip fabrication process (getting many more chances to fix any problems, and once you finally do, have the volume necessary to make it profitable). What a great idea it was:)
You get lot more data when running real world experiments .
For off world missions, the best examples are the Soviet Venus missions of how iterating and sticking with the goals helped do some incredible research which would be hard to replicate even today .
The benefit of not doing quick and dirty is why we got out The longevity of voyager or some of the mars rovers or ingenuity.
They were "launching cities" as one of their program chiefs said. Yes, when you can arbitrarily tax you population you can afford these loud propaganda headlines.
Hard to draw super hard conclusions. Could also be that the bets made on Falcon turned out to be particularly good, vs a more methodical approach Blue Origin took. The highly iterative approach _may_ be faster, but I don't see any evidence yet that it will _always_ be faster. Just depends on how good your bets are and how much in-flight testing you happen to have to do based on a design.
Would be interesting to see more detailed information like specific engineering issues being resolved one way vs another.
Falcon beat New Glenn to the punch, but New Glenn is probably more capable overall, so it's not an apples to oranges comparison. Completion of Starship would really help the iterative approach case though (ignoring the junk it leaves scatter around the world when it goes boom)
People need to remember that New Glenn is completely artificial in market terms. Blue Origin had literally infinite money, and if not sponsored by the richest man on the planet it could never exists. And New Glenn even if its 'better' then Falcon 9 (yet to be demonstrated) will likely never make back its development cost.
I think people just don't understand what an absurd amount of cash burn Blue had for the last 10+ years.
So when it comes to iterative vs methodical, this is a perfectly clear case. SpaceX did it faster and for an amount of money that is so much less then Blue that its hard to even compare the numbers.
Go back and just look at how many people worked at Blue, and then do the math on what their cash burn rate was just for people.
Rocket Lab is also taking a more methodical and less iterative approach with Neutron, which should be ready some time next year. If they make that work well, that will be another point in favor of a methodical approach.
There should be an in depth academic study on their two approaches, it seems like it'd be valuable.
To me at least, given the (probably) positive affects iterative dev has (overall) had on software development, my personal feeling is it'd be useful for most other types of engineering. But (as someone else also pointed out) iterative is much more expensive in hardware fields, given the high cost of materials, and you need to have a lot more funds to build hardware this way.
Spacex tends to "build rocket factories" instead of building one rocket. So they can launch and reuse hundreds a year. They're repeating this with starship.
It's hard to know what BO is doing because they're so quiet all the time, but to what degree is this scaling true for them also?
Going by the Tim Todds interview with Jeff Bezos it seems like BOs approach is very similar in this area. It looked to me that the machines they had there to build NG is set up to produce rockets in large quantities. He talked about their goals with the second stage, and that they’re looking at making a reusable version but that in parallel they’re also doing cost optimisations that may make it so cheap that reuse doesn’t make sense.
Was talking with someone else, yeah, focusing on a rocket factory instead of just building a couple of rockets does seem like a good idea. Allows you to build a lot of test articles during development, even ones that you'll launch like Space X, and during real flights, you'll have a lot of rockets available for real launches.
Yeah, Bezos has been putting most of his attention there for the past few years. And why not? What's more interesting, running a online marketplace (which still actually seems pretty interesting), or building rockets to fly into space:)
It'd be a win for me, probably not for him. Or investors, sadly, at least not in the short term. I would turn off the non-Amazon sales platform. It might be feasible to save the brand, but of course the immediate effect would be a loss of that revenue stream from all the random Chinese "brands" that flood it with cheap garbage and counterfeits.
Or at least make a persistent toggle switch in the UI where you can say "I never want to see a single product that is not shipped-from-and-sold-by-Amazon." And end commingling with any product that Amazon itself sells, if that is occurring.
That's why they don't let morons like me run big business :). I care about things that only matter when you are a small business apparently.
Ah, focus on quality of products being listed. And, maybe that means there is room for a high-end competitor to Amazon eventually (am not seeing this anytime soon, but maybe in a couple decades??)
Yes, but it's also a harder problem, aiming to reuse everything instead of just the first stage.
And they have at least reached orbital velocity on several occasions, so they could have physically orbited. They just purposely chose a trajectory that wasn't an actual orbit.
Agreed. And even if they don't ever fully reuse the second stage, they still could use this gigantic working rocket as a (probably still) very cost effective to transport things into space.
Of course, it could still happen, as maybe AI systems just need another few years to mature before trying to fully replace jobs like this...
... although, one thing I agree with you is that there isn't much info online on these quotes from Salesforce executives, so could be made up.