poslathian

We’ll see what the future holds but as an old timer, using LLMs to creat applications seems exactly the same as:

Python/JS and their ecosystem replacing OS hosted C/C++ which replaced bare metal Assembly which replaced digital logic which replaced analog circuits which replaced mechanical design as the “standard goto tool” for how to create programs.

Starting with punchcard looms and Ada Lovelace maybe.

In every case we trade resource efficiency and lower level understanding for developer velocity and raise the upper bound on system complexity, capability, and somehow performance (despite the wasted efficiency).

Place and routing compilers used in semiconductor design are not. Ironically, simulated annealing is the typical mechanism and is by any appropriate definition, imo, a type of AI.

Whatever you do in your life using devices that run software are proof that these tools are effective for continuing to scale complexity. Annoying to use also ;)

It seems this muscle has atrophied in nearly every niche it once thrived:

neighborhood kids self organizing play, multigenerational households and family run businesses, boarding houses where personal spaces are shared with strangers, participation in civic clubs and churches (which was as much about business networking and social prestige as anything but still organized collective effort to worthy causes), self organized physical markets (like the grove in LA in the 30s), chambers of commerce, local business (cartels) parceling up their markets, barn raising, “culture scenes” like the west village in the 60s…

perhaps most saliently to me personally, the end of faculty led university management.

Edit - the article (first I’ve seen on this story) hints that the Australian government is the biller. Did they load the plane up with ultra expensive per hour support staff? Was this plane their Air Force one equivalent or something?

Can someone knowledgeable explain how it’s possible that a 500k charter was his only option? I would have thought for 10k/hr you can charter pretty much whatever to wherever you want.

Several such pilots for this have been in flight the last couple years.

One problem with owner occupied tax breaks is that it incentivizes cities to find their revenue elsewhere, namely building office/commercial and apartments.

309fun!

I once wasted several months of schedule and payroll wrestling to improve a DNN model - which resulted in that model getting much much better thanks to tons of creative and interesting work.

We swapped the learned model for a perfect a oracle and realized the actual problem was elsewhere in the system and was easily fixed. None of those model enhancements were needed.

MS had yet to fully stabilize that lead a full decade after they had won the os platform standard for ibm compatible pcs. A platform standard moat goes way way beyond a brand advantage.

Azure, while significant, has no similar monopoly to support OpenAI. Do you really see a structural advantage to openAI beyond the Microsoft products integrating it?

Came here for this comment. I’ve always kept an eye open for anyone who has applied this insight to something. There is something also satisfying about how accessible accounting is vs any other domain you encounter invariants (like noethers theorem, group theory, or equivariant neural networks)

Thanks! This is a really great demo of how to bootstrap spatial volumes starting from just ordered pairs on a discrete set.

So cool. Hope to see more around this corner.

And the next paper will be this but with Gaussian splatting as the representation ;)

Weird, right? I couldn’t tell if they are trying to rebrand 2019’s yoloact model as a “foundation model” for marketing, something else with ultralyrics, or if there is hide-the-ball happening.

I thought I understood that the speculator highlights and view dependent color problems you mention are massively improved via adding spherical harmonics into each ellipse?

Beautiful. Agree. The juniors come in with a lot of technical education that would have been near impossible to learn on the job.

Effective teamwork in this setting is just as foundational as skills they acquired in school but they only begin learning it after they graduate.

Very rewarding to contribute to the growth in jr devs teamwork when it happens.

Is this any different than a beginner musician who learns to play their favorite songs but also write their own?

This is truly the wisest comment on this otherwise very questionable thread.

It is hugely challenging - and rewarding - to try and solve a constraint satisfaction problem that balances the risk, cost, upside that are a function of the huge set of decisions made in any complex undertaking.

It is massively easier to identify flaws and demand mitigations, which you can do ad infinitum.

Reddit.com/r/fuckcars Reddit.com/r/suburbanhell

It’s not the sad isolating culture leading to cars it’s the reverse. Who wants to sit outside with cars driving nearby?

Such an interesting principal to carry, I never thought about it before.

Now that I am, it seems kinder to fail people early out of paths they are ill suited to follow, no?

Why is this downvoted? This whole thread is scratching it head as to the “real reason” behind this. Here you go. Infrastructure is way too expensive to maintain for stuff you don’t really need.

OP, just to balance out these comments, you hooked me with the readme. DNNs have been moving way faster than pedagogy can keep up and these sorts of learning experiments are super helpful.

I think it’s way more likely that being desperate prevents you from innovating. That’s why people support the UBI…

Desperate people get myopic and risk averse. Stress focuses you on the next step at the expense of the long view. If escaping your pattern requires a longer term plan you are stuck. You might not even see the loop you are in anyway.

Isn’t that literally the poverty trap?

That last bar seems way too high.

UBI might become low enough to be practical with a mix of changing attitudes around living “the bare minimum” and motivation to fix absurd expenses when possible.

Living in a dorm room in a rust belt town with a cafeteria and 1940s-level clinic might look less dystopian when it’s the totally free option.

Get that level low enough and you can expect most people can actually succeed in creating about that much in value because it’s absurdly low and any person would struggle to create a hard 0.

Even without UBI, if it were legal, I think an ultra cheap lifestyle option would be a popular niche. Machines can easily support a 19th century living standard for almost nothing.

Thanks for this, well distilled. I used to be anxious about how the handful of folks I knew running breakout companies all leaned heavily towards asshole. Was it a necessary but not sufficient trait to achieve great performance?

I’ve come to learn that if you do the things on your list well, some people are going to see that as being an asshole.

Clearing roadblocks quickly ( and nearly all roadblocks are made of people-problems after about 50+), insisting on certain aspects of a product vision for years, and engaging in someone else’s details are all strongly correlated with annoying someone.

I also learned that the assholes I knew were just the usual mean, dunning Kruger style people that exist everywhere. Those traits ultimately meant their inner circle perpetually excludes the kind of really great people you’d want to surround yourself with.

I always assumed the key word here was “indoor,” and not “vertical.” Indoors is a place where it is much, much, easier to automate and control growing conditions. Both could add up to faster better cheaper. Given how amazingly huge; dense; and performant logistics has gotten, I can’t imagine the real estate angle is moving the needle very much.

The spreadsheets go and so do all the types of jobs that use them and similar. This could be a great outcome - all of us spending more time in reality — if someone can clearly articulate a vision for exactly we’re supposed to be doing there.

Best comment. And one you don’t hear nearly enough from typical product and eng managers.

I often wonder how we might estimate how much of the surplus created by economic growth and technology has been hoovered up by “alarp” in its many forms. I’m scared of the answer, and of the people who wouldn’t see a problem with it.

Super! Makes sense since Skydio is also amazing.

How much data is used for fine tuning? Since spectrograms are (surely?) very out of distribution for the pre training dataset, how much does value does the pre training really bring?