jgord

care to share the one glaring obstacle ?

slightly related .. I saw a talk on DCs in space, and it said median Earth orbit had a latency of 500ms .. but back of envelope seems to be : 15,000km above Earth would have around 100ms latency, comparable to internet ping times.

Not an expert, feel free to weigh in.

jgord

先月議論

not least because of the bureaucratic delays making nuclear slow to roll out.

jgord

先月議論

Im surprised they haven't doubled down on deep drill geothermal.

jgord

先月議論

imo, theres a reasonable chance we will have functioning fusion reactors before these are turned on.

jgord

先月議論

Their VanDyke Brown looks more like a Burnt Sienna to my eyes, but that might just be my screen.

You may also enjoy the Chromatopia book : https://www.goodreads.com/book/show/40554590-chromatopia

The author produces a very nice range of oil colors under the Langridge brand in Melbourne, downunder... its nice to keep these artisanal practices alive.

Would be handy to have the standard pigment codes. Ive been gradually moving away from using heavy metals such as Cadmiums. Haven't found the perfect red, although Napthol Red PR170 and Pyrrole "Ferrari" Red PR254 are pretty close to primary for mixing from a limited palette.

Its really surprising how you can get gorgeous brick-red browns and deep purplish blacks from mixing a near primary red and primary blue.

jgord

先月議論

Do we need a cryptocurrency for trading / donating LLM compute tokens ?

jgord

2 か月前議論

Googled it :

> Anthropic is paying SpaceX $1.25 billion per month (approximately $15 billion annually) to use the computing infrastructure at the Colossus and Colossus II data center clusters in Memphis, Tennessee. This massive infrastructure deal gives Anthropic access to over 220,000 NVIDIA GPUs (mostly H100 and H200 chips) for large-scale AI inference.

jgord

2 か月前議論

by inference .. do you mean renting GPU compute ?

jgord

2 か月前議論

gorgeous piece of human-computer engineering art.

superb.

totally usable in contexts where comic sans might be seen as kind of mocking.

jgord

2 か月前議論

Australia is proposing a CGT tax of 30% on "capital gains" .. essentially taxing the _income_ from wealth.

This is more of a fair comparable to reason about when comparing taxing wealth and taxing wages.

In nerd-speak, taxing the Derivative of Wealth is comparable to taxing Income.

You could argue that a fair comparison of wages and wealth would first subtract the minimum cost of living, so that wage tax is effectively a tax on the growing wealth of wage-earners. This would arguably be a fairer tax comparison - in both cases it is the derivative of wealth that is being taxed.

If a large portion of the populace spends all their income on basic food, rent and petrol then they have no chance of wealth increase, and perhaps should fairly be charged 30% of their $0 growth in annual wealth.

jgord

2 か月前議論

without reading .. my immediate guess is that one hand is needed for maintaining upright balance, while the other hand grasps something important ?

jgord

2 か月前議論

it would be nice if they declared War against global climate change.

jgord

3 か月前議論

imo, 2D is maybe 70% of the problem to solve. We build and we think about buildings in 2D, possibly because its simpler/cheaper, possibly because of gravity corraling us into spaces with flat levels.

Computationally 2D is simpler, less dof, also conceptually. I decided to bring the scope in to get some wins on the board and early revenue.

I did do a lot of prior work in pure 3D, detecting walls and pipes :

  https://youtu.be/-o58qe8egS4
  https://youtu.be/8fjHNDGKeu4

ans also have some demos of fitting textured triangles to make lite 3D web browser models .. a while before gaussian splats came on the scene.

You are quite right, pipes do bend and jump thru 3D in complicated ways - and so flat slices only pick up part of their journey.

My rationale is most people want an accurate floorplan first, and quickly, and this saves time and captures much of whats needed.

Obviously the same 2D technique can be used in sequence - floorplan first, then get a slice of a wall, and get vector lines for all the door and window openings, or the stairs using the same process [ pointcloud slab slice -> ortho xray image -> vectors -> CAD ]

Feel free to contact via email

jgord

3 か月前議論

We lived close to the Formula One track in Melbourne for a time. They gave out free tickets to locals, as a kind of compensation for the noise and construction disturbance, so my young son and I enjoyed a day of rummaging around and seeing the cars, logos, hotdogs and candifloss. But the best bit has always been the epic flypass of the 747 or A380 and very noisy fighter jets.

Melbourne has spent a lot on extensive bike pathways and new train stops, and recently made some tram travel free [ as a crowd-pleaser to counter petrol price hikes ], so its quite a pretty city to explore on foot or bike.

Bangkok and Danang have some great cafes .. the best seem to be when you wander a few sois away from the main shopping zones.

I especially like seeing the old wooden elevated Thai houses, which are becoming rare. Another way to find hidden gems, is walk along the banks of a klong - you get to see the underbelly of the city, without the makeup.

The locals in Bangkok tend to love the new shiny hypermalls and pristine train stations that segway into them. The air-con is nice after an hour of roadside bargain hunting.

In BKK, if you like bargains on clothing or bricabrac, I _highly_ recommend going to the top floor of the Pantip building across and west down the road from the shiny upgraded 'The Mall' Ngamwongwan. The weekend indoor market is crazy busy with affordable bargain stalls with the cheapest jeans, tees etc. Smaller but more enjoyable than the massive and more famous Chatuchak. If by chance you need alterations, there are a couple of great shops on the 5th floor, iirc - 60 baht hems, wow. The 4th? floor foodcourt is quieter than most. There is a whole floor of Thai buddhist good luck charm amulets. You'll have to run the gamut of outdoor stalls to get into the place, but that can be fun. There is also an incredible coffee shop down soi 27, called "High Coffee Roaster". I was stranded looking for my airbnb, and a local came out of a shop and asked me if I was lost .. then recommended a local cafe I could wait at until checkin. The cafe staff caught me smelling my coffee, as it was so good, and then gifted me a tiny dish of ground coffee specifically to smell .. incredible coffee and superb service.

jgord

3 か月前議論

What freedoms do we value ? freedom of speech, freedom of compute, freedom to own assets, to sell our work or give it away, bodily autonomy, freedom to travel, to read to learn ?

Amid the massive hype of the Web3 Crypto era, there was a kernel of useful innovation : that you can choose to have unique digital copies of things, and thus you can have a way of sending value that bypasses the middlemen, be they local thugs, bent politicians, violent regimes, benevolent dictators, or the dominant hegemony.

Having central big-Corp approve your content or sign your executable or take a vig on your sales, or license your hardware - these may be common, but are not a universal law of nature.

The internet itself is our best example of the value of technology open for all to use. Frankly, that is in danger.

Whether it is bogus age-checks in your OS, a hidden bios OS, or the move away from owning your own compute [ because the GPU / CPU and RAM are priced so high you have to rent them ], consumers need to pool resources and ensure open access.

Kudos to France for mandating a Linux OS for their public service workforce. Good on the Europeans for doubling down on renewables to insulate themselves from petrodollar volatility, and making sure portable devices have replaceable batteries.

Cory Doctorow has some great rants on enshizzification. Garys Economics YT channel has some great rants on why high inequality steals resources, see also Piketty.

The technocrats on this forum have an understanding of these measures the common person may not, and thus a moral obligation to weigh in on the issues and warn 'genpop'.

Resist, dont let the buzzkills wear you down.

jgord

3 か月前議論

Joanna, if your reading this, welcome back .. really miss your blog !

In my ideal future, Joanna would write about things like :

  - how do we secure LLMs leading up to AGI, presumably they can bypass our best attempt at airgap firewalls
  - is consciousness different from thinking from intelligence ?
  - how do we ensure AI is distributed so all humans can benefit, and not used as a multiplier of extreme wealth inequality ?
  - whats missing in current AI LLMs ?

I'll take a crack at the last one : imo, the current generation of LLMs is missing :

  - a bias for truth / fact .. or a mode switch to make it bias truthiness
  - reasoning by chains of formal language [ formal logic ]
  - reasoning by chains of probabilistic inference [ bayesian logic ]
  - reasoning by deep simulation [ stochastic modelling ]
  - spatial reasoning [ 3D model of space, machinery, physics. 3D reconstruction. model of humans and animals with bones, muscles, mass ] 
  - mathematical modelling [ proposing formulae, checking fit ]
  - psychological model of humans, of human populations, their needs / motivations / rewards and psychoses

Essentially AGI would require a proper merging of RL style [ NN learning from a stochastic simulation of future states ] and current ChatGPT style LLMs.

An implication of the above is that future AGI will need to run on fast branching CPU _and_ massively parallel GPU with a fast data path between them - ie. balanced compute.

However, the best part will be discovering what the real Joanna writes about !

jgord

3 か月前議論

Fitting lines to 'xray' scans of buildings - turning pixels into vector art.

Lets say you have a complex industrial plant, or datacenter you want to upgrade.

You scan it with lidar and get a pointcloud and 360 panorama images. This gives you a large dataset, but what you really want is a floorplan, a lite CAD plan showing the racks, cable trays etc.

You take the scan, slice the pointcloud and make an ortho image .. it really looks like an xray of a building from the top down.

Then someone has to manually trace that in CAD to make a useful 3D model they can use for designing the upgrade.

So Im automating the boring manual part - turning the xray plan pixels into vector polylines, using machine learning.

One of our clients scanned their datacenter, and we generated a floorplan that shows all the rack box positions, cable trays, pipes etc.

Other examples : drawing the weld lines of patches in steel storage tanks, drawing in all the steel girder beams in a scan of an old railway bridge, or the windows, doors, ceiling pipes of a commercial realestate refurb.

gord at quato.xyz

As part of this work, were looking at running our custom machine learning kernel on multi-core x86 CPUs.

jgord

3 か月前議論

concur .. better to have a 40-buck fx82 for daily math and use Desmos for graphing, than fork out 250 to 300 for a super-duper calc they wont use.

jgord

3 か月前議論

Economic inequality was relatively low in 60s 70s, and economic growth was high in late 80s and late 90s. That seems to be a good backdrop for human flourishing, not just in science/tech but also in art, music, film etc. Recommend Piketty / Gary's Economics for more opinions on this.

A couple snippets from my own experience / wow points :

  - browsing bookstores was phenomenal, hand illustrated romance, fantasy and sci-fi covers
  - Turbo Pascal, Dr Dobbs journal
  - Wolfenstein and Quake ushered in the era 3D interactive games
  - shareware and file sharing on BBS boards, via dialup modem
  - tapes, then CDs, then minidisc !
  - digital watches, scientific calculators !!
  - double your hard-drive storage every 18months !!!

jgord

3 か月前議論

Recommend "Halt and Catch Fire" series, if you haven't already seen it.

Its set slightly earlier in the PC rush.

The magic of that era, I think was that there was this new explosion of technology and a smart teen could figure out how things worked and tinker.

It coincided with a relatively low level of economic inequality, meaning garage bands and garage startups were doable, imo.

投稿

Digital Twins: the missing pieces we can solve with Machine Learning

AUS startups need to Recycle talent, more

Homeschooling as a right, and a needed practical alternative

Web search 30M rows sub-second, cheaply

Distributive Rule is understanding rectangles

コメント