tomas789

My point was quite the other way around. With those legacy tools it is very simple and most people would just build the feature. Serverless adds more friction if you don’t have a suitable backend. And this makes sense then.

I love how in the age of good old PHP, Django and alike this would be insanity. But with serverless it makes sense I guess.

For quite some time now I believe there should be a better alternative to git. I have never thought of building it on top of git. Neat.

What I think should be there is a daemon doing autocommity and automatic sync with server. Even if you work from multiple machines.

As mentioned here [1] they use atom interferometry to build extremely precise sensors of acceleration and rotation. This is everything needed to build an inertial navigation system (INS). This is completely independent of any external system. My guess is that this is what makes is so interesting for subs.

Probably the biggest issue with INSs is accuracy of the raw IMU measurements. The acceleration must be integrated twice. That means any error in the measurement gets exponentially larger with time. Rotation velocity measurements are integrated only once (some IMUs measure absolute orientation but it is not the case here) so the error does not explode as fast. Is is the combination with translation that is problematic. If you rotate around vertical axis by 90deg +/- 1% and then travel for 1000km +/- 1% the rotation accuracy will cause bigger error in final location.

To have an INS capable of long time operation without any other means of drift correction one has to nail many other things.

- Gravitational constant varies and you have to have an accurate map of it. We might not have a good enough one. I don't know. The subs could collect their measurements and perform a large scale optimization to increase the accuracy of this map.

- The rotation of earth gets added to the rotation speed measurements. The rotation is not constant (hence the leap seconds from now and then). We have to be able to extrapolate it well enough.

- Sampling rate of the quantum IMU has to be high enough. How usefull is the most precise measurement if you can measure only once in 10 minutes? Traditional IMUs sample at 1 kHz and more.

- Does the new quantum gyroscope add enough value to the accuracy of the final IMU compared to Fibre-optic gyroscopes? They are incredibly accurate as they are based on laser interferometry and then can sample at high frequency. Plus they are an established technology which partially benefits from economy of scale [5] because they are used in airplanes [6].

- The measurements of both rotation velocity and acceleration have to be measured in 3 axes. Are the relative orientations and translations of those axes stable enough as the device moves?

- Is the noise present in the measurements corelated to something or is it truly random? The corelation could mean an systematic error to the state estimate.

- You will likely also have to have an atomic clock on board labeling each IMU measurement as soon as it is made. But this is not a problem nowadays as one can have a small Rubidium-based atomic clock in a package starting to resemble a chip [2] [3] [4].

- And of course you will need enough computing power to calculate everything. I'm not sure how modern hardware they are allowed on board of nuclear-carrying subs. This is slight problem for things like Mars rover as due to the testing requirements the main chip can be easily 30 years old when the rocket is launched [7].

Also, the situation is not as bad as we can utilize the sensor fusion and correct lots of errors caused by IMU measurements. From the top of my head one can obviously measure water pressure which estimates depth (with some error). Sea bed topology could be other type of measurement but my guess is that this measurement is too crude compared to the accuracy of the quantum-based INS.

[1] https://www.imperial.ac.uk/centre-for-cold-matter/research/q...

[2] https://eu.mouser.com/new/microchip/microchip-macsa5x-atomic...

[3] https://www.sparkfun.com/products/14830

[4] https://www.accubeat.com/nano-atomic-clock

[5] https://emcore.com/wp-content/uploads/2016/12/EMCORE-Commerc...

[6] https://www.airbus.com/en/products-services/space/equipment/...

[7] https://bigthink.com/hard-science/perseverance-rover-brain/

A no-op API call to FastAPI takes more than that. Even when running on localhost. And FastAPI is one of the better contestants.

Amatures. They need to start their own country and then anything is legal. A business model that is proven by milenia.

Just because I can come up with something completely new does not mean it is a good idea.

I think there are two scenarios which both call for different solution.

One scenario is when your goal is to make something new. Something, that would feel different and have different ergonomics. This is like inventing a Python when there are mostly C-style and Pascal-style languages.

Other is when you are interested in a particular behaviour of the language. Like type system. Why should you bother whether you write let x = 5; or int x = 5; when all you are interested in is its type. This is likely the case of this paper.

A counter example for this would be Haskell. Before Haskell most of the researchers in functional languages started by defining CPU instruction set, then added their programming language, defined how backend of the compilation works and only after that they presented their point. This is sort of situation we see here.

Then Haskell came. They presented a nice language based on something others were familiar with (lambda calculus). Since then, two interesting things happened as I understand it. Poeple started adding more nuanced optimization to lambda claculus based backend. This is similar to what LLVM does. Second thing is that instead of comming up with their own languages, most adopted Haskell-style syntax. Just look at modern of functional languages. The mostly look like Haskell (well, ML).

Honest question. Why many researchers in the area reach for a completely custom language? There is a huge barrier for new language. All the tooling, bindings for OS, tailor made libraries for specific use cases and so on. It all has to be done almost from scratch.

Why not to start from what people already know and love and bolt your experiments on top of that? What I do with my DSL AmpLang is to start with Python, throw away parts I don’t like and add the things I’m interested in. It seems to work well enough and my language feels familiar for Python developers. I did not test that but many of existing Python programs would work out of the box in AmpLang. Python grammar is very simple. Even more so for 3.9 which does not have pattern matching.

They are all green - https://www.apple.com/support/systemstatus/

This is the first time I see an outage from Apple. I have personally seen one or two from most other big tech companies. If this is general pattern what is Apple doing different?

In our country we have a problem with exchanges giving horrendous rates to tourists (like 16 for 1 EUR instead of fair 24). Governor of central bank was asked why they don't stop it. He replied that they in theory could but seems like people are using such services and if they intervene then they are disrupting a free market. They are offering a service and they have customers. Should they stop them just because they have high margins?

What I want to say is that I don't like this practice either but should we cancel businesses just because we don't like them? They are not stealing or anything. I'd say they are in a light grey business but for sure not black.

And I guess there is a bright side to this all too. At least for large-ish businesses. If nissan.com was owned by domain squatter then they would sell it to Nisan for 1M or something and both sides would be happy.

I'd suggest to not use this technique as it gets quite confusing for others. You have to connect to something and you don't know whether it is .com, .org or what else. It is common to use different dummy names like githubusercontent.com and so. You can easily grab a random name like this and is way less confusing.

Nice concept. Similar techniques can be found in neuroevolution, where genetic algorithms optimize neural networks. One major benefit is the ability to compare the performance of both subpar and exceptional models on a relatively fair basis. Considering this evidence, I anticipate that this approach will gain greater significance as the quality of language models surpasses our existing benchmarks. However, it's worth noting that a potential pitfall arises when certain models outshine the rest of the group.

My experience was from Spain.

AirBnB used to be a great option compared to hotels. This gap is probably reversed at this point and hotels are becoming once again the preferred choice for many. AirBnB's disadvantages are mostly about very high price on per-night basis, high cleaning fees and many (often unreasonable) requirements from the hosts. Mostly regarding cleanliness upon departure.

That being said I think from my personal experience that there is a bigger problem with the service. I got a very bad experience from a relatively new all-5-star host. The host lied to us and AirBnB, tried to get more than $1000 for various claims and so on. At the end AirBnB did a very poor job at mediating the claim. Then I wrote a review. I tried to be as objective as possible. AirBnB simply deleted the review. They didn't ask us to correct anything or anything. They straight out deleted it.

I believe that by doing so AirBnB is misleading both guests and their investors. This creates an impression that the service is better than it actually is. That is likely inflating their stock price. I personally believe that there should be a class action lawsuit or Hindenburg research paper about this for them to do anything about this.

One more datapoint here. Good performance helps avoiding scaling problems. One example is an API I recently wrote. I was initially in Python and performance was OK I guess. I had to take care of horizontal scaling, how to manage the infrastructure, a bit of distributed scaling and all the other goodies that come along. By rewriting it in Rust and caring about performance I was able to run it on a single machine (which is ok for that product). That saved me quite some hustle. And my average response time went from 300ms to 29ms.

If the app would scale a lot I would still need to go into all that trouble but at this early stage the benefits of really simple infrastructure is immense.

I have like 20 cables at home with USB-C on at least one end. Quite often I plug it is and it doesn't work. Is is the charger? Is it the end device? Is it the cable? Other times it charges but surprisingly slow. It happened to me that I bought 120 EUR 100W GAN charger and was actually charging at 60W or so because I used wrong cable. Those devices will take out one variable.

My experience with code assistant is strikingly similar to my current experience with Vim. Sometimes, it makes me slightly faster. Other times the experience breaks down spectacularly on previously simple things. And last but not least, it is difficult to get rid of it.

The article links to a emarker reader [1]. This seems like something that milions of households should/will own. Do you know where can I buy a cheap and reliable one?

[1] https://www.reddit.com/r/UsbCHardware/comments/z2hncm/recent...

I'm sorry. This is not gonna happen. And it should not. Unless it is magically done worldwide. If 99% of labs stop training that it would be a huge boost for the remaining 1%. Do you remember what happened last time we asked poeple to stop doing something (going out)?

Ghidra is reasonably simple to pick up at the entry level. I use it just for fun to make a keygen for commercial software from time to time. Just to flex the muscle.

My take aways are: 1/ You can do RE without knowledge of assembler (which I know nothing about) 2/ C decompiler is useful and you will need to learn some patterns of how things get disassembled 3/ There are many good videos on youtube on how to get started 4/ There is a debugger to see what the program actually does but I never managed to get it running. That would be awesome feature to use.

Yes, they emit CO2 when burned. The goal is not no-emissions but net-zero. Question is where does the carbon in the e-Fuel come from?

That is true. The changes we will need to make will be significant. There is a change that some technology will cross the threshold of being the best-in-class or the-next-best-thing and the adoption will skyrocket. And those things might change very switfly.

Likely problem is with industries that are more difficult to electrify. Like aviation.

This is such a great move! Let me explain. What you care about is not EVs. It is carbon neutrality. You actions should be geared towards that. Not EVs. They are just one way of getting there.

Having commercially viable e-Fuels would have a few advantages. You can keep your old car and be net-zero, sport cars would be a thing, you can decarbonize other types of transport which are more complicated to decarbonize than personal vehicles. Like maritime, airplanes, long haul tracking and so on. Its like killing many many birds with one stone.

At the momeny, the fact that ICE cars would be banned could be a red flag for new projects trying to push for e-Fuels as it limits their customer base. This is important because at the beginning they will be more expensive that crude-based products so you need to find a niche market where poeple are willing to pay premium for net-zero. I think that companies like Google, Apple and so on would be willing to pay maybe even 2x or 3x for the fuel if it guarantees true net zero (no offsets).

As for the efficiency. It is true that the e-Fuel based chain would be less efficient than all-electric but there is a way around it. Efficiency is mostly important to keep the costs low. If you need 1kWh on the output and the efficiency is 20% instead of 90%, you will need 5 kWh instead of 1.11 kWh on the input. But what if the input is next to free? Two ways of getting that. One is by having a surplus electricity which we will have a lot as we inevitably have much more installed capacity in solar than peak load. The other one is to build renewables where they work very well (like Mauritania for solar), produce e-Fuels on-site and than ship them globally. Economy of scale will allow you to decrease cost of e-Fuels dramatically.

It might seem like a long shot for you at the moment but I gave it a lot of thought and discussions and research and I think that this very realistic and very likely scenario.

PS: If you want to start a business in this area, let me know. I'd be very interested!

I guess that nowadays CV is actually two fields that kinda share the same name. One would be neural-network based and the other one the good old-fashioned CV.

The answer depends on what kind of CV you're looking for. As for the old-fashioned CV I think that visual odometry, bundle adjustment and dense reconstruction are great things to play with and they have practical applications.

One hard-ish problem I used to work at is city-scale 3D reconstruction from photos. You can have street-level image data you collected to get the high detail and some aerial pictures to get the consistency for wide-area. A problem of matching street-level and aerial pictures is even interesting from research point of view (keyword is wide baseline image matching). With this you can build something similar to what Google Map has where you point your phone on the street and you get overlay directions. And if you call it mixed-reality metaverse, you can start a startup 6 months ago.

As for the NN-based I'll leave that for others.

Vimium is very interesting. I was probably inevitable that somebody made something like this. Maybe a bit of too steep learning curve for what am I trying to achieve.

I have already tried you.com but honestly the results were underwhelming. The suggestions of the results page were not very good. I was trying that while doing some Python. I guess other languages might be even worse.

You just have to touch some fires yourself to learn the lesson I guess :)

I did a master's thesis on visual inertial navigation just a few years ago. Around the time when ARKit and ARCore got released (what a time to be robotics researcher). At that point global shutter cameras were significantly more expensive. Basically you had to go for what is called "machine vision cameras". They were very robust, deployment ready and highly capable and configurable. A camera with similar parameters to this one could cost you anywhere in the neighbourhood of 1000 - 3000 USD. It is great to see that they are getting very affordable. I think this trend will help bring visual or visual-inertial navigations to the arsenal of average makers. Before, you had to have a very advanced software to compensate for imperfections of your sensors and the calibration complexity meant that only professionals could have them running reliably. We will se what people make with them.