tatjam

Precisely, I would say embassy is a satisfying middle-point between "baremetal" firmware and running something like FreeRTOS / NuttX that hides the event loop from you.

Writing embedded code with an async-aware programming language is wonderful (see Rust's embassy), but wonder how competitive this is when you need to push large quantities of data through a micro controller, I presume this is not suitable for real-time stuff?

I do agree with this, and in fact I do often use LLMs for for these tasks! I guess my message is more intended towards vibe-only coders (and, I guess, the non-technical higher ups drooling at the idea of never having to hire another developer).

Would you consider drudgery the in-depth thinking that's required to actually go and write that algorithm, think out all the data ownership relationships, name the variables, think the edge cases for the tests?

For me, the act of sitting down and writing the code is what actually leads to true understanding of the logic, in a similar way to how the only way to understand a mathematical proof is to go trough it. Sure, I'm not doing anything useful by showing that the root of 2 is irrational, but by doing that I gain insights that are otherwise impossible to transfer between two minds.

I believe that coding was one of the few things (among, for example, writing math proofs, or that weird process of crafting something with your hands where the object you are building becomes intimately evident) that get our brains to a higher level of abstraction than normal mammal "survival" thinking. And it makes me very sad to see it thrown out of the window in the name of a productivity that may not even be real.

Except the cooks don't exist anymore as they all have become head chefs (or changed careers) and the food is being cooked by magical cooking black boxes

It's perfectly framed in the loss of local-compute autonomy that seems to be the trend nowadays: https://andrewkelley.me/post/renting-is-for-suckers.html

If it wasn't obvious by now, the big capital doesn't really care about open source, hobby coding or small companies.

Claude, fix the memory leaks, or you'll go to jail!

Wellll you could technically jam their uplink channels, but doing so may get the US in your doorstep quite quickly

Confusingly, because it stops the "particle spawning" but not the animation! At first I thought it just changed the background to orange.

Exactly, for the thing that has been done in Github 10000x times over, LLMs are pretty awesome and they speed up your job significantly (it's arguable if you would be better off using some abstraction already built if that's the case).

But try to do something novel and... they become nearly useless. Not like anything particularly difficult, just something that's so niche it's never been done before. It will most likely hallucinate some methods and call it a day.

As a personal anecdote, I was doing some LTSpice simulations and tried to get Claude Sonnet to write a plot expression to convert reactance to apparent capacitance in an AC sweep. It hallucinated pretty much the entire thing, and got the equation wrong (assumed the source was unit intensity, while LTSpice models AC circuits with unit voltage. This surely is on the internet, but apparently has never been written alongside the need to convert an impedance to capacitance!).

This kind of stuff could trigger the next revolution in computing, as the theoretical energy consumption of computing is pretty insignificant. Imagine if we could make computers with near-zero energy dissipation! A "solid 3D" computer would then become possible, and Moore's law may keep going until we exhaust the new dimension ;)

Lean 4 seems to be pretty AI-usable, and you get insane guarantees (but LLM do seem to make very heavy use of "sorry")

As a quick example, compare doing embedded work with a C static uint8_t[MAX_BUFFER_SIZE] alongside a FreeRTOS semaphore and counter for the number of bytes written, vs using Rust's heapless::Vec<u8, MAX_BUFFER_SIZE>, behind a embassy Mutex.

The first will be a real pain, as you now have 3 global variables, and the second will look pretty much like multi-threaded Rust running on a normal OS, but with some extra logic to handle the buffer growing too big.

You can probably squeeze more performance out of the C code, specially if you know your system in-depth, but (from experience) it's very easy to lose track of the program's state and end up shooting your foot.

I think the key is that the LLM is having no trouble mapping from one "embedding" of the language to another (the task they are best performers at!), and that appears extremely intelligent to us humans, but certainly is not all there's to intelligence.

But just take a look at how LLMs struggle to handle dynamical, complex systems such as the "vending machine" paper published some time ago. Those kind of tasks, which we humans tend to think of as "less intelligent" than say, converting human language to a C++ implementation, seem to have some kind of higher (or at least, different) complexity than the embedding mapping done by LLMs. Maybe that's what we typically refer to as creativity? And if so, modern LLMs certainly struggle with that!

Quite sci-fi that we have created a "mind" so alien we struggle to even agree on the word to define what it's doing :)

I'm looking for it as a general development machine, sometimes I do stuff that requires GPUs so if they can get a competitive price wrt. building a custom PC then I'm all in for the convenient form factor!

I think most people would agree that AlphaEvolve is not AGI, but any AGI system must be a bit like AlphaEvolve, in the sense that it must be able to iteratively interact with an external system towards some sort of goal stated both abstractly and using some metrics.

I like to think that the fundamental difference between AlphaEvolve and your typical genetic / optimization algorithms is the ability to work with the context of its goal in an abstract manner instead of just the derivatives of the cost function against the inputs, thus being able to tackle problems with mind-boggling dimensionality.

You can just use the GPL, then it's free, but your labour cannot be so easily profited from by big corps

Haha love this "Not intended for radar applications. Core functionality needed for radar not included due to export control restrictions".

Wonder how they prevent usage as radar as this thing could pretty much be a drop-in missile seeker.

Well you can't reach a high orbit using air breathing engines because your impulse must be given within the atmosphere, and then your trajectory inevitably re-intercepts the atmosphere (unless you achieve an escape trajectory) and would decay quickly. You can get around this by packing a small rocket engine and circularizing on apogee!

Love how clever the logo is! I wonder why the RTC has the resonator exposed? I cannot find anything on the datasheet that explains its purpose

Also you have to consider that even top-notch lasers have a divergence (out of memory, I may be wrong by an order of magnitude) of 10mrad or so, and you are physically limited to not much better than that, so you need pretty insane powers to damage, let alone destroy, a satellite at LEO altitude.

Exactly, if you are interested in what you are doing, you will learn no matter what tool you use, be it AI or vim without syntax highlighting. A person that's trying to get rich quick, and sees software as a means to that end (and nothing more) will learn pretty much nothing vibe coding.

Pretty much all problems can be reduced to some number of matrix multiplications ;)

Galileo already has search and rescue (SAR) and emergency functions (EWSS)! GPS is trickier, as it's still sort of a millitary-first platform.

I wonder how could you keep the LLM from going bonkers as the game progresses? I have a feeling it's possibly better to re-create the prompts after some time, and have the LLM work more like one of those "reasoning models" with the game as something it can interact with.

Otherwise you run into the risk of "TOTAL NUCLEAR FINANCIAL LEGAL DESTRUCTION" ;)

I like the "dopamine hit" of changing a task from TODO to DONE that comes from colors. I use this very simple vim syntax file for that :)

syntax match TODOKey "TODO"

syntax match DONEKey "DONE"

syntax match BLCKKey "BLCK"

syntax match MAYBKey "MAYB"

syntax match Comment "\/\/.*$"

hi def link TODOKey DiagnosticWarn

hi def link DONEKey Type

hi def link BLCKKey DiagnosticError

hi def link MAYBKey Constant

hi def link Comment Comment

Looking at the supporting members, this appears to be supported by "both parties" across many many countries, what a sad thing to unite over...

But using pressure differentials is also sort of tautological. Lift IS the integral of the pressure on the surface, so saying that the pressure differentials cause lift is... true but unsatisfying. It's what makes the pressure difference appear that's truly interesting.

Funnily enough, as an undergraduate the first explanation for lift that you will receive uses Feynman's "dry water" (the Kutta condition for inviscid fluids). In my opinion, this explanation is also unsatisfying, as it's usually presented as a mere mathematical "convenience" imposed upon the flow to make it behave like real physics.

Some recent papers [1] are shedding light on generalizing the Kutta condition on non-sharp airfoils. In my opinion, the linked papers gives a way more mathematically and intuitively satisfying answer, but of course it requires some previous knowledge, and would be totally inappropriate as an answer by the AI.

Either way I feel that if the AI is a "pocket PhD" (or "pocket industry expert") it should at least give some pointers to the user on what to read next, using both classical and modern findings.

[1]: https://www.researchgate.net/publication/376503311_A_minimiz...

I'm going to guess the main take-away point is that the weights can be trained reliably if your transfer functions are sufficiently "stiff"? Not like you need the training for the operations presented, anyone could choose the weights manually, but it could maybe extend to more complex mathematical operations?

To be honest, it does feel a bit like Claude output (which the author states they used), reads convincingly "academic", but it seems like a drawn out tautology. For example, it's no surprise its precision is the same as floating point, as it's essentially carrying out the exact same operations on the CPU.

Please do correct me if I'm wrong! I've not read the cited paper on "Neural Arithmetic Logic Units", which may clear some stuff up.

IMO the difficulties are a bit off, if a 2D physics engine is 5/10, then orbital simulation cannot be 10/10 (even with the difficult stuff like J2 perturbation) ;)