This "symbolasm" looks exactly like what Blackfin did.
I think it's a very promising approach - it makes assembly a lot more intuitive, especially to people with background in higher level languages. It makes listings far more immediately readable, even without memorizing every instruction mnemonic. But it just came in too late. By the time Blackfin happened, coding in assembly was firmly an afterthought.
I'd like it to succeed, but I have my doubts that it will.
Blackfin is one of the things I looked at. I go quite a bit farther, with named variables, richer types (struct fields are named), block syntax, and generally capturing as much programmer intent as possible while maintaining an almost 1-1 mapping between statements and assembler instructions.
Time is a circle. Assembler could come back if it's an appealing target for LLMs. Currently it's not, but that's my hypothesis.
It is an appealing target for LLMs? Well, LLMs do make "handroll your hot paths in assembly for better performance" much more viable.
But that's a bit of a problem too?
Because, unlike humans, LLMs have no trouble whatsoever recalling all the mnemonics. Hell, they can often recall all the opcodes and encodings too and disassemble from hex "manually".
So if the primary advantage of this syntax is better human-readability and human-understandability? The need for it in a future where low level assembly routines are written by LLMs in a perf evaluation loop isn't that great.
We seriously need some "you own your hardware" laws - before the likes of Google tighten every bolt and implement enough "security" to firmly lock the users out of their own devices.
It's dead obvious that companies like Apple and Google and Sony and John Deere lock their hardware down to protect their revenue streams - not to help the users in any way. And yet, they're peddling the lie of "security" - and have been getting away with it for decades.
Now that AI is there to give even an average user a leg up when it comes to things like "I don't want to see ads in my TV's menus, can you do something about it", the companies will see more and more reasons to implement more "security" to stop it. And modern "security" measures are hard to defeat with technical measures - so this has to be curbed with legal action instead.
I don't think AI will save us from the ads. Eventually they will monetize AI by mixing ads into them if they aren't doing that already. Just like streaming got ads added into after being the escape from cable tv ads. And cable tv got ads added in after being the escape from broadcast tv ads. Its a cycle.
Stockholder value is now predicated entirely on things like recurring revenue, growth, and MBA short-term profit seeking. Enshittification of existing SaaS and 'smart' products is a feature, not a bug.
Unless something drastically changes in the US stock market, any publicly traded technology company of significant size is basically mandated to prostrate themselves to the altar of fiduciary responsibility to stockholders.
Or Driving? BMW had to pivot over initial outrage, but in 2022, they launched a "Functions on Demand" program that required drivers to pay a $18 monthly or $180 annual subscription to unlock physical heated seats and heated steering wheels.
Which will all have things like reprogrammable firmware and FPGA aspects to support paywalled proprietary features.
We already support it to some extent with licensed hardware codecs (HEVC, AVC, VVC etc...), licensed DRM like HDCP, and individual device licensing for audio technologies like Dolby Atmos, DTS:X and video technologies like Dolby Vision.
I’ve just realised: could shareholders sue for this? E.g. if I’ve bought the stock pre-enshittification, then the conpany decided to prioritize short-term profit, and then some time after the stock price falls (and that’s a big “if”, sure).
Embodiment for general purpose goals like "grounding, alignment, and AI safety" just seems to be one of those things that some people want to be true. There's no actual reason for it to be true. There are no findings showing it to be true. People just want it to be true anyway.
I'm very skeptical of that. Wishful thinking is a treacherous thing.
LLMs provide some incredibly solid "general purpose commonsense reasoning" primitives. And the likes of Astra, Opus 5.5 and Gemini 4 have demonstrated generational leaps in "spatial problem solving" - including robotics tasks. Those don't even have dedicated "sensor input embedding" and "action output diffusion" ends rolled into them at model level - but you can give them control over robot arms, and they'll have a decent chance of, slowly, finding a way to complete new tasks you'd previously would have to tune a robotics model for.
Previously, I have considered "generative video models" to be a decent candidate for the foundation of future robotics models - they, by necessity, perceive video, learn the kind of world dynamics required to plan and execute actions in real world environments. And they can begin to learn those dynamics off embodiment-free general purpose web-scale datasets.
But non-robotics LLM generalists have come to show they're no slouches either. They're not at all optimal for things like rapid precise manipulation. But they find more and more use in robotics pipelines anyway. Makes me wonder what something like "Astra reasoning engine but with robotics architectural extensions" could accomplish.
"Matching the dexterity" is a false goal. It's focusing on the wrong things.
As teleoperation readily shows, even a very dumb primitive hand can be salvaged if you get a very advanced control policy to drive it.
Sure, it would still take a lot longer for a well tuned robot with just a pair of UMI-style grippers to tie shoe laces than it would a practiced human. But a robot doesn't have to work faster than a human. "It can do the same task for cheaper" is often good enough.
The right thing to focus on is: AI. We already have robot bodies that are good enough to do many economically valuable tasks. We don't have the AI that would make those bodies perform to their potential. Modern robotics is an AI problem above all.
Human "sense of balance"? Human vestibular system is a sad joke if you compare it against any off the shelf smartphone IMU. Proprioreception is a joke when you compare it against current sensing in every motor and absolute encoders in every joint. Human hearing loses to a half-decent microphone array.
Even human vision, the one perception channel evolution has put an awful lot of weight on, is beat pretty resoundingly by the kinds of camera units that ship in smartphones nowadays.
The two senses robots don't have an edge in are: touch and smell. But I'd struggle to make the case that smell is critical at all, and touch? Hard to beat it directly. In practice, however, it's already partially compensated for.
Sure, a robot with a basic UMI-style gripper doesn't get rich tactile feedback - only a few channels of feedback from arm joints and finger motors. But you know what those actuated UMI gripper get that humans don't? Wrist cams. A robot can see the object it's manipulating - right near the point where the manipulation happens. That matters a lot for precise manipulation tasks.
Humans don't have wrist cameras not because they're undesirable, but because evolution sucks at wiring. We don't suck at wiring.
The fact that humans can pull off general purpose manipulation tasks with the bodies they have, and the best of robots can't? A testament to how much of a gap is there between modern robotics AIs and the evolved wiring of human brains. The intelligence edge is what makes the human body, flawed as it is, work as well as it does. Everything in the human body is half-assed and then salvaged by the brain in post. But that gap is closing as AI tech advances.
Your reply to my comment downstairs got chewed up by HN. Maybe some sort of LLM detector thought you're speaking too much like an AI?
The reason why I'm not "team sensors" is that we're nowhere near to being able to as much as squeeze out even one mid-range camera sensor for all its worth. Things like AI assisted scene reconstruction show just how much data is there in even a small series of photos - when the AI is there to back the raw measurements with sensible assumptions.
The difference between "a plausible reconstruction" and "an accurate measurement" is academical rather than practical. You never get perfectly accurate measurements (because real world sucks), so a well designed "plausible reconstruction" will often have better error averages AND bounds than a naive measurements-first pipeline - simply by taking better advantage of spatial and temporal redundancies. Adding more assumptions to the pipe from there gives you even better performance.
I expect approximately none of the near-term gains in robotics performance to come from the sensors themselves. There's still room for improvements there, mind. There is some cool sensor tech that could result in a better performance ceiling. But we're nowhere near that "sensor ceiling".
Today's robotics sensors are so far ahead of what robotics AI can reliably utilize that trying to push for "we need better sensors" is a mistake in nearly every case.
Human "dynamic range" is a hack - the limits look impressive, but they're only reachable through a lengthy adaptation process. Modern camera sensors with HDR, backside illumination and analog gain controls get pretty close to that - and much, much faster.
In practice, if you want really good performance across an extreme dynamic range, it's usually easier to stack multiple specialized sensors within a single sensor unit. A filter-free reduced resolution sensor for high QE in low light conditions (or at high framerates), and a daylight-optimized sensor that gets high resolution, full Bayer, IR-cut and a polarized glare filter to get you as much data as it can in high light environments. This gives you straight up superhuman range with no adaptation time.
Do the raw low light images look nice? No. But human brain itself takes a lot of postprocessing for "looks nice" to happen. Salvaging good images from series of noisy measurements is an AI problem. We've had some pretty capable sensors for a while now, but it took AI-powered computational photography to truly make low light photography happen on a smartphone.
I'm aware of the "detection of individual photons at above-chance rates" experiments with human vision. But modern backside illuminated CMOS, if plugged into the same setup, would also be able to detect photons at above-chance rates. Not to mention the actual photon event cameras. Very niche items, but they can detect photons pretty reliably - not merely "more likely to be right than wrong".
There are also thermal imagers and image intensifier stages, if you want to take the "niche extreme sensors" route. Not commonly done. But goes to show just how much ass engineered sensors can kick.
The limits aren't in the sensors - it's in making sense of the data they give you.
You're overstating how well wrist cams can compensate for a lack of touch. It's really hard to get the instant estimate of weight and rigidity we humans get when picking up an object with a wrist camera, especially for rigid end effectors. They can guide a grasp in ways that would require touch for a human and help detect that you've dropped an object but don't compensate for many aspects of grasp that prevent the drop in the first place.
"Estimate of weight and rigidity" is exactly the kind of thing that's performed rather well by low-dimensional sensing of actuators themselves.
Backdrivable actuators know how much current are they sinking - and thus, how much force are they facing. So even the simplest two finger gripper can easily "know" that one object is squishy and the other is not. The response curves are different. Same can apply to arm actuators for weight estimation.
Now, what this doesn't actually get you is an estimate of how hot/cold an object is, and how moist an object is.
Where cameras come in is specifically the "guide a grasp" part. Force sensing alone can tell a robot "the grasp is failing". A camera pointed at the grasped object can tell "why".
In theory yes but in practice backlash and stiction make drive currents much less useful than the raw force at point of contact for any realistic gearing. And an important part of weight estimation is that force lateral to the point of contact. Now, your arm motors will be drawing more current as they move so again in theory you can distinguish that, you've still got the weight of the whole gripper added in.
I spent years of my professional life on these topics[1] and from my experience motor drive current usually isn't enough to narrow down the weight of an object in a typical picking task even to within a factor of 2 to detect when you've accidentally picked up two items from a bin instead of one.
Oh, it gets extremely annoying when you try to use those kinds of indirect, noisy, smeared, nonlinear, relative measurements for absolute sensing.
The power of modern AI robotics is that, much like the biological mess it's trying to replicate, it doesn't care that much about absolute sensing.
Imperfect sensors fused by a high performance control model with a good foundation of world dynamics and self dynamics? Gets you impressively close to "perfect sensors" in practice. Being able to take a big bundle of noisy nonlinear ambiguous measurements, and implicitly reference them off each other and over time to get to a usable control loop is powerful.
It's the kind of power human nervous system wields to force the horrid mess that is the human body to perform.
For "detect when you've accidentally picked up two items from a bin instead of one", a wrist camera is a far more natural sensor than a bundle of actuators - especially for lightweight items. Wrist cameras are awesome. But even a bundle of actuators, if fused well enough, can often do it.
The problem is that designing algorithms to sensor fuse to this level by hand is pretty much a no-go. The kind of sensor wrangling hell I would only wish upon an enemy. Which gets us right back to general purpose robotics being an AI problem at its core.
Animal olfactory sensing is more or less just detecting the presence of certain molecules in the air.
We already have devices that do that better than humans and even ones that detect things we can’t smell (carbon monoxide, for example).
Though I suppose you could say that “smelling” is the result of the synthesis of the accumulated signals from the olfactory system in the brain, and more than merely sensing their presence.
Sure, olfactory sensing sensing is "just" sensing of volatile molecules. And we can build plenty of different molecular sensors. Including for things humans can't detect naturally.
But, just like being able to build a load cell and a thermometer doesn't translate to "skin-like sensing", we can't go from "chemical sensors" directly to smell.
For one, things like sensor diversity, density, longevity and manufacturability matter a whole lot. Can you give a robot a weather station type "pressure + temperature + humidity + PM etc" air quality sensor? Yes. Can you give that robot a sensor with enough molecular coverage and fidelity to replace the human sense of smell? Almost certainly not. Can you even get close on a sensible budget? Not with the tech we have.
Getting a system to perform molecular sensing with discrimination against many molecule classes - the kind of operation human smell performs - is very much not trivial.
Now, whether that's worth pursuing at all - I'm still leaning towards a "no". It's a very niche capability.
Sensors on industrial robots aren't THAT great. Robots people seem to use torque sensors on both sides of a motor, meaning they need all of input, output, and commanded torque to know the actual torque value that the software needs. Things like that. I think GP's point stands.
Very similar. Might even say it's the same thing viewed at a different angle.
We don't know the details of OpenAI's implementation, mind. But the "limited layerwise depth" problem was known for a while now, and both OpenAI's approach and this are getting at it. And that's far from being the first attempts to address it.
Training and stability were the main issues that stopped it from being adopted rapidly and widely the way something like MoE was. But I think it was always a matter of time.
It is a shame that nuclear power has stagnated - we could have been moving towards fossil fuel free grid half a century ago. We've lost decades of progress there. But it's a good thing that solar power is coming into its own more and more.
Where we lost decades of progress was governments pouring enormous amounts of R&D money into nuclear instead of renewables. We got a nuclear a head start that turned out to be a dead end. We could have gotten solar, wind and storage cost competitive with fossils decades earlier. And as a side effect kept the west competitive in energy tech.
I don't think fusion will happen, at least not in the "power too cheap to measure" dream of the 50s.
The problem with fusion is not technical (although we've not yet solved the technical challenges, which are considerable) but financial/economical.
Let's assume all the physics are sorted out later today. Now all we need is $100 billion to build a demo plant. Because we live in a real world, it will take 10 years before we can break ground.
Probably 5 more years of construction, testing, etc. So your $100b should start earning a return 15 years from now. If all goes well.
And returns should be pretty good. Assuming no great breakthroughs in storage tech between now and then.
Of course, returns will be seasonal, and mostly at night. We can expect solar to supply daytime demand. (Hey, solar is a form of fusion energy as well!)
So yeah, we need $100 billion now, perhaps more later as costs spiral, theres no prospect of a return anytime soon, actual returns are hard to forecast, and frankly life-span of the asset is hard to predict.
All of which is to say that it's unlikely to be privately funded. Govt maybe. Assuming the project survives multiple govt funding rounds. And, of course, being govt funded, timelines will drag and costs will overrun.
Unless of course, well, China. Cost is no object, experienced at getting major stuff built. Fewer delays. Money no object. But even there, why choose fusion when solar is so cheap? What strategic advantage does it hold?
Betting on fusion today means betting that it gets to production faster than storage. Indeed that it beats storage hy 30 years. And that's some bet to make.
Similarly, battery storage won't happen. At least, not in the "power is basically free thanks to solar" dream of the 70s.
The problem with solar is not technical (although we've not yet solved the technical challenges, which are considerable) but financial/economical.
Let's assume all the physics are sorted out later today. Now we need $500 billion to build 15 mines around the world for whatever new magical element we need in massive amounts. Because we live in the real world, this will be fraught with global pushback, because finding the political support today for creating insanely massive mines in favor of hopefully one day breaking even on carbon neutrality is a fool's errand.
Probably 25 more years of litigation, and you should start getting access to mines which would allow you to pollute to insane degrees so we can fight climate change in a century.
Yeah, nuclear is fine. It just at this point it can’t really compete with solar+battery for how quickly you can bring things online. The 10 or 15 years required to build a nuclear plant is pretty devastating for it when you can build up solar in 1-2 years instead.
It also can't compete on cost. Nuclear is crazy expensive and prone to cost overruns, so no market party is willing to build it without the government taking on all of the financial risks. And with the rapidly evolving power landscape the long payback time only makes things worse.
It’s not even that it’s overregulated but that the amount of power it brings online at a time is just too big.
Individual wind turbines aren’t very big, solar panels get laughably small, and while they are not efficient at their small form factors they are very easy to scale up and down. Nuclear is the exact opposite, representing a very large single point of failure, and even under pre-renewables hard to absorb; some of the first pumped hydro was developed to take excess nuclear capacity because it cannot scale up or down very easily.
Its not just matter of regulation if it was that easy to build China would be building 100s. And by saying China I am referencing their manufacturing capacity and them being a command economy that does not have the same financial profitability concerns. It is not like they have not tried they have but nuclear technology the world has currently have does not translate to mass manufacturing.
Agreed. It’s very cheap to mass produce solar panels. The components that go into nuclear power stations must have a very high precision and low failure rate.
The cost to produce such components that must safely operate in very demanding conditions is prohibitively expensive in comparison to producing a solar panel.
Solar panels will always win because you’ll never have the required demand to mass produce nuclear power station components.
Nuclear has an extensive startup cost, which makes it an extremely walled garden. This is the only place where it's prone to overruns. Once it's built, of course, it's by far the most efficient, most reliable, and most secure.
You can also bypass most project overrun concerns if you use already-approved reactors instead of trying to use a brand-new one, but of course this pushes back the schedule for profitability.
It's probably worth pointing out that 90% of global uranium supply comes from 4 countries.
Given the current instability of oil prices, where considerably more than 4 countries are producers, it's perhaps dangerous to bet the farm on future fuel supply.
Edit: for reference each reactor consumes about 140 tons of uranium each year, and current production is lower than demand by about 10%.
So if there's a growth in nuclear reactors (which there is, in China alone) I guess we're hoping for an equivalent growth in uranium supply?
Do yeah, you're not wrong. Currently nuclear, once built, is very reliable. But equally it has a "single supplier" problem as well.
Nuclear isn't uncompetitive due to some inherent technological reason. It's uncompetitive due to overregulation and gold plating.
Solar used to be completely uncompetitive, but improvements in technology and increases in scale made the economics work. Nuclear got kneecapped before that could have a chance of happening.
If you read the literature analysing historic nuclear costs around the world, this just isn't true. Nuclear reactors are complex 'mega-projects' requiring coordination across a vast number of actors, specialist work under the constraints of a construction sites, and which - because of lower numbers of reactors, changing and adapted design, diffuse consortia - have failed to push down costs through learning-by-doing. A study just over a decade ago found that ~98% of nuclear projects involve schedule overruns. Because nuclear projects involve extreme capital costs, this means: (i) vast sums have to be paid in interest over the build time; (ii) states/taxpayers/consumers usually bear some of the risk.
Compare a nuclear project to a solar panel. Solar is cheap, simple, commoditised, flexible, and is riding on almost miraculous learning curves. If dispatchable energy can be solved, beyond just gas/coal plants, ideally with long-duration storage based on the conversion of electricity, then this is going to be an infinitely better option. It's also just not clear what role nuclear plants have given that renewables need highly dispatchable power to ramp up and down according to their variable output, not baseload nuclear power.
> It's uncompetitive due to overregulation and gold plating.
This is an oft repeated point by pro nuclear folk, but it’s not true. There are many regulatory regimes all over the world, and new nuclear is competitive in effectively none of them. Even in single party state china with few regulations , nukes are a declining share of total power every year.
They’re just giant big complex pieces of machinery. With huge ongoing security, insurance and deferred decommissioning costs.
> Nuclear isn't uncompetitive due to some inherent technological reason. It's uncompetitive due to overregulation and gold plating.
That's the go-to nuclear apologist argument, but it doesn't hold water. Nuclear is struggling globally, even in China. That points to inherent problems. Regulation also doesn't explain why costs have been so widely underestimated. That's a function of dishonesty of the cost projections, not of regulation.
I'll add that regulation is the price you pay for implicit subsidy like government insurance against accidents. You don't get the latter without the former.
Germany has been spending a lot recently to build new power generation. Of course it’s expensive.
France is now on ancient fully depreciated nuclear reactors that are soon in dire need of replacement. Of course it’s cheap. But they have huge investments ahead of them to replace, upgrade, refurbish, etc. They have identified that that need to expand renewable energy to 50% of power generation, at least.
We’ve also seen recently that Germany has had to save France during hot summers when their reactors have had to shut down.
What’s the lesson from France? Build lots of nuclear all at once many decades ago when labor was cheap and then stop investing and letting your infrastructure rot, and you’ll have cheap power for some limited time? Is that something anyone can or should copy today? No.
Now, even VALCOE (value-adjusted LCOE) of wind and solar is coming down well below all other fossil and nuclear alternatives. So how can you say betting on renewables was the wrong at this time? It has been expensive to transition but long term it’s drastically cheaper than everything else. And we haven’t even started getting the benefits of how repowering old renewable power plants is significantly cheaper than building new ones.
You're reading from the playbook of bad nuclear bro arguments.
The implication you're trying to draw there is a non sequitur. That situation doesn't contradict what I said. Moreover, France is going to be dragged kicking and screaming into the renewable age because they can't afford to subsidize nuclear builds enough to make a nuclear approach work.
It's possible that this "phrase hash" method was used for a different case - like another, unrelated and possibly unreleased algorithm. And then it was confused for the one used for those curves.
I know for sure that Qualcomm LTE/5G modems are also SDRs under the hood.
Like Wi-Fi chips, they don't use SDR mode during most of the normal operation. They have dedicated hardware units that do a lot of the decimation, conversion, modulation and demodulation efficiently. But the "SDR mode" is still there for things like factory calibration.
Makes me wonder how many other common MCUs and SoCs are just waiting for someone to unlock SDR in them.
TX is less "not possible" and more "this dev doesn't want to enable it" from what I gathered. Probably for "I don't want people using this to shit all over the spectrum" kinds of reasons. So you can get TX - it'll just require extra hacking.
Yes they are. And they have a much wider tuning range (I have one that tunes from 240 MHz to 7 GHz). Problem of standard hardware is that every path has dedicated SAW/BAW bandpass filters that yeet anything not in uplink/downlink. And you need an unfused chip for custom Hexagon DSP stuff (so no phones).
Their DSP/SoC silicon includes a wideband transceiver? But phone chips have external bandpass filters? Interesting. Mostly I've seen baseband processor + external transceiver (wideband or not) with its own filters.
I know the LTE modems had ADCs and DACs in SoC - ZIF with analog signal running straight between the tuner and the SoC. But 5G systems rolled ADC/DAC into the transceiver chip itself - with only the digital processing living in SoC.
Is the "QLink" setup they use in 5G systems now just a thin protocol running over LVDS and streaming I/Q between the transceiver and SoC DSP as digital signals?
Never actually went that far, but I did wonder if you could plug a QLink tuner chip into an FPGA to use it as a broadband SDR frontend.
There are also chips like NXP layerscape which do the sampling and hardware acceleration on a chip which streams to host (though still need another transceiver).
The motivation for this is of course beamforming, angle of arrival, spatial streaming, etc. You connect multiple antennas with different views back to the base station.
Yeah, I've worked on such chips where there is essentially a DSP with extra Wi-Fi specific hardware blocks on a shared memory bus. The DSP runs its standard assembly language and makes calls to the blocks, which are themselves configured with config registers or even a kind of microcode style assembly.
Those hardware blocks could be useful for certain applications but most importantly for triggering captures only when there are frames to process and giving you both the frame + the sample.
So for me the perfect hack would allow both normal Wi-Fi tx/rx and triggered or continuous rx sampling.
When the solution becomes "you need a nanometer precision stepper and an electron beam to modify the SoC's circuitry and dismantle the 'root of trust' that's making your hardware betray you to the corporate interests", you've lost.
We can't rely on hardware staying hackable. We couldn't ever since "secure boot" became a thing.
I think it's a very promising approach - it makes assembly a lot more intuitive, especially to people with background in higher level languages. It makes listings far more immediately readable, even without memorizing every instruction mnemonic. But it just came in too late. By the time Blackfin happened, coding in assembly was firmly an afterthought.
I'd like it to succeed, but I have my doubts that it will.
reply