datameta

SEEKING WORK | USA REMOTE | NYC HYBRID

  Languages: C/C++, Python, JavaScript, Kotlin, Dart, VHDL, SystemVerilog
  Embedded Systems: PowerPC, ESP32, STM32, iCE40, ECP5
  ML: TensorFlow Lite Micro, PyTorch, ONNX, OpenCV
  IoT: BLE, WiFi, MQTT, LoRa

Hi, I'm an embedded systems engineer/architect bridging the full compute spectrum of ultra-low power IoT, edge systems, and HPC.

My latest projects have been in full-stack Edge R&D for SLaM, memory controller firmware dev for POWER servers, and chip bringup/testing for Z mainframes.

I'm passionate about modular synthesis and tinkering with all sorts of analog circuits and DSP. Currently upskilling in analog PCB design and refining my abilities in FPGA development.

Services I offer: custom firmware implementation, hardware/software integration, full-stack edge AI architecture.

Feel free to reach out directly at dm@datameta.info

> generalizability doesn't come from multi-modality but by scaling a single modality itself

Could you expand on what you mean by this?

But are great as part of a cannabis strain profile!

I believe it is cost per unit density that allows magnetic tape to be more suitable for archival storage than HDD, which surpasses it in I/O speeds. Magnetic tape storage is still usually in RAID-like configurations and while tape only survives for 15-30 years, data survival by migration is typically done well before medium degradation.

From what I understand, alignment and interpretability were rewarded as part of the optimization function. I think it is prudent that we bake in these "guardrails" early on.

I think this is interesting enough for a post in and of itself: https://arxiv.org/abs/2505.22954

Can a lawyer offer their civilian opinion as to which supercedes/governs?

If students' optimization function is getting hired, what happens to the quality of skill and knowledge of grads?

As far as we know there is no direct competitor in learning to doing it repeatedly, in different ways. Now whether a mixture of structured and truly unstructured learning is superior is another discussion.

Personally, when I find myself fully aware of acting against my best interest (especially repeatedly), "stupid" doesn't seem to be totally out of place.

I don't find the author's argument, that students who essentially skip learning via LLM could avoid fairly being labeled "stupid", particularly convincing.

Not that I think it to be a particularly useful label, but I don't find awareness of self-sabotage to preclude one from such a label.

This is similar to how I believe the label "smart" alone does not carry much use.

You mean corporations regulating the government? Oh wait...

I can't tell now if you're being intentionally condescending.

Ah, this is what happens when one dredges up old knowledge and doesn't seek an updated understanding before sharing. Thanks for the correction on the colloquial classification.

I suppose it was pointless of me to ask if what I received was a wiki link, no offense. Thanks.

I love this example! The unforeseen synergies of different methods attempting the same outcome cannot be overstated enough in my opinion. Each strategy has its own diminishing returns, and stacking them covers gaps each has.

I think a great illustration of this today are SLaM methods that almost always seem to combine a low-drift high-noise technique with one that is high-drift and low-noise.

I did not mean to imply that steel quality has in any way continued to suffer. He was talking about the process of the early 20st century in comparison to the late 19th when steelmaking was handled more end-to-end from maker to user.

I am talking about enshittification of digital applications and services compared to the earlier years of the information age.

And the strings of the guitar are an intersection of both interpretations! There are several core wire alloys used and many different kinds of winding wire to achieve different tone and magnetic response.

Fun fact, a common older winding alloy for acoustic guitars was called 80/20 Bronze despite actually being copper/zinc alloy and therefore Brass!

Is that because we see no way to bootstrap equipment down so many magnitudes of scale, not even close - or is it something else?

Where could you recommend I read more about how they are derived scales?

The full read particularly resonated with me from the perspective of a metaphor to today's rapidly changing engineering practices.

> “Time was,” he lamented later, “when a man made steel, decided what it was good for and told the customer how to make the best of it. Then, with time’s quickening step, he just made the steel; he engaged another man, who knew nothing about steelmaking, to analyse it, and say what it was good for. Then he engaged a second man, who knew all about hardening and tempering steel; then a third man who could neither make steel, nor analyse it, nor harden and temper it—but this last tested it, put his OK mark on it and passed it into service.”

In a way it warms my heart that obfuscation in such a manner, perhaps even enshittification, is not a new experience for those watching a trade modernise. Several things come to mind: npm and python library dependency hell and LLM as a catalyst of skill atrophy in experts simultaneously with the enablement of a whole new middle layer of proprietors.

Also the article led me to think more about the idea of simultaneous invention - I used to believe it to be redundant and wasted work. But this still can lead to different outcomes even with identical formulations or methods. Getting an invention into use in the world is perhaps as great a feat as the invention itself. I now believe any worthwhile invention deserves more than one champion.

In today's hyperconnected world it is easy to discover that someone else has beaten you to the full flourishing of idea into invention, but I find that doing the novel work oneself with one's own mind and hands still provides the unique learning opportunity which can allow one to invent yet again, albeit now with a widened skill and knowledge horizon.

My naive notion on this is yes, iff the new method is computationally or memory-wise of lower complexity

Infinity, aka 2 or more. I agree that those are truly three distinct classes of quantity/identity

To what extent are the Planck length and Planck second confirmed smallest discrete units?

Ah, makes sense! Thanks

Certainly it can make you a 10x video releaser, but not a 10x Youtuber.

I don't understand the reasoning behind that besides market exclusivity.

Only insofar as both building types are recognized externally as inextricably linked to the culture, right? Sauna is deeply rooted in Finnish culture but not quite to the level or multipurpose use of ger.

Does every other form of human communication have an analog in digital systems that has both the capability to be "better" while also putting a lot of resources into modeling the relatively nonlinear human version?

Yikes, you've just perfectly articulated a trajectory that I've been using subconsciously as one of the primary reasons why I want to keep my coding craft sharp.

Human speech evolved with biological constraints and through neurological adaptions to emit and understand the nonlinear output that has lexically fuzzy areas to the untrained ear. So I think it's a rather "lossy" analog to digital conversion because the computer is simulating understanding of a form of information transfer that it itself is not constrained by (digital systems don't have vocal cords and could transmit anything).

Seems ethically sound to me.