davemp

No one is going to argue against those vague statements. He obviously does not agree with people that put more weight on making things compile errors instead of runtime errors

> Compile errors are better than runtime crashes

I don’t think you’re being fair here.

Looks like Mr. Kelly agrees with you. From the zen of zig:

- Runtime crashes are better than bugs.

- Compile errors are better than runtime crashes.

- Incremental improvements.

- Avoid local maximums.

Considering every memory safe language that I’m aware of panics on index out of bounds, there’s spectrum here.

Personally, I’d like to see refinement types that allow people to elide the runtime overhead/panics. Though that may be too much complexity to stomach.

Without explicitly entering the language holy wars. I think there’s an interesting trade-off here.

Static analysis based approaches restrict what is possible to express in the language (See: doubly linked lists). But can have lower dynamic overhead.

Ways to work around expressiveness limitations include strategies such as using indices into an array instead of pointers. Which in turn incurs the dynamic overhead from bounds checking.

Maybe zig could find a way to elide enough of the capability overhead that careful programming could get a <2x penalty vs C. Maybe explicit opt-in built in types or something. This would essentially be the reverse of unsafe blocks, but then there’s no escape hatch.

Another key pro to capabilities is that you can actually have a memory safe interface without having to leak lifetime and other info across boundaries. This would be great for compile times and places where open source is not tenable.

Regardless, I’m very happy to see another independent school of thought approach memory safety. I’ll forgive Mr. Kelly any gruffness, these topics can be quite annoying to discuss in public.

I don’t think people are arguing to stop researching AGI. Moreso against sales people trying to use the concept of AGI to sell products that are very much not AGI. Or devoting so many of our resources into such a pursuit that it causes harm to real people.

This is obviously complicated by the fact that LLMs/Agents are useful by themselves, but that’s not really the topic at hand.

clueless and stubborn and confidently wrong

Uhuh. I really shouldn’t be replying to this type of comment from a throwaway.

But the extremely powerful semantic search that we get from LLMs isn’t enough. I don’t think anyone is credibly arguing otherwise?

Agents already are a layer on top trying to bridge the gap. But they’re really just using LLMs as a heuristic to explore extremely NP problem spaces. The notable successes with agents so far are when we can provide them with a solid verifier and preferably additional context hints on the steps to take in the problem space. See the test oracle problem on where this gets us.

So forgive me if I think that it would be enough of a jump in computational complexity to remove those guard rails that it’s not feasible. But don’t say that I’m clueless, stubborn, or confidently wrong.

There is a vast gulf between theoretically possible and technologically feasible.

If you can’t provide a realistic path to achieve something, you’re asking people to believe in science fiction.

You could tell me that a rock’s molecules are comprised of protons, neutrons, and electrons. Blood is also entirely protons, neutrons, and electrons; so theoretically, one could rearrange stone into blood. But without an actual method to do so, it sounds like you’re telling me that you can squeeze blood from a stone.

the human brain exists, it is not made of magic, it can reproduced

Yeah. It only takes 9 months and ~18 years of training…

But saying that the human brain cognitive capabilities cannot be reproduced on other types of substrates is stupid at this point

Let’s be clear. Everyone is talking about silicon transistors here. That’s what we’ve got.

Digital computers have real limits. Sensors and other sources of training data have real limitations. It’s not clear that we can organize them in a way to reproduce organic brains.

Totally insane that they enshrine this stuff in law without a reference design and corresponding data to validate its effectiveness.

Yeah I think that’s the difference. On the east coast in July, the big safety risk with fireworks is generally to your own person.

EDA tools have been using ML techniques for >10yrs afaik

It definitely isn’t simple, but it’s a pretty well trod path. If the FCC or state equivalent doesn’t have folks who can write the spec that’s a huge problem. I would be surprised if an existing spec doesn’t already exist that just needs to be applied to this scenario.

The streamers should be responsible for the signal. If the device front end has crazy frequency response or the backend does weird DSP tricks, that’s on the device manufacturers.

Article reeks of LLM mannerisms in a pretty obnoxious way. I read it anyways and surprise surprise, it boils down to:

- Ridiculously NP problem space that we guess and check with ML heuristics.

- Problem: not enough training data

Cool enough work, but the LLM generated fluff in the article really takes away from it.

Picking up on some dunning kruger effect here.

Programming isn’t even a field in the same way as prob&stats. Computer science does in fact have non-deterministic sub fields such as information theory.

Just because it could be worse doesn’t mean it couldn’t be better.

Would you want to take a pay cut if your employer was having financial problems?

This happens constantly in the form of layoffs…

Ah yes. Just need to displace 16 million people’s livelihoods. No big deal.

I assume because most windows installs are corporate IT garbage that if anyone cared about performance they could just turn off one of the three endpoint protection services or tune the backup service down and get better results than processor upgrades.

One third of all software code is written by AI.

I find it interesting that using lines of code as a metric is making a comeback.

I’m pretty certain there are physiological limits that you can’t just muscle through and stress _can_ be an indicator that you’re reaching said limits.

usage != positive opinion

I don’t like driving in traffic yet I do it pretty much every day. Why don’t I simply not drive?

Kind of goes to show how out of touch and insular the tech exec sphere can be. Almost everyone I interact with in reality has a deep distain for LLMs and their touted trajectory.

It’s not a public study though. We’re not going to get trust worthy numbers about labor or token cost.

So they can get training data I assume.

I think it’s becoming hard to ignore that the Internet has fundamental flaws from a game theoretical view. I hope that we can skip the step of having Google as the feudal lord who saves us from anarchy though.

How about we start with some accountability for entities that host fraud? The main reason we can have relative anonymity in public is part trust and partially because you can get physically taken out if you cross the line. I understand there are some real limitations with enforcing accountability on the Internet, but perhaps that’s where we should be focusing.

The FCC or whoever is almost 100% just looking at power/time/location. Those 10 devices will look like 1 device.

What -apart from cost- prevents a user who wants more bandwidth from installing 10 devices in parallel and alternate each radio so none of the radios exceed their allowed transmit duty cycle?

Folks with badges knocking on the user’s door. It is pretty trivial to locate stationary signals.

I’m not sure how you could suggest starbucks and mcdonalds outcompete based on product quality or ambiance. Have you been to a random mom&pop coffee or burger spot recently?

Chairs aren’t magic, we know roughly what’s good or not. You don’t need a PhD to choose a good chair. A mom and pop could just copy the chair too.

The chains outcompete on marketing, leverage with vendors, etc.

They put a huge amount of effort into baseline quality because it’s incredibly hard to pull off across 1000s of stores staffed by people who don’t have any incentive to care.

I'll start of by saying I really hate C (also love it), and welcome improvements; but I have a few criticisms:

- Sensor agent is such a rancid name for a remote sensor that I feel a need to public say so. Please don't use marketing names for things that already have more descriptive names.

- Rust uses a full RTOS and C uses the mediocre ST HAL (vendor specific). Immediately apples to oranges. Also I've never heard of the C JSON library and it looks sketchy at a glance so that will also hurt the comparison.

- Streaming slow sensor data with a 160MHz 786KB/2MB MCU is not a good test in the slightest. You could probably use something like micro python here and be done. No one is reaching for bare metal C here. Also no one serious about performance is using JSON serdes. If you're using bare metal C, you're likely trying to push the limits of your hardware or doing something so simple that you won't be tempted to reach for terrible third party libraries.

- Does the Rust code base use the 'unsafe' keyword anywhere, including the RTOS? If so, it's not memory safe without additional formal verification.

Overall I'd say this paper has approximately zero value wrt its stated goal of comparison.

Good for you? I’m glad you have languages that fit your needs.

In the realtime/high assurance systems world, where garbage collection can be a huge source of non-determinism and overhead; we don’t have great options.

Zig is really the only language (idk about Odin?) trying to take the same approach that C did in giving you absolute control over a minimally abstracted CPU model. Us folks who need/want maximum control/performance should be allowed to have nice things too.

It's not by writing syntax that you get there. It's by creating software and gaining the experience of seeing it go wrong.

I’m not sure there’s any data to support this statement. Maybe if you hedge with “not by syntax alone”.

I am aware of studies that showed the act of physical writing letters improves retention.

I would not be surprised if this extended to physically typing out your curly braces, naming functions, etc.

The argument that training AI on scraped data is "fair use" and the resulting model outputs are "transformative works" has held up in courts.

Nope. Nope. Nope. That has explicitly not been ruled on yet. Transformative means that you don't need a fair use defense. Anthropic has only gotten away with their outputs being called transformative so far because they put a dubiously effective filter in front to block the most egregious infringing outputs. No one has actually challenged this afaik.