These policy tradeoffs are interesting (and tricky) - if you have a huge number of different parties that are unorganized yet impacted by a policy, how do you ensure you have adequate representation from them when compared to a single well-organized party?
> The Petition recognizes that there currently are unlicensed part 15 devices operating in the Lower 900 MHz Band, but it is unclear regarding the extent to which the proposed reconfiguration would impact potentially millions of such devices. With respect to part 15 devices, NextNav states that it is completing technical analyses and “will work with unlicensed users to understand their spectrum requirements.” Id.
> NextNav does, however, seek the removal of the current requirement that it not cause unacceptable levels of interference to part 15 devices. See Petition at A-6 (proposing to amend § 90.361), A-11 (proposing to add § 90.1410(c)).
Requesting public comment is perhaps better than nothing (and likely better than just allowing lobbyists to influence policy behind closed doors), but it's a hard problem to quantify whether the collected comments are representative when one side is more heavily resourced and organized than the other?
Is it even possible to "work with unlicensed users to understand their spectrum requirements" in a way that doesn't ignore a potentially substantial long-tail of varied usage?
The thing that's really legacy in GitHub is the code review flow, which is likely one of the surface areas that corporate customers use the most.
Actions and Codespaces have been huge and transformative, so it's disappointing that core functionality like code reviews hasn't seen the same level of improvement.
Basic features that Phabricator had nearly a decade ago are still missing:
- highlighting copied/moved code blocks
- gutter indicators for code coverage
- stacked diffs
- not collapsing large changes by default (what a ridiculous default! "A lot of code has changed here, meaning it's likely bugs may be here - I'll hide that for you to make your review easier!")
And I don't recall how well Phabricator supported it, but handling rebases sanely by carrying over comments and showing diffs from prior PR versions across rebases would be amazing. The number of times I have to re-review an entire PR in GitHub because it can't show what changed since the last review if the author rebases their branch...
To be fair, they did improve several things in the PR flow over the last decade but there's definitely a lot more that could be done. I just hope they keep their API team well-resourced so that code review tools like Graphite and Reviewable can keep working! (Disclaimer: I'm the founder of Reviewable.)
This is 100% true. For some reason GitHub has put very little love into code review even though the Pull Request is probably their most important flow.
The good news is that many people have built better code review interfaces on top of GitHub. My favorites:
- CodeApprove (I created it, so yeah I like it)
- Graphite
- Reviewable
- GitContext
Check them out! You’d be surprised how much better they are and how quick they are to set up.
This is absolutely false in California, please don't spread dangerous misinformation.
See CVC 22400(a):
No person shall drive upon a highway at such a slow speed as to impede or block the normal and reasonable movement of traffic unless the reduced speed is necessary for safe operation, becauseof a grade, or in compliance with law.
No person shall bring a vehicle to a complete stop upon a highway so as to impede or block the normal and reasonable movement of traffic unless the stop is necessary for safe operation or in compliance with law.
The above was referencing stopping on a city street ("Powell between Bush and Sutter"). You're talking about stopping on a highway. These things are not particularly comparable.
In the California Vehicle Code section 360, a "highway" is defined for the purposes of the vehicle code as "a way or place of whatever nature, publicly maintained and open to the use of the public for purposes of vehicular travel. Highway includes street." [0]
Succinctly stated and something that resonates strongly with me.
In the last internet revolution (web search), results started high quality because the inputs were high quality - bloggers and others just wanted to document and share knowledge. But over time, many interests (largely commercial) figured out how to game the system with SEO, and quality of search results has decreased as search's incentive structure led to lower quality data being indexed.
We're at the start of the LLM revolution now - models are trained on high quality inputs (which may be as rare as "low-background steel" in the future). But the models allow the mass production of lower quality outputs with errors and hallucinations; once those get fed back into new models, are we doomed to decreasing effectiveness of LLMs, just as we've seen with search? Will there be LLMO (LLM optimization) to try to get your commercial interests reflected in the next generation models?
I think we've got a few golden years of high quality LLMs before that negative feedback loop really starts to hurt like it did in search.
I doubt it's going to go the same way as search. You can't run Google on consumer hardware, but you can run LLMs locally.
At worse, newer models will get worse and you can just stick to older models.
You could also argue that proprietary models gated by an API are better than anything you can run locally, and yeah maybe those will get worse with time.
They're not going to get any worse than what you can run locally though. If they do, open models will overtake them, and then we'd be in a better position overall.
> You can't run Google on consumer hardware, but you can run LLMs locally.
You can't run an up to date model locally. When I ask Googles models they have knowledge from stuff just a week ago, without using search. You wont get that from a giant local model.
Unlike the low background radiation steel, high quality content will still continue to be created. Arguably even at a faster rate with these powerful tools.
The proportion will drop ofcourse but that only means curation will become king. With search this curation was difficult, because the value was so low you could only do it profitably if it was completely automatic, and that was difficult.
An optimistic take would be that LLMs will make curation so much more valuable, that it will be done much better. If the wider world gets to use this curation to limit spam, and highlight good work, it would be amazing for the world.
You're forgetting that we have mechanisms in place to curate curation and farm attention toward the crap, not toward quality. It will be harder and harder to break the surface tension of this gooey gelatin wrapper we've placed over creative activity.
People will pay money for curation towards quality, probably quite a bit if the rest of the landscape is 99.99% noise. They won't pay much money for curation "toward the crap".
But that will be reputation-based, and names can be sold. Look at the brand-holding conglomerates we have today that don't make any of the original product, they just license the name out to anyone, regardless of the quality of the finished good. From mattresses to magazines, we keep seeing this. Why wouldn't we see this in curation sites?
Sure. I suspect it might be hard to actually recruit people like this considering how the people who got sampled are hugely pissed off at AI firms rn, but money will go a long way here.
I don't think it impacts AGI timelines much. Worst case scenario, we just cut off training at 2024 data. But it's not like another ~5 years of internet data is the magic that will get us over the finish line to AGI. We should have enough data already. We will get to AGI with synthetic data (e.g. AlphaGeometry for code, or simulators e.g. RL inside unreal engine), world model from video, algorithm improvements, and 100x more compute.
I'd add that majority of LLM-generated stuff put onto the internet isn't pure junk. It's endowed with human-created context and curation. If someone submits some LLM-created code to Github, it's because it works. A small fraction is pure noise (e.g. state-actors spamming Twitter) but that should remain a minority.
I think it’s clear that LLMs cannot be the end state of this technology, and we will need systems that can reason and develop hypotheses and test them internally. These systems may benefit from more curated datasets (such as those collected before the bullshit wave began) along with real world interaction data from YouTube and robotics. Such systems could eventually be used to rank web pages for their bullshit level, which of course would present a risk of censorship but when used properly could lead to insightful data handling.
It’s just really clear that a giant text averaging machine can only go so far, and while we do see some higher level emergent properties, we’re going to have to move beyond the current state of the art in the next decade, and such future systems may be much less affected by the internet’s bullshit. Even without the bullshit singularity I think such measures would be a necessity and we would see them developed soon.
> It’s just really clear that a giant text averaging machine can only go so far
It's not really a text averaging machine, it's a pattern matching machine.
Right now the "depth" of the patterns it can match can only go so far, but in a few years with more advances in chips and memory the depth is going to increase and the patterns it can match will fan out accordingly.
it is a statistical model. If everyone is saying X and is wrong, and one guy says Y and is right, the llm will bit out X. Because that is the most probable thing in the dataset. It literally is a text averaging machine
Not just real world data from videos. AI models need feedback from many sources: humans, code execution, web search, simulations, games, robotics, math verification, or from actual experiments in the real world. All of these are environments that can take the output of a model and do some processing and return feedback. The model can learn and search for solutions, creating its own training data as a RL agent.
Since all deployed models produce some kind of effect and feedback from the world there is an opportunity there to collect data targeted on the current level of the model, the most useful kind of data. That's why I think AI will be ok even with the proliferation of bots online. It's not 100% pure synthetic data in a loop, it is a agent-environment loop.
tl;dr Models learn better from their own experiences, not ours.
But bullshit hallucinogenic output and fakery at colossal scales doesn't just pollute the pool of static information available on the web. It also warps the minds of the humans you're relying on to verify reality on the next training loop. Not only that, but we can also see the emergent breed of humans who believe that writing is similar to arithmetic - an unnecessary skill that can be handed off to a calculator. Or that making a movie shouldn't require knowing anything besides asking for what you want to see. How is someone like that - someone who wants to rely on bots - going to tell a bot what is or isn't true? How can they even have a pre-pollution baseline understanding of reality?
I just finished rereading Do Androids Dream for the first time in 20 years or so, and was astonished at how similar his andys really are to LLMs in the polluted / destroyed reality there. How confusing and corrupting they are to organic life. PKD describes the androif brains as neural networks with thousands of layered pathways and trillions of weighted parameters, and it's as if he was able to accurately conceive of what linguistic and "emotional" strengths and weaknesses those constructs would actually have, decades before LLMs existed. And there's this one amazing line where Deckard calls them "Life thieves". What else should we call what Sam Altman and others are building - but theft of human ingenuity, creativity, and basic reason for living, and the utter annihilation and suppression of people like this 16 year old kid who dare to hope they can contribute something more original in life than being a servant of a tech company building this shit, or a tiktok influencer who writes prompts?
What should that kid hope: That their work becomes noticeable enough to be immediately stolen and their name turned into a prompt?
Life thieves.
Just as a further aside, I had dinner tonight with a friend who's a fairly famous animator in the commercial realm, and I brought up this post. He's just sure the kid's screwed and the genie is out and creative is basically over. He's turning to building wooden clocks.
But his reaction, and the reactions I see here every time this comes up, remind me of something else. They remind me of how people react when someone is robbed. Everyone has some reason why it's bad but not that bad, it was inevitable, it'll be okay, etc. Or they go around wondering what they're going to do now. Or they paper it over with optimism. Surprisingly few people get robbed and are willing to realize they were robbed, and become wildly pissed off about it. Most people have some sort of flight reaction, as evidenced by e.g. the promotion of "prompt writing" or learning to use a paid API to do what was formerly your own creative job that now spits your own work back at you.
I say sue the shit out of all these content thieves.
The comparison is only valid if AI ends up being monopolized. A continuously evolving ecosystem on the other hand has a better chance of adapting to those pressures. I am sure there are search engines that don’t index the SEO crap, but I don’t remember the names, and they have other flaws.
Open source AI needs to get a lot of investment for this to be mitigated. Relying on market incentives to drive development without the possibility of forking is too dangerous given the high stakes.
This comment/sentiment/idea has already been thoroughly discussed, shared, retweeted, posted, and echoed a couple thousand times online already. Humans like you have been redundantly saying the same things in varying degrees of novelty, like this, for centuries.
That's what the topic of the post we are discussing is. He is basically reiterating the point. Which is a good point. And there have been zero good arguments made (as far as I'm aware) to why this fear would be unfounded.
I'm pointing out the irony. The Internet is already saturated with us continually metaphorically JPG compressing original ideas into textual redundancy, with many people having only been exposed to those copies of copies.
I guess some of it might depend on how good the AI-generated content gets, and also how good the AI gets at detecting AI-generated content.
If the AI was good at detecting it, it wouldn't matter if the AI-generated content sucked, yes?
Even a low probability of detection would help. Let's say our algorithm is 50% likely to detect AI junk. That means that half the junk data won't make it in to model. Even 20% would probably be worthwhile, especially if it also threw away human-generated junk (and let's be realistic here: there is, and always has been, no shortage of terrible and/or wrong human-generated content).
Let's say those crappy filler paragraphs that get stuck between pictures on meme clickbait pages... I suspect most of those people have already been replaced, but that prose was horrible long before the current LLM boom.
I suspect there is a lot of effort being expended right now on ways to ensure the training data (whether human or AI generated) isn't shite.
Humans are smarter than machines. We "sense" bullshit in patterns that machines cannot yet intuit. I can tell when something's AI generated. It has a smell or a flavor that's different from human work and it's a variable pattern because, like a southern accent, I can tell if it's Georgia or Alabama, Gemini or Claud, even if very capable machines cannot.
I've been daily driving FF Android for a few years now and I've had the opposite experience: the vast majority of pages work and render fine (including HN) and it's an extremely rare occasion that I switch to Chrome to use a website. Even then, I often find that Chrome isn't any better and the underlying issue was the website's mobile handling in general (e.g. touch events working differently than mouse events, or just a completely broken mobile-only component swaps)
On a related music/audio note, I recently came across videos of the teenage engineering TP-7 which looks like a really cool application of similar haptic feedback concepts. Unfortunately it's out of the price range where I can justify picking one up just to tinker around with it, so I've only seen videos of it so far.
Designer here - I will admit that I just do this as a bit of a hobby and I'm not an EE, so I don't have a great answer for you, but generally speaking since this is a high resistance/low KV motor and there's only so fast you could spin it by hand realistically, I think the voltages you could generate that way shouldn't be so high as to cause damage in most cases. The TMC6300 datasheet does show it directly connected to the BLDC motor so that's what I based this design on.
There are probably some best practices that I'm missing though (diode on the main supply maybe?), so if anyone knowledgeable wants to chime in, I'm all ears! (Or I suppose pull requests are welcome too).
Thanks for the swift response! I've been interested in haptic feedback for a long time, but so far have not come across any general way to calculate the acceptable power dissipation for a given motor and driver circuit. Physical controls have so much to offer in ease-of-use, but the chief downside of them currently is that there is no easy 'two-way binding'.
Motors often run at high rpm and then shut off, coasting to a stop. I would expect any standard bldc controller circuit to be able to handle any potential generated by manual movements without difficulty.
The UX in your project is fantastic, well done. The motor kicks to simulate clicks, for example, are a really nice touch. I still don't get how the PCB detects flexing, I'll just put it down to magic.
MechE here. The way strain gauges work is with a long conductive trace that loops back on itself like a snake, or a radiator. This creates a conductive line that passes back and forth many times along the direction you want to measure strain in.
When the surface that the gauge is attached to flexes, the PCB itself actually stretches and shrinks by a very small amount, on the order of a thousandth of a percent. This causes the long conductive trace to get slightly longer, which measurably increases its resistance.
Agreed, this seems a likely explanation to me. I run a small hobby electronics business in my free time and although I ship some items to the EU I explicitly don't ship to Germany due to the additional complexity there; it's simply not worth my time for maybe 3 additional orders a year.
I'm sure the regulation is well-intentioned (and could even be great in aggregate effect too!), but a side-effect is that it biases the market in favor of larger companies who can afford to hire compliance specialists.
If you look at it from the German perspective, I think a person who is in/from Germany and has a small online hobby electronics shop would not be impacted by this the same way you are - thus, I believe the side effect you mention could be seen as biasing the market in favor of domestic small entities, as opposed to international ones, which might be good for a self sufficient economy. Feature not a bug, basically, is my suspicion. Although I suspect the intended target is more likely to be on Aliexpress than Tindie.
I made a rotary input device that provides software-defined "virtual" detents and end-stops, implemented using a BLDC gimbal motor. It can dynamically switch from completely smooth unbounded rotation, to having detents with configurable spacing and strength and "end-stops" that spring back if you try to rotate past them.
It's got a round LCD on the front of the knob (wired and supported via the hollow shaft of the motor) and uses the flex of the PCB and strain gauge sensors (in the latest revision, simply SMD resistors whose resistance changes when stretched) to detect when the knob is pressed down.
HN folks might appreciate that it communicates with host software on the computer via protobuf-encoded USB serial messages -- nanopb is awesome for embedded C protobuf support, and having the defined schema, autogenerated serialization code, and compile-time type safety is so much nicer than ArduinoJson or hand-written binary protocols!
I'd love to get it hooked up to some real software eventually (video editors or home-assistant control are my 2 main ideas), but it's really just been a fun project to tinker with and try out some new ideas and parts I've never used before.
Hah, yeah I think it may have. I've noticed an increase in comments over the last few days, which usually suggests the algo is showing it to new people again, and it got featured on the Adafruit blog this morning which probably reinforced it further, but no idea why the recent resurgence.
> The Petition recognizes that there currently are unlicensed part 15 devices operating in the Lower 900 MHz Band, but it is unclear regarding the extent to which the proposed reconfiguration would impact potentially millions of such devices. With respect to part 15 devices, NextNav states that it is completing technical analyses and “will work with unlicensed users to understand their spectrum requirements.” Id.
> NextNav does, however, seek the removal of the current requirement that it not cause unacceptable levels of interference to part 15 devices. See Petition at A-6 (proposing to amend § 90.361), A-11 (proposing to add § 90.1410(c)).
Requesting public comment is perhaps better than nothing (and likely better than just allowing lobbyists to influence policy behind closed doors), but it's a hard problem to quantify whether the collected comments are representative when one side is more heavily resourced and organized than the other?
Is it even possible to "work with unlicensed users to understand their spectrum requirements" in a way that doesn't ignore a potentially substantial long-tail of varied usage?