The boring but true answer is that the only thing people should be protesting for is a change of the electoral law. Everything else is downstream of that.
In the US, it's a de-facto duopoly on power, held up by a number of "winner-takes-all" rules. Politicians of either party will do everything in their power to keep "outsiders" (i.e. people/parties that are not entrenched in the two-party system and might actually drive positive change) from ever gaining a foothold.
In Germany it's the famous 5% rule that virtually ensures that every new party must maximize populism or perish.
I'm sure it's very similar in most other "democratic" countries.
Laws aren't perfect. In fact they often are buggy as hell. The electoral law is certainly no exception. However it is ultimately the law that matters most as it determines who can raise to power and who can't. Ensuring it fair and democratic should be the #1 civic duty.
I'm not sure "keeping out outsiders" is a bug. The US is experiencing what it is like to be governed by an outsider with no previous political experience and who thinks things like "laws" don't apply to him, and who thinks experts can't be trusted and puts unqualified people in charge of the military, science and health. Politicians need to develop -- they should start with a local position, and "graduate" to a national-level position before they even attempt to rule a nation.
Notably, we could do that while still abolishing first past the post. Requirements for holding a previous position could be added while simultaneously reforming the federal (and hopefully also state) systems to be compatible with multiple parties. I imagine it would be sufficient for each level to require a single term served at the previous level - city or county, state, and federal.
The downside is encouraging career politicians, but the upside is that if you can't win increasingly high stakes elections over a period of 10 years or so then you probably have no business being the president of a country this size.
I think this take highlights one of the core problems our democracy faces - winning elections and governing effectively are entirely different skill sets. These things may even be, in part, antithetical.
I merely intended it as a reasonably general proxy for relevant experience whose ruleset would be difficult to weaponize. I agree that in theory there almost certainly must be better methods than elections by which to select legislators, leaders, and other official positions. However I'm not aware of any in practice, particularly when the inevitability of bad faith attempts to abuse the system are taken into account.
Otherwise we would have loonies like the Grey Panthers (old people party), the “Spiritual Party”, or the extreme right-wing “Republicans” (AFD is moderate compared to those) being able to vote on laws etc.
Of course that also cuts out some parties that I have supported in the past, but the system allows a lot of parties to participate that aren’t _that_ populist (e.g. the Greens, the Left, the Pirates (I think they managed to get a seat or two in the past))
Of course it’s not perfect, but I still think it’s one of the best flawed systems we came up with so far.
We should keep iterating on it but very slowly and carefully.
The obvious problem with the 5% rule is that voters who don't like any of the established parties are faced with the decision between voting for something they don't like or most likely throwing away their vote.
As someone who writes algorithms for a living I can think of ~ 100 ways to resolve this bug without limiting the original intent. You don't have to be a rocket scientist to come up with one. However the fact that this %5 rule hasn't been changed tells you everything you need to know about the legislators.
It doesn’t shut them out though, they can still effect change on smaller scale - for example smaller regional and state governments.
They don’t need to sit in the nation wide parliament and add noise.
The rule also encourages cooperation and compromise which is arguably also more democratic than everyone staying in their more extreme position
What’s “fraudulent” about lying in a marketing blog post? Private companies can, in most situations, say whatever they want about their business. They are almost certainly lying (as are most of these companies) by categorising any revenue as “recurring” but it wouldn’t be fraudulent to do that. Their investors are surely aware of the real financials and have no problem with this public posturing because it serves only to increase the value of their investment.
ARR drives startup valuations. For example people might argue if you have $100M ARR the startup is worth 10x, so $1B.
Now that’s a pretty good incentive to drive up ARR, no matter how.
Welcome to the world of creative accounting!
One-time projects? Count as ARR! 3 months contract with no extension? Count as ARR! You gave a discount of 50% for the first year? Make sure to count the full price as ARR!
While I have no issues believing that the outlined strategies are effective, and I sincerely congratulate the author on his journey, there is a flip side to engineering "social normalcy" that IMO the author is missing:
Any reasonably "normal" person (anyone that's not severely autistic) will find there are people that we effortlessly connect with and many others we don't. It's the natural state.
Now in any sufficiently intelligent and psychologically OK person the act of eliciting / pushing emotional connection with people from the latter group (where there's no natural connection) should trigger a certain amount of internal disgust.
The fact that it doesn't seem to be the case with the author would indicate that he's more of an outlier. Based on his writing he does seem intelligent and psychologically OK, so there might be other factors at play. My point is that his journey might not be transferable 1:1.
Can someone ELI5 to me (someone who doesn't have the time to keep up with all the latest research) what this is and why it's a big deal?
It's very hard to guess from the github and paper. For example, there is OCR in the title but the abstract and readme.md talk about context compression for LLMs, which I find confusing. Somebody care to explain the link and provide some high-level context?
Suppose you have an image with 1000 words in it, and suppose for simplicity that every word is 1 token. Then the image is “worth” 1000 tokens.
But under the hood, the image will have to be transformed into features / embeddings before it can be decoded into text. Suppose that the image gets processed into 100 “image tokens”, which are subsequently decoded into 1000 “text tokens”.
Now forget that we are even talking about images or OCR. If you look at just the decoding process, you find that we were able to compress the output into a 10x smaller representation.
The implication for LLMs is that we don’t need 1000 tokens and 1000 token embeddings to produce the 1001st token, if we can figure out how to compress them into a 10x smaller latent representation first.
Excellent, thanks. So basically this is saying: "our pixels-to-token encoding is so efficient (information density in a set of "image tokens" is much higher as compared to a set of text tokens), why even bother representing text as text?"
Basically. Some people are even saying, hey, if you encode text as an image then you don’t need tokenizers any more, and you get more expressivity from the graphic styling.
Another takeaway is that you don’t need to pass a tensor of shape (batch_size, sequence_length, d_model) through your transformer. Not every token needs its own dedicated latent embedding. You can presumably get away with dividing sequence_length by a constant.
This isn’t super ground breaking but it does reinforce the validity of a middle ground between recurrent models, where context is compressed into a single “memory token”, and transformers, where context is uncompressed. 1 < n/k < n
He threatened the ex-intern by saying they have "a lot of money to pay for top lawyers" (goodness!), the incident gathered a lot of attention, got caught, and was forced to say sorry.
Replit is in the news because of the Vercel fiasco. And it's jarring because of how they've tried to take advantage of that situation.
Got it. I was going to say it sounds like less than a fiasco but I see this is in context of someone criticizing the Replit CEO for a specific action so now I get it. Apologies for being dense. It's a reminder that people leading companies are just people. And if we have a purity test on everyone we better be prepared to have a purity test on ourselves. And not just in our own eyes but in the eyes of everyone else. And that's where it quickly falls apart.
It gets even crazier when compared to other IP law:
Engineer makes an invention: Write 30-Page patent application. Multi-year patenting process with USPTO, pay 1000s of $ if DIY, 10x that if using an IP law firm. Multiply by 4x if going international. With luck, patent gets issued 3 years later. It protects you for 25 years, but only if you have deep pockets for an IP lawsuit in case someone does copy you -- and with uncertain outcome.
Artist releases a song: automatically enjoys 100+ years of protection, even for minor samples, hooks, melodic elements. Lawsuits are easily won as long as you can prove you are the copyright holder.
I have my theories about how we ended up in this state of affairs but no jurist with a sliver of common sense can seriously claim that this is fine.
I think it's worth mentioning that with a patent, nobody else is allowed to use the patented idea. This holds even if they have never heard of you before, and were entirely unaware your patent (or your version of the patented concept) existed. You are granted a monopoly.
With a copyright, people are allowed to do anything similar to you, so long as they do not derive their work from yours. Someone unaware of your work cannot infringe your copyright. You do not have a monopoly on anything, you are just protected against someone deriving their work from yours directly.
This difference probably factors into how easy it is to win a lawsuit: for a copyright infringement, you need to show they made a copy. You wouldn't be bringing suit in the first place if there were substantial doubt in that area. It also factors into how easy it should be to get a patent vs a copyright: a patent closes off much, much, MUCH more idea-space than a copyright.
> With a copyright, people are allowed to do anything similar to you, so long as they do not derive their work from yours.
That is patently false for music - a songwriter's claim today that they never heard a slightly similar hook from 40 years ago generally doesn't hold in court.
The burden of proof of infringement is on the plaintiff in either case.
If you have an obscure patent nobody has ever heard of, you can win the suit by showing that the defendant had the same idea you did - you don't need to show their work derives from yours.
If you have an obscure song nobody has ever heard (or even an unpublished one!) you are going to have a difficult time proving infringement for a similar riff.
Perhaps you're most familiar with litigation around very famous songs, like the Under Pressure / Ice Ice famous case? In those cultural-saturation cases the court may make the assumption that the later party could have been exposed to the work of the earlier one, but that is not the case universally.
Harrison's song was "My Sweet Lord", the song allegedly infringed was "He's So Fine" by some girl group in the 1950s or 1960s. It shouldn't be hard to hear either one. My recollection is that the songs did some pretty much alike.
It's clearer than you think. Did Dua Lipa hear an obscure Florida reggae band's tune that isn't available on any major platform to listen to any more? It took some time, but eventually the court agreed she'd never heard the song she supposedly copied:
How is that different from a patent then? I don't think anyone would sue you for patent infringement against your hobby project. IP only typically becomes an issue when you want to make some money from your work, and then someone claims that it's actually their work, and thus should be their money.
Also, patents are for frankly more important things. It's a much bigger deal if life-saving medicine or a more efficient car are locked out for 100 years than Winnie the Pooh wearing pants.
Copyright and patents are very different things. Lumping them under the disingenuous umbrella term "IP" only serves to muddle the waters and create FUD. They are not property rights.
It's best to criticize each precisely and surgically. Know the terms, know the rules, the exceptions, etc. Know the history, know the original purpose of these laws. That kind of broader knowledge in broader society is what can help. The big corps are interested in having a vague blurry idea around "IP" that just makes you scared and think "wouldn't download a car" and has a chilling effect of thinking that all "that stuff" is electrified and better not touch it, and that it's just natural that there's "intellectual property" and it's just minor details whether it's copyright or patents or trademarks or whatever else. Property rights are ancient. By associating copyright with that, they make it seem that it's also just as fundamental and civilization-grounding as private property, when most of intellectual history had no such concept. Derivative works, tweaking ideas, splicing them in new ways was just normal.
A related disingenuous propaganda term is "content consumption", again creating the association between e.g. reading a book or listening to a song on the one hand and eating food, or using up soap or fuel on the other.
What? Copyright and patents are exactly the same thing. Making "you copied me!" actionable at a court of law, by statute, when before that there was no such legal fiction of "intellectual property" or any other exclusive rights to reproduce a thing.
Copyright and patents are absolutely not the same thing at all.
A patent makes it illegal to use a particular idea, by any means. It is a limited-time universal monopoly on a set of specific "claims" (the selected applications of the idea). It has nothing to do with whether someone copied you or not.
A copyright is a prohibition on someone copying you, with certain exceptions where they are allowed to do so ("fair use" or "fair dealing"). Copyrights also, in the USA, bring certain protections against people attempting to use a work in a way you did not intend even where they are not duplicating it.
A trademark is a prohibition on someone causing confusion by copying something you use to identify yourself, or by using/referencing it (without copying) in a context you don't wish.
A trade secret is a criminal prohibition on someone intentionally causing another to divulge certain information that you contractually banned them from divulging.
A security classification is a ban on certain ideas/information being shared with parties that a government agency did not wish them shared with, or used in certain contexts. This is not a protection available to individuals, only to governments.
A license is an intentional weakening of one or more of the above types of protection, potentially with attached civil penalties in the event the bounds of the license are exceeded. It can thus, when accepted, limit behaviour beyond what would normally be allowed by one of the IP types above.
All six of these IP types protect against Person B doing something that is in some way related to an activity Person A did earlier, but they are really quite different in what they cover and how. I don't agree at all they're the same thing.
Also the purpose is important too, to know how to move forward when the tech landscape shifts.
Patents are there to incentivize making it public how inventions work. Inventors would tend to keep their inventions as trade secrets, in fear that someone else will reap all the profits. This would slow the compounding effect of later inventions building upon previous ones. Therefore, patents give exclusivity to the inventor for a set period. In exchange, during this period competitors and others become aware of how the new tech works and hence they can prepare for the time when the patent expires, and then a host of derivative tech can appear. It has been perverted to an absurd parody where the vast majority of patents are not intented for actual use, never get licensed and simply provide cannon fodder and deterrent stockpiles in the lawyer wars between companies. The patents are so broad and vague that legally speaking tech companies are constantly tramping on each other's ground, resulting in a kind of stalemate truce where they agree not to sue, because the other would countersue.
Copyright is there to incentivize creative authorship and its dissemination to the public. By giving exclusive rights, the author can pull a revenue stream, making cultural/artisitic/intellectual creation more viable and hence spurring intellectual and cultural activity in society. The goal is to make authors incentivized to create. Not to drive up the stock prices of mega-publishers and music labels.
Trademarks are there to avoid confusion for buyers as to the identity of a seller or by falsely implying endorsement.
What's common to these is that they are overwhelmingly there to help broader social interests, a common good, in accordance with the enlightenment, somewhat romantic ideals of the optimist zeitgeist of the time when the concepts were defined. They involve restricting individual rights, such as free speech (you can't recite this or that poem in public). That restriction was done reluctantly, to serve higher purposes, like a thriving intellectual exchange and technological progress on a societal scale. It was not about trying to make things as cushy and profitable for huge conglomerates as possible.
>A patent makes it illegal to use a particular idea, by any means. It is a limited-time universal monopoly on a set of specific "claims" (the selected applications of the idea). It has nothing to do with whether someone copied you or not.
Just FYI, as a layman with no opinion on whether they are "the same thing" in this context, this paragraph sounds self-contradictory. It sounds like you're describing copying somebody's idea, and then you say "It has nothing to do with whether someone copied you", so I ended up confused as to your meaning.
Let's say I come up with the idea of using a fan to blow a balloon into the air. I get a patent with the claim "a device made buoyant in air and propelled by forced wind".
Someone else comes up with the idea of blowing a dust bunny into the air with their breath.
Their idea, which has nothing at all to do with my idea and is certainly not a copy, infringes my claim because what they are doing matches what I've claimed as the core idea in my patent. They didn't "copy" me, for two reasons:
1. Their idea was had independently of mine, with its own creativity
2. Their idea is fundamentally not the same as mine, but because my enumerated claim is sufficiently broad to cover their concept too, it's still infringement
With a copyright, their additional creativity would have made their work not infringe. With a copyright, you can't claim anything broader than the work you actually produced.
> With a copyright, their additional creativity would have made their work not infringe.
But not generally. You can't sell a new Hogwarts-based book with Harry Potter characters. This is both due to copyright and trademarks. But already copyright blocks derivative works even if there's additional creativity with a new plot.
You can't sell a Harry Potter book, but you can sell any number of Magical School for Wizards books. They can use similar themes to Harry Potter, similar pacing, even similar artwork styles.
They just can't copy the characters, significant chunks of text, or images.
That's the difference between a copyright and a patent in a nutshell: you copyright an implementation of an idea, while you patent the idea itself.
I think you want them to be meaningfully different, for whatever reason, but at the end of the day, they both come down to "if I did the thing you already did, and that you laid claim to through some form of artificial statutory fabrication of rights, you can sue me".
Whether that means me exploiting having heard your song by playing your song myself, or exploiting your invention I examined by building it myself, they both come down to: statutory fabrication of fictitious "you can't do because they did already" rights, that at common law could have (rightly) only been achieved through keeping the thing a secret (e.g. still present to this day in say trading algorithms, and in software through the now ubiquitous SaaS model) and contacts (i.e. NDAs) flowing from that.
See my other reply parallel to yours. There's no principle of "I did it first, therefore it's my property!". For example, if that was so, you could report an invention and get a patent for it without disclosing exactly how you did the thing. After all you did do it first, so it should be off limits by the (non-existent) "I did it first" principle. Instead, patent law requires "sufficiency of disclosure", meaning that you MUST disclose enough information that another skilled person can recreate the invention from the specified information. You get the time-limited exclusivity in exchange for disclosing the method so that others can work on top of it, refine the technique etc, so when the time comes that the patent expires, there will be improved versions. It is explicitly there to inspire others to work on the thing afterwards, just with some time delay.
Blurring distinct laws and their nuanced purposes into some generic "I call dibs!" principle is exactly what the propaganda part is. Because that creates a kind blurry haze in people's minds that even fills gaps that none of the existing laws currently block out. So people will feel like "that just feels illegal, but I can't exactly say what it violates". A kind of FUD around doing all manners of free intellectual activity in society.
"property in general" is essentially a "legal fiction" because there are no natural property rights other than "I successfully guarded my pile of stuff against thieves". heck when you think of a complete body of law for instance, its all "fiction", if I were to use your terminology.
Outwith the U.S.A. there is a thing called a design right that applies to that.
In the U.K., the design right took its initial form in 1787 and applied to printed patterns on fabrics; so it has been on point for clothing design for over 230 years. (-:
Indeed. Good example where value comes from. It's all the same crap, but since you've seen My Crap worn by Anne Hathaway a couple of times in ads at the airport, I can charge 10x price.
Classic case of how democracy isn't, in practice, majority rule.
If you put this demented situation to a vote, it'd lose 9-to-1.
Frankly the whole concept of copyright is absolutely stupid though, the equivalent of escalating schoolyard "stop copying me!" to actionable at a court of law. But since when did something being absolutely frothing at the mouth retarded stop it being an entrenched part of the world we live in.
I apologize for this extremely dumb question, but how is this a "server"? As far as I'm aware Blender is a local app. It can run without an internet connection. If an LLM wants to call into it, it needs to call its local python API.
Is this just unlucky naming or am I missing a critical piece?
MCP is a spec that is attempting to standardize a communication pattern for registering and calling tools from an llm. Part of the spec is a server that exposes specific JSON-RPC end points with a registry of the available tools, resources, and templates, and a way of executing them. That's the server, in this case the server acts as the interface into Blender.
The pipeline for LLM to MCP and to the app looks like,
LLM -> chat app -> MCP client -> MCP server -> specific app (Blender)
The chat app doesn’t know how to talk to Blender. It knows about MCP and links in a client. Blender exposes its functionality via a MCP server. MCP connects the two.
A server-client architecture can run on a single computer. You just need one piece of code to act as the "server" and one to act as the "client". Technically you don't even necessarily have to involve the networking stack, you can just communicate between processes.
A bit "off topic" but this is called AMA, so why not:
To which degree do you see your own profession (law practice) changing with AI? As an outsider it looks like most of the legwork, research work or basically anything a non-senior attorney would spend their time on could soon be done by LLMs. How accurate is that impression? What future do you envision for law-related professions?
I can only speak about immigration law and AI is changing it significantly. This might also change but right now I think a lawyer is still needed to understand the nuances of the law, to understand the risks of different approaches, and to determine the best strategy, but more and more, initial legal research and initial drafting are being done by AI.
This was of course a calculated move. The founders of Glass are not that stupid. They knew the original author would complain in the loudest way possible and cause a viral outrage, which would give them a ton of eyeballs and exposure.
Engagement hacks, outrage, eyeballs, distribution, attention at all cost. Welcome to tech in 2025.
In the US, it's a de-facto duopoly on power, held up by a number of "winner-takes-all" rules. Politicians of either party will do everything in their power to keep "outsiders" (i.e. people/parties that are not entrenched in the two-party system and might actually drive positive change) from ever gaining a foothold.
In Germany it's the famous 5% rule that virtually ensures that every new party must maximize populism or perish.
I'm sure it's very similar in most other "democratic" countries.
Laws aren't perfect. In fact they often are buggy as hell. The electoral law is certainly no exception. However it is ultimately the law that matters most as it determines who can raise to power and who can't. Ensuring it fair and democratic should be the #1 civic duty.