Doing a RAM swap and then selling the resulting unit as if it were a Raspberry Pi of the higher RAM capacity might be construed as trademark infringement[0]. And in general it's fairly difficult to make arguments in antitrust cases that boil down to "yes, your honor, everything they're doing is within their rights but I don't like it".
[0] At the same time, the scammers could probably get away with this by putting "Upgraded to 16GB!" in the title of their eBay listings or whatever, in the same way that fraudulent storage devices are labeled as "upgraded" (read: lying about their capacity). That doesn't impact the antitrust analysis though.
I certainly agree that such a reseller should clearly state the modification, and not pretend its an authentic RPi 8GB unit.
But if this is the real concern, they could just display the fact that "Unoriginal RAM detected" to inform the user its not a vanilla RPi Foundation product, so buyers understand the situation, but continue booting normally.
For every user asking questions and facing issues for such-or-such firmware or software support they need in their projects, that same user would have appeared regardless if their Pi came from RP directly or through a reseller.
Every board sold would have been an extra user, the reseller isn't adding full users (although they may increase the indirect customer base!), so even gaining a user through a reseller increases the amortization divider for the support burden.
Now lets consider RAM chip support burden, the original Pi RAM chips can have issues generating support burden, and so can the reseller RAM chips. Unless RP foundation refuses to upstream link training / RAM bringup patches from resellers, there is no real additional support burden.
In fact if RP were a little more creative, they could have tried contacting the resellers to explain they can continue their modding business as long as they advertise it clearly as modified, and to inform them they will be receiving units that didn't pass the RAM tests from the automated lines, since they're going to desolder the chips anyway, this can save costs for RP foundation.
in those jurisdictions where either price discrimination or locking hardware are not illegal, it is typically perfectly legal to reverse engineer for compatibility purposes. So nothing prevents such resellers from upping their game and learning a thing or two about link training and general SDRAM bringup.
Once they have acquired the knowledge legally in such jurisdictions the product just flows in the other jurisdictions anyway.
Courts upheld the monopolistic position of microsoft etc. for decades, but respect has no finality, "re" + "spect" to review, so later such companies were fined for their decades long transgressions.
But during those decades it may appear legal to the average consumer.
All of this is pretty irrelevant from a legal perspective.
Reverse engineering is legal, but manufacturers don’t have to help you figure out how to reverse engineer their products and they’re allowed to design their products in a way that makes it difficult.
There’s generally not a lot of jurisdictions out there that make stuff like hardware lockdowns of this nature illegal, especially for companies that are not in regulated industries or are not monopoly/oligopoly players.
> All of this is pretty irrelevant from a legal perspective.
> Reverse engineering is legal, but manufacturers don’t have to help you figure out how to reverse engineer their products and they’re allowed to design their products in a way that makes it difficult.
I'm not asking for help to figure this out, and of course RP is allowed to shoot themselves in any foot they like.
Consider Europe where such reverse engineering is perfectly legal for compatibility purposes. Let's consider an idealized scenario, there were no security bugs or other issues with the product that would justify reverse engineering RP blobs in public. When they do something like this, they are GIVING CARTE BLANCHE to public reverse engineering efforts to attain compatibility with other RAM chips. Why would you provide legal cover in some jurisdictions for others in publicly reverse engineering your own products by introducing artificial incompatibility?
Also from a consumer perspective, RP tries to place itself front and center in maker communties, emphasizing modularity, not just hobbyists, also companies build and design end products using RP units. All of these companies must now be wondering where RP Foundations exact red line resides: can we resell an RP board in a Kiosk we made? or only for as long as it doesn't compete with RP's future Kiosk modules? Today its the RAM upgrade, but tomorrow RP modules may refuse to boot some popular Kiosk hardware intentionally. What prevents RP Foundation from eventually "not copying" end-application hardware like a Kiosk design from their customers, and then enforcing compatibility only with RP Kiosks?
So not only are they providing legal cover for public reverse engineering efforts, they are scaring part of their customer base.
> There’s generally not a lot of jurisdictions out there that make stuff like hardware lockdowns of this nature illegal, especially for companies that are not in regulated industries or are not monopoly/oligopoly players.
That's a very shallow interpretation of law, looking only at what RP is or isn't allowed to do.
There's generally not a lot of jurisdictions out there that make stuff like reverse engineering for compatibility or security applications illegal either.
The part replacement isn't being blocked? The OP's CM5 with replaced RAM is still a working CM5 according to him. And fully functional compared to before the replacement.
"Each chip is programmed with the timing settings and storage size settings of just the RAM we ship it with, and doesn't know how to boot with anything else." I don't know if you could call that "blocking." I mean you could call it that but I don't know if that's legally what it would be.
So European institution following this discussion would be wise to ASAP formally request RP Foundation to describe it either as a blocking security mechanism, or just a compatibility issue.
Between current inaction and eventual action, it settles the position of RP foundation, so they can't switch back and forth between "these people are circumventing security" vs "its just an unintended incompatibility", as long as they don't reveal their position its neither here nor there.
>This sounds insightful until you find yourself debating with the CEO for the eleventh time about how his high-school understanding of the physics involved breaks down in a particular case.
> ... This is just a list of someone's ideas for equivalently difficult (and transformative) unsolved problems in biology. ...
I think GP was saying the proof of the pudding is in the tasting: the longer a problem has provably resisted resolution the more difficult it is considered...
bombastically decorating a problem as equivalently difficult does not make it so.
Well, if LLMs actually turn out to solve some of the Millennium problems, at least some of these biology problems are almost certainly at least as difficult. Certainly solving any of them would be Nobel-worthy if biology nobels existed (and solving some of them would likely easily be worthy of a medicine Nobel). It's not like these are new problems whose difficulty is unknown!
The Millenium Prize problems didn't only withstand decades/century of resolution, they also withstood the same test of time in ridicule of the challenge statement.
This list of "Millenium Problems for Biology" contains such brainfart level "analogies" that there the list will be ridiculed, for the question / challenge itself displays a lack of understanding of the subject in question. Science is also asking the right questions.
Consider for example:
> 10. Protein Amplification Chain Reaction:
> Demonstrate exponential amplification of arbitrary peptide substrates.
> Specifically, demonstrate input-protein-dependent synthesis of new, full-length, sequence-faithful covalent polypeptide copies from amino-acid monomers without a nucleic-acid template or preformed cognate scaffold, in a single pot reaction. For the challenge to be considered complete, at least 100 random peptide sequences of at least 50 amino acids each must be preregistered, synthesized, and pooled. It must then be shown that the abundance of these peptides in solution can be amplified at least 1000x with at least 90% sequence accuracy on a per-residue basis. Reasonable modifications may be added to the peptide sequences to facilitate post-amplification analysis if necessary, provided they are not active in the amplification. Methods that rely on explicit sequencing of the peptide are not permitted. Methods that rely on reverse translation to generate a nucleic acid intermediate are not permitted, because they are duplicative with a separate Millennium Problem.
The analogy is very clear: to amplify DNA or RNA one uses PCR, basically throw the desired product in a cauldron with monomer building blocks, then by repeated heating and cooling the lone monomers find their permitted locations on a complementary pre-existing strand, and form the new polymer strand.
So it seems natural to ask for a generalization to protein polymers, except every biologist or chemist knows its nonsense: proteins don't have a complementary strand! You can't demand chemistry or physics to magically copy without a complementary template!
You may ask "but if that were true, how can we already have PCR for RNA?"
Well pretty simple: while this is done routinely, its only possible indirectly: convert the RNA to double-strand DNA, use PCR on this DNA and then convert the amplified DNA back to RNA!
The demand to not involve sequencing or the hypothetical reverse translatase from one of the other problem statements turns this one into a non-existence theorem, but the challenge doesn't describe a winner for demonstrating its impossibility!
I assure you that any chemist or biologist being asked why we dont have PCR for protein, will understand your lack of knowledge, and explain how PCR works, so that you understand that PCR was only possible because of the complementary strand!
This list will be ridiculed for being not even wrong.
thats a poorly implemented reward script, if it leaves you exposed to the mining pool with this gentleman's agreement.
the script could have been designed 2 phase, so one first submits a hash of the solution & submitter address, so even if miners front-run the submitter, they just helpfully pay the transaction fee!
I've recently been working on this exact problem due to my desire to create puzzle challenges for Simplicity, the smart contract programming environment that I work on for my job.
Since Simplicity runs on Bitcoin-like blockchains, someone can swipe the witness data from the legitimate winner's proposed transaction, and create a new transaction (perhaps with a higher fee) using the same claim data and sending the prize to a different address.
Anyway, I ended up implementing a two-phase commit mechanism in which you pay a deposit to temporarily lock the prize so that it can only be paid out to your address. If you then make a valid claim, the prize can be paid to you; if you don't, you forfeit your deposit.
(I think this was suggested by Russell O'Connor, the inventor of Simplicity, but it may have been a widespread idea in the smart contracts world. I don't know whether there's a straightforward way to implement it with Bitcoin Script, which is what this older prize would have needed.)
I don't claim to know the nuance of what you're trying to address with this.
Wouldn't it be simpler to simply protect a bitcoin private key with the encryption that you are challenging people to break?
Off the top of my head, the only downside I can see is that someone could drain the wallet without publishing the key, but people like to brag, so it seems unlikely to be a problem in practice.
That would work if you were only demonstrating encryption or creating challenges related to encryption, but some challenges don't map very easily to a cipher (though you might be able to get them to map to a cipher key).
In Simplicity (and in a sense in Bitcoin Script) there's a broader concept of "if you show you know information X, you're entitled to this money", but it has this specific issue that if the information or the entitlement to receive money for knowing it isn't unique to a specific recipient, there will automatically be a witness swiping or front-running risk for architectural reasons.
A gentleman's agreement between the western bloc, China, India, Russia, ... ?
It would be a lip service agreement, and all would continue the machine learning race...
The mere suggestion of slowing down AI progress signals to adversaries that there is some novel power just identified: do you think adversaries would agree and slow down, or calculate a little harder and longer in order to also figure out what it unlocks?
Do you ever wonder how Anthropic or OpenAI can release a new state of the art model and then a few months later there's a Chinese open model that's nearly as good?
Why not calculate the number of bits necessary before Sol goes supernova? The clock will only be as useful as long as the local star lasts, and when different star systems get populated they or their planets will use local clocks as well.
In theory laws could be authored requiring entities to register pragmatic assumptions with foreseeable violation dates, so that they can be added to a sorted queue of assumption violations to be crossed.
Let's be super safe and assume 128 bits will be sufficient for as long as the universe lasts. Whether humanity will still be around is a matter of debate.
if language models (current GPT-style or any future model with different architecture) can be viewed as a compression of their output corpus (the corpus it emits by providing random starting native contexts weighted by the model's likelihood of that native context) then one is saying the corpus contains the same information bits as the model, the model just takes less data bits, while the corpus represents inflated data bits.
If learning requires communication of data, one could ponder if it is more effective to transmit learning data as suitable model coefficients instead of corpus monologue or dialogue (say with a teacher).
This may sound absurd, but is it really so far fetched to believe one could formulate "download"-compatible LLM architectures, so that the user or student can play a kind of memory game to train "download"-architected model weights by simple reflex games?
Suppose instead of token vectors we used token matrices or token multivectors as in geometric algebra (not to be confused with algebraic geometry!).
Word2Vec couldn't do general language modelling because it used vectors, but when using token matrices or token multivectors one could postulate the following partition function:
given a dictionary of tokens, and their corresponding (matrix/multivector) M one could define the partition function
exp(-|M1 * M2|^2)
For example it could be alphabet level (to demonstrate the concept):
For every string one can compute the corresponding M in terms of the elementary character matrices:
M_"car" = M_"c" * M_"a" * M_"r"
both matrix products as well as geometric multivector products are associative but not commutative:
But M_"car" != M_"rac" since matrix/geometric products is not commutative.
The result of non-linear but layer-less and order aware architectures could promise the following:
Feasibility of uploading model weights: a random grand tour of 2D projections of the square matrix space, projecting the token positions down to 2 dimensions along an changing axis, a "game" could correctly project 95% of the tokens, requiring the user to identify the incorrectly placed tokens, which jump back to their correct position when clicked.
This means a user can learn (since most of the time tokens are rendered in the correct position of the 2D projected cloud), and as a user learns their performance will go up.
If a user could eventually (after playing for a long time) correct 90% of the token positions on random 2D projected planes, then the user has effectively stored 90% of their coordinates.
Suppose such a user reads text in their mother tongue, then the joint likelihoods in natural text will correspond to matrix products of those weights. In other words the brain will learn that whatever it has learnt playing the token-game aids the brain in predicting a next token.
Once it has learnt to utilize the format (matrix / multivector coordinates), the apparently pointless skill of positioning tokens in the cloud learnt during the game, it would also help predict the next token in languages the user didn't know, say when looking for ingredients on a product package. I believe such a user would swiftly discover they actually know those languages, and I believe transmission of a compressed format would take less time than transmission of the same knowledge in expanded output-corpus form.
I wonder if Terrence Tao has an opinion on such a possibility:
1) does he consider it (im)possible to have token-first formatted LLM's without MLP layers etc reach similar levels of performance as the current crop of LLM's?
2) supposing it were possible, would he predict a user "downloading" weights to his brain by a reflex-game as feasible?
3) does he believe a human brain would be able to make use of the downloaded weights and would the brain notice the utility in predicting tokens?
4) does he believe that once the brain has noticed / learnt how to make use of the model weights, first for predicting the next token in the mother tongue, it would subsequently have learnt how to predict in other tongues?
I predict it would be possible to quantify the effects objectively, as a thought experiment:
assuming functional proteins, gene regulation is about binding affinities of cellullar chemical species to promoter regions, so 200k promoter regions x number of chemical species present in humans give us the number of binding affinities that describe the behavior of human cells (essentially all cell types excluding some germ-line cells, but including brain cells).
the thought experiment: apart from ethics nothing prevents humanity from going to all the remaining wild-type or near wild-type humans (when did 5% of the local population start wearing glasses? as a measure of wild-type-ness), and collecting as many individual genomes as possible, for a reference data set for the table of binding affinities. For each individual genome there is 1 table of binding affinities. Now we start averaging the value of each location in the table by averaging this table over all the collected genomes.
So theoretically we could collect the bulk of the gene regulatory information / skills encoded in the genome for wild-type humans. Next we can traverse the medical literature on autism and collect all the reported or suggested features, survey answers, diagnostic indicators for the full autism spectrum, and for each population in such a separating indicator, fetch example genomes from the purported indicated and unindicated groups, and compare their averaged binding affinity tables until you resolve or exclude any genetic correlation.
Or people downvote because I can't wait for science to ... progress?
reply