At the very top of the article:
"Achieving world-class AI inference performance through Japan-developed 2nm 3D-stacked CPU and server integrated, developed, and manufactured in Japan"
"Japan-developed CPU..." plus the server itself (parts, power supplies, case, etc.) integrated, developed and manufactured in Japan. It sounds like they're weasel-wording it a bit, but I don't think the CPU itself is manufactured in Japan, only "developed."
> What about giving it to your secretary with a list of what to shop?
"When a Comet user directs the Assistant to locate an item on Amazon.com, the Assistant takes screenshotsof the browser view, sends those screenshotsfrom the user’s computer to Perplexity’s servers, and receives instructions from Perplexity’s servers on how to navigateAmazon.com. In other words, the Assistant cannot operate wholly independently; it relies on direction from the user and instructions from Perplexity’s servers."
Obviously all browsers rely on direction from the user. But they also typically rely on instructions from the browser maker's servers. Traditionally, you get all of those instructions in a single download that's been pre-packaged (the browser program itself). But if part of the browser's logic requires more computational resources than most consumers have, what's the problem with "outsourcing" that bit to Perplexity's servers?
As a consumer, I don't think it's very prudent to trust a company with that kind of access to your data, but it doesn't seem materially different from the kind of trust you have to give to Chrome.
> if part of the browser's logic requires more computational resources than most consumers have, what's the problem with "outsourcing" that bit to Perplexity's servers?
It's not longer my agent. It's a joint agent of my self and whoever else is giving it instructions. (This is Amazon's argument.)
Also, Amazon told Perplexity to fuck off. If your assistant is told to stop doing something in a cease and desist, and then keeps doing it, the fact that you asked them to do it isn't a get-out-of-jail-free card.
My guess is we'll find a balance where local models can act as user agents while corporate ones have to meet certain requirements to gain that safe harbour.
Icosahedron is the classic? I have never ever seen an icosahedron globe in my life, neither real or digital. What in the world happened in the brain or artificial neural network when that title was picked for an icosahedron “globe”?
My personal favorite is Authagraph, when folded they make a really neat looking shape. I bought one years ago at a paper store and it’s still standing.
That usage is not a generic payment usage. This is a use case it is good for. But if 8 billion people should use Bitcoin for their everyday shopping, that is a totally different case...
Well you wouldn't use Bitcoin itself to exchange on, you'd use side layers like Lightning payments for the near instantaneous exchange which then gets aggregated into one of the upcoming blocks.
MasterCard was around in the 1950's, in 2026 they do not still use the same method of exchanging money as they did back then and have had to make big changes to their systems as time has gone on, Bitcoin does the same thing except it's not controlled by a banking cartel but by all it's userbase.
> except it's not controlled by a banking cartel but by all it's userbase.
Bitcoin is absolutely not controlled by its user base, but by another cartel (made of mining pools and exchanges). You, as a user have literally no say in any decisions taken by bitcoin except voting with your feet, which is exactly what you're suggesting against Visa & MasterCard's (which aren't banks by the way…)
The exchanges and miners do have a large influence, but the node runners (the users) have a voice as well. The node runners can get together, and refuse to peopogate the blocks of of the miners/exchanges that are being bad actors (BIP 148 for example) and also (BIP 110) that's currently making the rounds.
If a large enough percentage of the node runners need refuses to propogate a block from a bad actor, then another miner can mine the same block and get the reward.
The miners don't need “node runners” to get access to other miners mined block. That's not how the bitcoin protocol works at all.
In an ideal world the best non-miners can do is prevent some transactions by preventing them to reach the mempool, but you're not going to block an exchange this way either.
The bitcoin protocol is actually very friendly to centralization, the only mechanism that was supposed to prevent it was the PoW, which failed because ASICs exist. And as a result, in practice it's been very centralized for more than a decade now.
You can’t self certify any products including a radio (almost everything today due to WiFi and BLE), products for children or medical equipment. An accredited lab is required.
The pre certified module do not exempt the product from testing. It only means that you don’t have to do the conducted testing and that it’s a higher probability for passing the other tests. There are still lots of chances to fuck up and a full radiated test of the product is required.
In the EU (and UK with UKCA), CE certification is down to whoever makes the product available on the market. There's no get out for pre-certified modules. If you put a pre-certified module in a box and sell it, you have to be responsible for the CE mark.
You need evidence that the whole product complies with the EMC regulations. There's some judgement involved - if you put a certified circuit board in a plastic box, it won't change the EMC behaviour so you don't need to re-test. But if you solder a module onto a larger circuit board, you are almost certainly changing the EMC behaviour. You should definitely be re-testing something like that.
The pre-certified module does save you some time in the "intentional radiator" testing - assuming you use the same antenna it was tested with as a module, you can probably make the case that the behaviour is the same.
Will anyone ever notice or care for a few thousand niche devices? I doubt it. But to sell in the EU legally you do need to do that testing.
No, you actually don't have to do the testing, except in some special circumstances (e.g. childrens' toys). Generally, the "CE" mark is a self-certification done by the manufacturer or importer that certifies that the product meets certain (category-specific) criteria. You can do the proper thing, test your product and then certify it. But it is also possible and usually legal to just issue the certificate based on a self-assessment and deal with the fallout later, if there ever is any. That's how basically all non-EU manufacturers and importers get their CE mark...
Most products don't require testing by a notified body. That means anyone can do the testing - or even you could do it at home. The CE mark is self-assessed - you don't have to get a lab to issue it to you.
But electronics does have to conform with various electromagnetic compatibility rules. There's effectively no way to prove that without doing EMC testing. Most non-EU manufacturers will still do EMC testing in a lab, perhaps in China. That test report becomes part of the product's technical file.
My point is that if you take a module from a 3rd party and integrate it into a larger electronic product, it no longer has the same behaviour from an EMC point of view. You have no evidence that it meets the harmonised standards for electromagnetic compatibility. Realistically, probably nobody will ever notice. But if they do, you don't have evidence that your product complies with the standard.
Products for children, medical equipment and all things with a radio (nfc, Bluetooth, mobile radios, WiFi +++) requires an external accredited lab.
You can test electronics without a radio yourself, but it requires a very extensive lab. Few companies can do this. But you can save money by using a non accredited lab.
>> except in some special circumstances (e.g. childrens' toys)
> Don't nearly all electronics fall under it too, this is the second item after children toys in the link you posted.
No. That page just says: There are tons of EU laws about everything, including electronics. By putting a CE mark in your product, you certify that your product complies with all applicable EU laws.
In most cases, this certification can be a self-assessment, meaning that you can just assign the CE mark by whatever internal process you deem sufficient. I.e. you are allowed to rubber-stamp, but of course you are fully responsible for all consequences.
In some cases, for example childrens' toys, but only for some electronic components (electronic toys, medical stuff, high-voltage components, components that emit EM radiation above a certain threshold, ...), self-assessment isn't enough. In those cases, you need to obtain a certification from a notified body, which is an external organisation doing some design review and testing, as required by the respective laws.
It's all laid out further below under "How to obtain CE marking" and "Do you need an independent assessment?"
Very interesting. I tried to make a tree based agent two years ago but back then the LLM couldn’t manage to navigate the tree effectively. But today everything is different: better LLMs, and we have better understanding on how to instruct them.
I’ll try it and maybe I’ll be inspired to make my current home brewed agent more document focused again.
One drawback is much higher token usage because we get a very good saving in the cached data.
A little heads up: in your web page, the header download button force double screen with on mobile (iPhone 17 pro). No big problem but always an annoying when a page scrolls in two directions.
The issue with COBOL code is that it’s hidden. It’s mostly internal systems so little code available for training. HTML, TypeScript, JavaScript, C, etc, are readily available, billions of code lines.
reply