I have absolutely no insight knowledge, but I think it's not a bad assumption to have that, it's costly to run the models, when they release a new model they assume that cost and give per user more raw power, when they've captured the new users and wow factor, they start reducing costs by reducing the capacity they provide to users. Rinse and repeat.
They're "optimizing" costs wherever possible - reducing compute allocations, quantizing models, doing whatever they can to reduce the cost per token, but vehemently insisting that no such things are occurring, that it's all in the users' heads, and using the weaseliest of corporate weasel speak to explain what's happening. They insist it's not happening, then they say something like "oh, it happened but it was an accident", then they say "yes, it's happening, but it's actually good!" and "we serve the same model day by day, and we've always been at war with Eastasia."
They should be transparent and tell customers that they're trying to not lose money, but that'd entail telling people why they're paying for service they're not getting. I suspect it's probably not legal to do a bait and switch like that, but this is pretty novel legal territory.
There are frequently claims that Anthropic is somehow diluting or dumbing down models in some subtle way. Unfortunately it’s tough to validate these claims without a body of regularly checked evals. This test set should hopefully help settle whether Anthropic is actually making changes under the hood or whether the changes are all in people’s heads.
> We never reduce model quality due to demand, time of day, or server load
Forgive me, but as a native English speaker, this sentence says exactly one thing to me; We _do_ reduce model quality, just not for these listed reasons.
If they don't do it, they could put a full stop after the fifth word and save some ~~tokens~~ time.
Yes, Dario is responsible for some of the weaseliest of corporate weasel wording I've ever seen, and he's got some incredible competition in that arena. Those things aren't the reason, they're just strongly coincidental with the actual reason, which is to slow the burn rate and extend the runway.
They wouldn't do it "intentionally". It would be an mistake "accidentally" made by a Developer or AI, that under the right conditions allows Zoom employees, etc arbitrary file reads on the host...
The article doesnt' claim it's executed straight up either ("can result") but it's pretty ambiguous:
> When the project is opened, Visual Studio Code prompts the user to trust the repository author. If that trust is granted, the application automatically processes the repository’s tasks.json configuration file, which can result in embedded arbitrary commands being executed on the system.
In the screenshot the task is named "node" - so it's a bit like embedding a malicious Makefile target as a backdoor.
Except harder to spot since it's in a obscure .vscode/somethingsomething json file. (And probably you can easily fool GH Copilot to run it)
In tasks.json, which I use for automatically `git fetch`ing on a few projects. While I don't recall it's interaction with first run / untrusted folder dialogs, it's entirely automatic on second run / trusted folders.
Does it matter that much? I don't think there is any "safe" build system. Users will try to build project sooner or later. With Maven it is easy to add a plugin with harmful payload as dependency, you won't spot it in "source", unless you carefully review every dependency. IDEs need containers/isolation and they need it now. Instead we got that "Do you trust this project" dialog.
Thank you for offering an additional perspective. The article reads like the tariffs were a complete and total failure. You are saying it is too soon to tell, correct?
For many if not most of the affected imports, it would never ever make economical sense to try to rebuild a local industry from scratch, at least not without major subsidies. And as for local alternatives that already exist, they carry a premium higher than any tariff (because a local workforce likes to be paid living wages), even if magically scaled up to achieve some degree of economies of scale. Western economies are truly and irreversibly post-industrial.
This take depends on the munificence of the local retailers deciding all on their own (because there's no external pressure apart from other retailers in the same position) to lower the price of their goods (and therefore reduce their profit) once the tariffs have forced the foreign goods off the market.
I don't believe it for a second.
Here's something that has happened before though. During the American War of Treasonous Aggression, the southern states decided to impose tariffs on the export of cotton to the UK. That caused massive hardship in the UK in the short term as it made the goods that the cotton was a raw material for impossible to sell. UK jobs were lost, UK workers starved etc.
So the local UK manufacturing companies found markets to supply them elsewhere in the world. They set up the supply lines, they reached agreements, and life went on. UK jobs were gained again, UK workers stopped starving. Life was good.
When the American War of Treasonous Aggression was over, the southern states wanted to drop their tariffs and start supplying the UK again. It didn't happen - there was no need to return to an unreliable partner when everything was set up just fine and dandy as it was, and the US cotton was essentially unsellable. US jobs were lost, US workers starved etc. The cotton industry never recovered from that.
Tariffs are a "fuck you, make me" slap in the face. Do it to people you don't like, but if you do it to people who are supposed to be "allies" (obviously not allies in the above case, but the consequences here were just as real), the consequences can be ... quite concerning for the tariffers.
Tariffs/subsidies should be weaned off over time. Their success absolutely does not depend on "local retailers [why are you even talking about retailers when it is all about manufacturers?] deciding all on their own to lower their prices". The external competitors are still there and will just come back when tariffs are lowered, so domestic manufacturers are forced to become cost competitive.
> Tariffs/subsidies should be weaned off over time.
The economic problem is that so long as the production cost imbalance exists, weaning off the tariffs just creates the same market forces that created the trade imbalance (and export of jobs) that created the situation in the first place.
I.e., if it inherently costs $5 to make a "widget" in Elbonia [1] and it inherently [2] costs $25 to make the identical (in every way) "widget" here [3] then while a tariff of $20/widget would make both equal in price, any reduction in the tariff will make the Elbonian made widget cheaper, and a purchaser will be incentivized to buy the Elbonian made one over the "made here" version because they, individually, save money by doing so.
So to maintain the widget making industry "here" the tariff has to be maintained. Any reduction and the cost incentives of "made in Elbonia" reappear, and the local manufacturer sees a corresponding drop in sales.
[1] Chosen only because it is not a real place.
[2] Meaning the local manufacturer cannot possibly produce one for less, due to higher costs "here" (e.g., energy costs, raw materials costs, labor costs, insurance costs, etc.)
Historical background
Russia launched a full-scale invasion of Ukraine in 2022, which followed Moscow's 2014 annexation of Crimea Common knowledge citation[*]. Russian leaders have repeatedly demanded that Ukraine renounce NATO membership and accept changes to control of Donetsk, Luhansk, Kherson and Zaporizhzhia — demands that form the backdrop for any negotiated settlement cited in reporting Citation 9 from dawn.com[9]Citation 10 from meduza.io[10].
This tracks with what I've seen. Greater "capability" can imbue a greater ability to lie, cheat, deceive oneself, or others, and generally create all sorts of complicated problems, be them internal or external
Layman in cryptotography (that's 99% of us at least) may be encouraged to deploy LLM generated crypto implementations, without understanding the crypto
reply