No, that’s for “frontier LLM development” which somehow includes examples like distributed training infra.
Based on how sensitive the classifers are, any data scientist / MLE is probably going to encounter cases where some silent degradation happens and you never know about it.
It does nothing to protect against distillation attacks, because distillation attacks are far less interested in the topic of AI research than just generally getting tons of diverse output from the model. It might be that Mythos was (accidentally?) trained on internal Anthropic documentation on how Mythos was trained, and thus it could leak secret sauce? Doubtful; it feels like its less about the specific attack of reverse-engineering Mythos, and more about being a general sophon against any model training at all; that Anthropic's official position is now that they're the only ones who should be training models.
For doing some reporting stuff internally, there isn’t a certification. But there are definitely humans who have to certify financial statements and communications for financial offerings.
SLAAC and link-local is very different from DHCP/NAT/etc. in IPv4 world. Link-local addresses are pretty arcane in IPv4 while they are a central idea in IPv6.
That’s fine. As pointed out elsewhere, DHCP was relatively new when IPv6 was introduced. But it is a learning curve well past just knowing the difference between NAT and stateful firewall.
But pro-AI posts never seem to pin themselves down on whether code checked in will be read and understood by a human. Perhaps a lot of engineers work in “vibe-codeable” domains, but a huge amount of domains deal with money, health, financial reporting, etc. Then there are domains those domains use as infrastructure (OS, cloud, databases, networking, etc.)
Even where it is non-critical, such as a social media site, whether that site runs and serves ads (and bills for them correctly) is critical for that company.
The scams are more sophisticated than getting gift cards to pay the IRS. A number saying that it’s from the bank will say they need to verify some account information.
I have had to actually verify my “investment profile” with a major broker in order to unfreeze some trades, in a high friction process. To the extent that a sideloaded app that looks exactly like the bank app has a low friction install, then people can get fooled and irrevocably lose savings.
If the lock-down is opt-in, almost nobody will opt in to it. If the lockdown is opt-out, then whether scams still happen depends on how much friction there is in opting out.
Freedom to install other unsigned sandboxed apps has a solution: Banks could use passkeys and other non-phishable methods. Sideloaded apps in Android can’t get to the bank app’s passkey.
Passkeys or hardware tokens get worries about the enshittification of the theoretical recovery process. Which, if that’s the case, I guess we should hope for/pay a better world, at least with banks and brokers. For them specifically, for account recovery allow either showing up in person or using ID checks.
Both for personal accounts and business accounts (i.e. with Business Email Compromise), I believe the onus should be on the bank to use non-phishable methods to show the human-readable payee from their app for irrevocable transfers.
Electricity price regulation, at least for transmission, has been a thing for states for 100+ years and federally since the 1930s. Pipelines and railroads also have price regulation of some sort.
Monopolies, in these cases natural monopolies, can in fact exist. Look at the Micro supply and demand curves. As a general rule over those 100 years, there has not been rationing of electricity. There are natural blackouts and today an unplanned surge in demand (as happens in every industry such as chips after Covid), but generally the price regulation did not cause some kind of gas lines.
Medicare has overhead, but you’re not saying whether it is more than commercial insurance. The admin expense/profit portion of commercial insurers also don’t take into account provider admin costs (not to mention the huge amount of time patients can deal with denials, appeals, etc.)
reply