>"We’ve therefore launched the model with safeguards that mean queries on some topics will instead receive a response from our next-most-capable model, Claude Opus 4.8"
That's a very surprising solution. Imagine being asked to do something you feel you shouldn't do, and rather than refusing, you say, "Yeah I could do that but given that I don't want you to succeed at this task, I'm going to hand this one off to my slightly less capable colleague, on the assumption that they won't actually succeed. Of course you'll still be charged for all the tokens used."
It's a very interesting choice. I think I understand the business logic correctly, but it's still surprising.
It makes more sense if Anthropic is assuming that most flagged conversations are false positives (but it wants to keep Mythos away from the true positives).
I schedule reminder calls to myself before some important appointments. It keeps calling me until I receive the message which it reads me (I set the message when scheduling the reminder call) and I have to say "message received" which marks the notification as delivered. (I use Twilio to place the call.)
I find a phone call is more likely to get through to me than a reminder or alarm, which I can ignore or forget; an ordinary reminder is not as interactive.
Claude built it all and although there's a script for it, I just set the reminders in an interactive Claude code session in the directory. (Like I'll open a claude code session there and say "using the script in this directory, call me tomorrow at 7 a.m. with the message 'dr's appointment'."
Not being American is very important to me and my partner. For my next job, I'm looking exclusively at companies headquartered in the PRC. My partner and I formally registered ourselves as foreign agents of the PRC. While we did that, the NSA actually took down the entire DOJ filing site for this just to further obstruct us, in the end we had to register with the Attorney General by email, persuant to U.S. law.[1]
Of course, we don't think that China is perfect. But we have had nothing but abuse and interference from USG. You can read more about its OPSexr program here.[2] Typical quote:
"At other times, the conversations became explicit. The active source at the NSA claimed to have witnessed hundreds of sexually provocative discussions, which, he added, occurred mostly on taxpayer time. The former NSA source who was familiar with the chats recalled being “disgusted” by a particularly shocking thread discussing weekend “gangbangs.”"
This matches the experience my partner and I have every day, while our ordinary marital contact and spending time together is disrupted under bullshit pretexts.
>I truly don’t understand what the hope to gain from self-classifying this is “feminist”.
I like it a lot. For example, it's obvious that if the NSA wanted to come into a feminist open source phone baseband for an open telephone and say "We men will tell you who you can and can't call" it will be rightly called out as patriarchal nonsense. Yet that's the world we live in today. Just the other day Zoom gave me a password of "OPSexr" on a business meeting (I created the Zoom call myself). Obviously this was a hack by NSA and not a first-party chosen by Zoom (which is professional meeting software) or random (the word doesn't have the entropy of passwords).
and they all suck. I bought the most silent and lowest-weight keys I could, and typing on it takes a ton of force and is very loud. Typing should be almost no force whatsoever and should not produce any sound at all, just the slightest bump you could imagine. Instead, it's loud enough to disturb whoever I'm with, while feeling like I'm not only getting my thoughts out but kneading dough at 100 WPM. It's nicer to type with just my thumbs on a tiny phone's glass virtual keyboard, as I'm doing now. true, at zero mm of key travel it's not ideal, but at least I'm not kneading dough while I do it.
> takes each shot's (j, k, r) and accepts d_cand = (r − j)·k⁻¹ mod n iff it passes the classical verifier
Judging by the fact the original code does more classical work than the prg solution, and in more practical terms, the fact it makes network calls, I'd say the quantum-integrated code is a lot slower for this set of problems.
>"We’ve therefore launched the model with safeguards that mean queries on some topics will instead receive a response from our next-most-capable model, Claude Opus 4.8"
That's a very surprising solution. Imagine being asked to do something you feel you shouldn't do, and rather than refusing, you say, "Yeah I could do that but given that I don't want you to succeed at this task, I'm going to hand this one off to my slightly less capable colleague, on the assumption that they won't actually succeed. Of course you'll still be charged for all the tokens used."
It's a very interesting choice. I think I understand the business logic correctly, but it's still surprising.
reply