I disagree, the claude models seem the best at tool calling, opus 4.5 seems the smartest, and claude code (+ claude model) seems to make good use of subagents and planning in a way that codex doesn't
Opus 4.5 is so bad at instruction following (30% worse per benchmark shared above) that it requires a manual toggle for plan mode.
GPT 5.2 simply obeys instruction to assemble a plan and avoids the need to compensate for poor steerability that would require the user to manually manage modalities.
Opus has improved though so the plan mode is less necessary than it was before, but it is still far behind state of art steerability.
> Nowadays, newly-onboarded Discord corporate users receive a laptop and at least one Yubikey alongside that laptop. IT onboards users to Okta and instructs them to register at least two WebAuthn authenticators; typically, this is their Macbook’s TouchID/Windows Hello sensor and also their Security Key C NFC.
> We also instruct corporate users to set up Okta Verify for use only as a fallback MFA in the event that all their authenticators fail at once. This way, we never have user accounts lacking at least one strong form of multi-factor authentication.
So OS level keys, a yubikey for roaming, and Okta Verify for fallback
In addition, Okta Admins can also recover accounts, so loss of the Yubi doesn't mean the account is locked out forever. You can easily provision a different 2FA method.
When we deployed we banned Verify (didn't want any OTP), but encouraged TouchID, and the Yubi. If someone was locked out we could temporarily enable Verify, or reset their Macbook or Okta access so they could reregister into either.
But,in deploying 1500 or so yubikeys over a 5+ year period we never saw one actually break. Employees would often say they'd broken, but troubleshooting normally was user error.
The worst we saw were a few cases where Yubis needed unplugging and replugging (sometimes being left out for an hour or so).
It effectively is if you're ready to ship your dependencies. If you're happy to depend on major versions that come with the system, qt isn't bad either.
$0.13/prompt can only be useful for artists/end users. Anyone thinking about using this at scale would need a 20/30x reduction in price. But there's still no API available so I think that will change with time. Maybe they will add different tiers based on volume.
That is a fair point. I don't think the pricing is unreasonable, but it feels limiting. You could try 1000 variations until you find what you need perfectly, but in that pricing model users will be induced to use less the tool, not more.
I'd prefer an option to pay like 200 usd/year to use unlimited. And maybe have a price per use only in the API.
edit: this pricing model also makes it expensive to learn to use the tool.
Until you consider the level of demand for this product, which is surely higher than OpenAI can scale to with the number of GPUs they have. If they price it lower they’ll be overwhelmed.
I have an older Surface tablet (that I recovered from a recycle bin, TBH) and I've been using Chrome on it, but I was annoyed by some of the latency and tried Firefox the other day. It had all the problems of Chrome, plus jittery scrolling and some extra random lags thrown in, so I noped right out of it. I've got nothing against Firefox as a concept, but it's apparently too hardware intensive for that scenario.
Are you sure it isn't the hardware? Perhaps there's a reason it was in the recycling bin. Not my experience whatsoever, even with "old Surface tablets".
> Theft of iPhones basically stopped being a problem because features like these makes the device a mostly useless brick once it’s marked in Apple’s systems as stolen.
People are still routinely tricked to enter their iCloud credentials into fraudlent websites designed to look just like iCloud, allowing thieves to wipe the device.
I've had this happen to three people I know. Works like this: they contact you via your contacts, pretending to be apple care specialists, ship a link, then social engineer you into entering credentials, then they disappear.
(I don't really know how they get to contacts, but this was the case in all three instances)
Can confirm, had my XR stolen. Contacted on whatsapp via a number that I put as the lost message (displays when you turn the phone on).
Story was, this girl from Italy bought my phone on ebay but then saw my message and then took it to the Apple store. Someone from the apple store would contact me to arrange recovery of my stolen phone.
Couple hours later I got a message from another number asking for me to log in to my iCloud account and arrange an appointment using this link. Opened the link and about to type my details in then I saw the domain wasn't quite right. Looked into it, realised it was a phishing attempted and bailed on that.
I know they can't access the phone whilst it's still on my iCloud account so it will remain there, in lost mode, until the heat death of the universe.
It’s not really surprising that organised criminals are trying to get around the anti-theft measures (although crazy to hear that you know that many who’ve had their phones stolen). But as you can imagine, running such social engineering schemes takes time, and success is not guaranteed, so even in this case, the anti-theft measures are making it less attractive to steal iPhones.
