TFW something you feared was your error or a weird quirk of your system turns out to actually be a bug 🎉
Nick@nickbaumann_
@_simonsmith not just you! 🥲 this one's on us -- hang tight
English
Simon Smith
6.4K posts
Simon Smith
@_simonsmith
EVP Generative AI @klickhealth
Katılım Ağustos 2023
796 Takip Edilen3.8K Takipçiler
@nickbaumann_ Thanks! Honestly, just glad it’s not me. I was worried I’d be like this n of 1 and spend days debugging.
English
@_simonsmith hey Simon -- this is a bug we're aware of & fixing right now. thank you for flagging, I'll keep you posted for when it's fixed
English
Codex mobile is failing for me. It keeps trying to do auth via a Google Account, but my ChatGPT login is via Apple. Anyone else having this issue?
English
Deleted all browser history for today and yesterday and got out of the Gmail-associated account, but still no option to log in with Apple. Gets directed right to Gmail login.
English
Frustrating. It can see the connection, but when I click to connect, it routes me through Gmail. And since I foolishly wanted to see what happened if I followed that route, I’m stuck with a connection to a ChatGPT account I created accidentally with my Gmail login and there’s no way out.
English
@sasha_po_ua Didn’t work. Still locked to the nonexistent Gmail-associated ChatGPT account.
English
@_simonsmith Log out of Codex on desktop and log back in, then try opening Codex on your phone again. It worked for me.
English
There is no way out of this now. It wants to connect but with a Gmail account not connected to my ChatGPT account. I can’t reset this flow. Canceling doesn’t do it. I even deleted the ChatGPT app and reinstalled it. This seems problematic.
English
Definitely seems like a bug with authorizing Codex mobile. Clicking to connect forces me to go through the auth flow, but the auth flow forces me through Google login, and I don't have a Google account associated with ChatGPT. I log in with Apple. I made the mistake of starting the auth flow to see what would happen and now it won't let me out of this and is forcing me to create a new account.
English
I think workspace-write with network access is the answer. I will find out tomorrow. If it works, the confusing thing is that this isn't necessary when I'm running the automation manually by asking Codex to do it, versus when it does it by itself. Anyway, let's see what happens tomorrow.
English
Please, Codex experts, help me figure this out. I have an automation that runs every morning. It needs to first rebase. No matter what I do, I can't get this to work. Sandbox issues. I've tried all kinds of permissions changes. What am I missing?
English
This is similar to how Claude does memory and I think is preferable to the current row-style database memory in ChatGPT. It lets the models build a more holistic picture of you rather than anchoring on discrete facts, and lets you more easily review and edit that understanding.
Tibor Blaho@btibor91
OpenAI plans to replace saved memories in ChatGPT with a structured summary of ChatGPT's memory on September 1, with an option to add or update something about yourself directly in the personalization view Plus new mentions of "ChatGPT in Viber" in Security settings, an option to "Send email directly from ChatGPT with a connected provider", and a new announcement that "Finances in ChatGPT is currently rolling out to Pro users in the US"
English
One good thing about the shift to usage-based token billing from seat-based is that labs have an even stronger incentive to compete. Having users locked into annual seat licenses is very different from having to fight for every token.
English
The current era of tokenmaxxing will be swiftly followed by an era of rigorous token accounting.
English
Got the new memory in ChatGPT today. First test really nailed it, giving a recommendation that pulled on memories going back almost a year. Also glad I can finally TROUBLESHOOT issues that come up when ChatGPT keeps using context from somewhere and I can't figure out where.
English
Update: "New" means newer than the last checkpoint, but not newer than the Glasswing announcement.
English
So there's a new Mythos Preview checkpoint and its performance here looks like a similar jump from Opus 4.6 to the early Mythos Preview.
AI Security Institute@AISecurityInst
Our cyber range results illustrate this step-up. Since our first Mythos evaluation, we received access to a newer Mythos Preview checkpoint. On a 32-step corporate network attack we estimate takes a human expert ~20 hours, this checkpoint completes the full attack in 6 /10 attempts.
English
I keep hearing of companies that plan to combine departmental token and headcount budgets, with department heads having to allocate that budget between people and AI. Am I in a bubble, or is this a trend?
English
I have sadly stopped using ChatGPT Atlas as I expect it to be killed soon. Switching to the Gemini sidebar in Chrome for contextual questions, and Codex with Chrome for more complex browser work. I await the super app and will adjust tools again based on how it works.
English
AI capabilities are rapidly outpacing our benchmarks. METR tasks lengths saturated. AISI task lengths saturated. FrontierMath approaching saturated since 1/3 of problems are broken. Now let's measure real-world impact. Businesses built. Diseases cured. Physics solved.
English