I almost always use medium. xhigh thinks too much, becomes an ouroboros, and starts redirecting itself. I think it makes more technically correct mistakes or reads too narrowly into what I’m asking.
I still want to customize what I’m using but most users at our company don’t need to know which one they’re using. I just set global config to medium with fast_mode false. I don’t want to write a paragraph and think of every caveat to stop it from eating its own tail or explaining to nontechnical users when to use what.
“Goal complete - Output: need clarification from Nick to research how payments are processed.” - xhigh when goal is to research how payments are processed.
Fast mode isn’t that much faster to be worth it for most of our users.
xhigh does get my sql schema stuff right more often than medium. And I trust it more with analysis.
DB for tickets/issues, memories (curated and default), teams, outlook, git, and iMessage.
Can codex just know my entire life.
Phone call context and in person meeting context and it would be perfect for business. I’m still using gpt-4o-transcribe for apple notes when not on teams. Codex needs all of my context.
@jxnlco Sometimes I like to queue 10 next steps sometimes I queue and steer. Sometimes I steer. It all depends on the models thought and understanding of what I asked
@theo I just tell codex to refer to query history and give it schema lookup & path tables. Rarely makes mistakes - mostly wonky business logic and then I update the skill..
Can we choose to make skills only discoverable when calling the ‘@Plugin‘ or ‘/skill‘? I don’t like that they’re eating up to 2% of context. TEAMS, OUTLOOK are excellent. Would be nice if they had more graph permissions natively without me make them. Also would be nice if I could have the ‘manage’ style screen first. So hard to explain to nontechnical users how to find their Apps & Skills that aren’t on the Plugins page. Like telling them they have to click the GHE app not the GitHub plugin. I do 90%+ of my work in codex so it’s hard to say everything we do.
Meet DiffusionGemma!
An experimental open model that explores a fast approach to text generation, released under an Apache 2.0 license.
Moving beyond sequential, token-by-token processes to generate entire blocks of text simultaneously. Here’s what’s new with DiffusionGemma: 👇
5.5 refuses to import oai tool into the workspace. When I open the file excel says that it needs to repair it - minor inconvenience. Would be nice if all of the tools it needed for plugins didn’t need to exist in the workspace. Same issue for presentations, docs, etc. I use medium mostly + review for me.
@Dimillian Tell it to commit and push. Sometimes if I want all local changes I might use the ui because it only wants to push its specific changes not caring that the other three threads all contributed. Hard trade off bc that is good sometimes. I don’t really notice it as much with 5.5
@zswang24@jxnlco@OpenAI Can I edit it to add a custom app/plugin. Codex is the first model that isn’t too lazy to do sql without spoon-feeding it. Or would you do @ data analytics @ sql-tool?
I find myself using codex less for code and more for everything everyday.
I’m committing to 4. 3 internal across different teams - Gitlab, GHE, GitHub Business, and personal GitHub.
I’ll ask codex to consolidate and make one for me.
Need GHE US residency support on the GitHub connector or ability to configure auth URI
@TheRealAdamG Let me sell Enterprise Codex to our customers 🙏 and a few points on BAA environments. I’ll do most margin on implementation/connecting their data
@jxnlco Letting the GitHub plugin connect to our ghe cloud and internal GitHub natively. It has it in memory that it can do it but it still says token is expired even when it is not. Then I have to repeat “escalate to me.” The GitHub connector doesn’t let me change or add multiple URLs
@gdb SVP I report to asks will Codex do XYZ and I love to say “just ask it”. I can feel the job security when he replies “I love it - it just works” lol
I blocked /fast in our enterprise codex. We’ve never hit our enterprise limits and suddenly used all of it in 4 days with everyone using 5.5 fast. I can’t really tell the difference. Make it easier for legacy tenants to get codex only seats - can’t stand making Codex do token to credit to $ math lol