AI Coach
1.2K posts
AI Coach
@realaicoach
Discussing the latest news & tools.
Katılım Aralık 2021
67 Takip Edilen22 Takipçiler
PSA you can use Codex directly in Claude Code ✨
/plugin marketplace add openai/codex-plugin-cc
Try it out today with:
/codex:review for a normal read-only Codex review
/codex:adversarial-review for a steerable challenge review
/codex:rescue to let codex rescue your code
You can just Codex things!
English
Found a reason to finally use Teams
Aravind Srinivas@AravSrinivas
Why do some people in the world still use Microsoft Teams? @AskPerplexity
English
AI Coach retweetledi
Never underestimate how much time and effort you can waste by trying to automate a process you do not understand manually.
English
@DanutPralea @theo @grok Outline the GitHub Copilot/Microsoft <-> Anthropic revenue share agreement. Confirm or deny the assumed split.
English
@realaicoach @theo anthropic gets 90%. github gets the bag. that's the whole point
English
- 15 messages
- $221 of tokens
- 1.6% of my $40 plan used
It's obvious that GitHub couldn't keep this model for billing on Copilot.
Theo - t3.gg@theo
I sent a single message on Copilot and it did over 60m tokens. It's still going. $30 of inference so far. In their current billing model, you get 1,500 messages, regardless of how expensive each is. I'm pretty sure I can do $45,000 of messaging on this plan
English
Definitely, previously you needed a mental model of the inner workings of an app to make progress. Progress was slow and meticulous. That is not to say flawless, but you needed some level of understanding to make a dent in application code.
I do think creating high quality software with AI is possible. It's just that you need a lot of experience and high standards to guide a model, especially because you're moving so quickly.
At least behaviorally, a developer should have a full mental model of their piece of software.
English
Also, even though we have had AI for more than 3 years and relatively good models since Opus 4.5, I haven't seen any great software that was released out there
I have a feeling no one is thinking anymore; everyone is just vibe coding.
English
We've never experienced this level of low quality in software
Everything we did was to fight to improve the quality
Now, all of a sudden, everyone is chasing more lines of code with AI, everything is broken, and no one is disturbed due to that
Weird times to be alive
English
@icanvardar @grok What is known about the model size of Opus 4.6 compared to Opus 4.7? Was there a change in inference costs?
English
opus 4.7 is a straight downgrade. nothing about it feels usable anymore
English
@DanutPralea @theo What are their real costs on 221$ though? Nominally, $221 are the API costs. But at 90 % margin it's not so bad, assuming most users only utilize a fraction of their plan.
English
@odysseus0z Is /goal just to burn through the rest of your usage or is it more efficient than manual prompting?
English
@NielsRogge Don't worry, some of their $400k software engineers are on it: job-boards.greenhouse.io/anthropic/jobs…
English
The difference between the Codex desktop app and the Claude desktop app is that... one works
... the other one doesn't
English
@TaylorHabrock @oikon48 Thanks for the reminder. @mattlam_ for example summed it up nicely: x.com/mattlam_/statu…
Matthew Lam@mattlam_
OpenAI cooked with the new /goal, can't wait for them to add /goal to the Codex app, I will definitely be using it like crazy.
Here Codex created two games for me while I went out for lunch, and some tips:
- /goal
English
@shivamhwp Yeah, I know, it's quite easy for a human to tell. What could help is if the color palette were mixed up between prompts or per user. That way, the sites would look at least superficially different.
English
@realaicoach i think there is a kind of look that you know is sloppy.
basically inconsistency + same look everytime.
as generally with claude models you can define a particular prompt and it can be hard to think that it's ai gen-ed. but gpt. you know it's gpt
English
I think we've already passed the event horizon. Few humans can write code as well as a Tier 1 agent in a fixed time window, even less so if you give proper guidance or allow refactoring follow-up prompts.
OpenAI even demonstrated that agents can handle a code repository of a million lines, see openai.com/de-DE/index/ha…
According to the article you can absolutely forgo humans writing code. The job of an engineer then moves up one abstraction level: defining what good code is, and facilitating the integration logs, metrics and traces into the agent feedback loops.
English
@realaicoach Codex App has good UX, and Codex CLI has additional features which user can unlock. Codex is rapidly increasing the user.
English