Server-Driven UI, compilers, language design, and reactive programming. Shared Client Foundations at @Netflix
Austin, Texas
- Tweets 32,893
- Following 853
- Followers 24,972
- Likes 7,102
Still early but, as my daily driver, thus far it feels like Grok 4.7 is a step backwards from 4.6 and 4.6 was a step backwards from 4.5.
So strange. Hoping I'm wrong. I've been using 4.6 enough now that, at least for my use cases, I'm confident it was vs. 4.5
5
10
1,348
4.7, FOR ME MIND YOU, seems to be skipping very obvious things. Taking shortcuts and half-assing things. It reminds me of why I needed Plan Mode. Which I know flies in the face of what they've *said*, but thus far my experience.
1
2
176
Pure speculation but maybe their drive to try and beat the other top models led to it not being as great of a daily driver; maybe it's better on paper at some things but in practice not good enough to use as the daily driver. I expect my daily driver to do tasks that aren't particularly hard, but aren't just trivial either. 4.5 seems like it's better at that, thus far. Will give 4.7 another couple weeks then switch back to 4.5 if that pattern holds. I was *very* happy with 4.5 as a daily driver. Again, emphasizing this is all highly unscientific. It could just be coincidence.
2
156
Productivity. If you use AI *heavily* do you feel like you are doing:
(poll character limits make wording weird)
74%
more in same amount time
4%
less in same amount time
7%
(mostly) unchanged
15%
see results
54 votes • Final results
1
814
Me asking AI: "How hard would this be to migrate?"
AI: "Massive, multi-quarter initiative."
*do migration using AI in a day*
3
1
18
1,332
Why, since Grok is my daily driver, do I still find myself when I ask it get opinions from Sol and Fable that I tell Grok "you're smart too"?
Because...it works? It otherwise blindly agrees with them.
4
8
1,335
Me trying to remember what coding was like before AI, and I've been doing this for well over 20 years
6
12
1,615
Kidding (somewhat) but seriously: if you're not contributing PRs across the stack, to repos you don't own, you're doing it wrong. I don't mean blindly vibe coding changes. I'm saying: it takes less time for you to have AI figure out where the bug or missing gap is and fix it, you review the changes, have it then submit the PR, than it does for you to waste time trying to Slack them and ask for someone to eventually find the issue and fix it and land it: 1 day vs weeks, if that. Don't let org boundaries hold you back anymore.
1
3
587
Especially if you've developed the understanding of where AI is good and where it's bad, SKILLS, etc. you can almost always know where it clearly fixed the issue vs nope it's more complex and needs humans. Even in the latter it's almost always faster than just asking someone else to fix it for you.
276
If anyone non-Netflix streaming service UIEs follow me: the biggest feature request I have is to stop having back button leave your player if I just finished scrubbing. I'm trying to dismiss the player controls, not leave the player.
4
825
Best advice with AI coding: don't use the same company's models to review your code as what you used to create it. e.g. use Sol to review Grok, or the other way around.
9
2
18
2,603
