oh? this is inefficient code? but I used Garry Tan’s GStack and Claude Opus 4.7 on ultrathink!
| Home | Orlando, FL |
| Website | https://eshumarneedi.com |
| Threads | https://threads.net/@eshumarneedi |
| Bluesky | https://bsky.app/profile/eshumarneedi.com |
Eshu Marneedi
@EshuMarneedi
- 207 Followers
- 126 Following
- 6.6K Posts
Technology blogger, unskilled jester, raconteur. Writer of stories long and short. Pastry connoisseur.
Props to GPT-5.5, it gave me such a good edit of my Tim Cook/John Ternus article in 57 seconds while Opus took 10 minutes and came up with almost nothing.
(I obviously review all edits, and none of them change the content of my human writing.)
I am of the personal opinion that Gurman should issue an apology — or at least, an acknowledgment of his error — to the Financial Times editors. They were right, almost on the money, and he was not. https://daringfireball.net/2026/04/delicious_claim_chowder_regarding_the_cook-ternus_ceo_transition
New Metric album is pretty good. Definitely a good summer album.
I wrote about how the modern AI industry has forgotten what I call the “essence of computation”: letting humans be humans. https://eshumarneedi.com/2026/04/24/the-essence-of-computation.html
Updated: my best models list. GPT-5.5 earns runner-up in the image analysis and knowledge work categories due to its lower hallucination rate and improved image analysis. You can really tell the model is a new pre-train — it truly is much nicer to work with. https://eshumarneedi.com/uploads/2026/models.html
GPT-5.5 thoughts:
- Hallucinations are now fixed. It’s now on par with Opus 4.7.
- They broke LaTeX rendering in the Mac app. Fun!
- It’s still not as nice to talk to as Opus. If you work *with* the model, Opus is still the way to go.
- It’s extremely fast. Best model for quick, terse answers.
Just to be clear, an industry where a text editor that makes API calls to AI providers is worth $60 billion is not an industry worth taking seriously.