Jess 👉 AI & Tech
Region: US
Monday 29 December 2025 18:09:13 GMT
2927
141
7
11
Music
Download
Comments
Against The Norm :
I notice that as well
2026-01-12 05:52:45
0
BigRanch :
When coding agents I have GPT break each section down into singular steps, with an overall verification of progress at the end of each section. Helps the AI with memory during coding progress and during each section of build. Previously GPT would suggest up to 5 steps during each section build but would lose its place if it encountered an issue. Ultimately leading to looping and getting stuck or making recommendations that did not apply to the overall build goals.
2026-01-20 02:51:29
0
Home Bluston :
The future is small domain specific lm's with some mechanism to direct queries to them.
2025-12-30 19:41:14
0
AI & Automations :
true story 👌😎
2025-12-31 12:58:21
0
Tim :
No idea what a production-level codebases are, but ChatGPT and Claude both fell on their respective faces when I asked them to make changes to a word document in a series of revisions. Sometimes, they’d make the requested changes without problem. One time, though, I gave ChatGPT the latest version of the document (which was then 14 pages) and asked for a more changes. It confidently announced that it had made the changes and returned the document. When I opened the document, I saw that the 14-pager had inexplicably become 3 pages. Claude had similar problems. 🤦♂️
2025-12-29 19:58:10
0
Jimmy's place :
Vibe coding.. Doesn't work.. Maybe in 1-2 years from now..
2025-12-31 10:52:44
0
Jaden Stock :
Claude code with Opus 4.5 is by far the most impressive model I’ve seen. GPT-5.1-codex-max is second. Sometimes one or the other ends up getting a little stuck so I end up pivoting between the two. Claude is also like 3-5x more expensive. For deep research I’m always just starting in ChatGPT.
2025-12-29 20:31:03
0
To see more videos from user @therubberduckiee, please go to the Tikwm
homepage.
