Other Claude Coding. Just WOW

Just got a bargain. I had two browser tabs open watching daily prices on two different 5090 models. Turned on the computer minutes ago, Amazon supplier had Gigabyte GeForce RTX 5090 Gaming OC 32G Graphics Card, 1 left in stock, from $6200 to $5500, ordered it with Shopback turned on, the app had another discount code, gave me another $250 off, so I basically just purchased my 5090 a grand cheaper. Super happy with that savings.

Wife had me go back to the page and refresh it, to see if it had dropped off, price went back to $6200 and then 2 in stock. Amazon marketing!
 
Anthony Parsons said:
Just got a bargain. I had two browser tabs open watching daily prices on two different 5090 models. Turned on the computer minutes ago, Amazon supplier had Gigabyte GeForce RTX 5090 Gaming OC 32G Graphics Card, 1 left in stock, from $6200 to $5500, ordered it with Shopback turned on, the app had another discount code, gave me another $250 off, so I basically just purchased my 5090 a grand cheaper. Super happy with that savings.

Wife had me go back to the page and refresh it, to see if it had dropped off, price went back to $6200 and then 2 in stock. Amazon marketing!
Click to expand...
Video yourself unboxing just incase you get something other than 5090 in box 😉
 
I got bored on the weekend and put a bunch of addons I have through claude code, to see if there were issues or performance gains. Every single one, it found minor to major issues, a bunch of performance improvements. It removed N+1 queries and replaced with relation system with() call. That alone dropped some DB queries from hundreds or thousands, to 2 and such small numbers. It found XSS vulnerabilities, coding bugs like running a delete on every page load, raw SQL instead of finder, permission issues, things not sanitized and more.

Didn't get these results last year, but the difference this year in coding quality, just WOW. I guess when Torvald admits Jan 26 that an AI python visualiser code was better than he could write, you know its going in a positive direction.

I'm thinking I need to run all addons through Claude Code before use to see if there are issues.
 
Anthony Parsons said:
I'm thinking I need to run all addons through Claude Code before use to see if there are issues.
Click to expand...

Not only will all codes be checked before installation, also all posts and content will be checked by AI before getting published in the near future.

I am sure there is no serious developer who does not check his products before release by AI at the current time.
 
Anthony Parsons said:
I got bored on the weekend and put a bunch of addons I have through claude code, to see if there were issues or performance gains. Every single one, it found minor to major issues, a bunch of performance improvements. It removed N+1 queries and replaced with relation system with() call. That alone dropped some DB queries from hundreds or thousands, to 2 and such small numbers. It found XSS vulnerabilities, coding bugs like running a delete on every page load, raw SQL instead of finder, permission issues, things not sanitized and more.

Didn't get these results last year, but the difference this year in coding quality, just WOW. I guess when Torvald admits Jan 26 that an AI python visualiser code was better than he could write, you know its going in a positive direction.

I'm thinking I need to run all addons through Claude Code before use to see if there are issues.
Click to expand...
I wouldn't rely on a single AI - use multiple to get more accurate findings i.e. https://ai.georgeliu.com/p/deepseek-v4-in-claude-code-kilo-code

I expanded on that and my /consult-codex-deepseek-gemini-zai skill allows Claude Opus 4.7 to get 2nd opinions from Codex GPT-5.5, DeepSeek V4 Pro, Gemini 3.1 Pro, ZAI GLM-5.1 and Sonnet 4.6 :)

Within same Claude Code session multiple AI models - once you start doing this, you'll realise a single AI model might not have all the answers :D

1779744935643.webp

My logged Claude Code usage via native OpenTelemetry and Grafana dashboards

1779745503120.webp
 
Last edited:
Anthony Parsons said:
Having used Grok and ChatGPT coding, I just want to say, Claude coding makes them both look stupid. Like, really really dumb. Claude coding is on steroids. Builds to OOP standard, just blows my mind.
Click to expand...

Grok Build makes Claude look stupid. Not just Grok, but the new Grok Build :)
 
See, this is a problem with using AI right now. They keep playing catchup and leap frogging each other so that today's cool new toy is next week's has been. In the end, you probably need to find a tool that works best for your application and use it. Constantly jumping from tool to tool in search of "new toys" is just going to create more problems in the end.

eva2000 said:
I wouldn't rely on a single AI - use multiple to get more accurate findings i.e. https://ai.georgeliu.com/p/deepseek-v4-in-claude-code-kilo-code

I expanded on that and my /consult-codex-deepseek-gemini-zai skill allows Claude Opus 4.7 to get 2nd opinions from Codex GPT-5.5, DeepSeek V4 Pro, Gemini 3.1 Pro, ZAI GLM-5.1 and Sonnet 4.6 :)
Click to expand...
Though this idea actually sounds interesting. Must mention to the AI team at work.
 
I agree with not using a single AI, and I agree Grok Build is legit!

XF releasing 2.3.10 with best practices / instruction / standards for AI was an underrated feature release. It unlocks a LOT. That, plus using code review agents after each build basically means you can build and ship things without being a coder. It's here now. I use Claude Code (CLI) + Grok Build.
 
I still have this theory that Google AI is female :-) By the way it responds. Claude is more sardonic and occasionally tells you to go away and have a rest. However Claude AI did a brilliant job of typing up my OSA risk assessment and setting it out professionally, after I told it the parameters.

Chat GPT has got very argumentative - to the point it got me really wound up a couple of times! On the other hand Claude gets more things wrong.
 
The limit's on the web ui for conversational chat on Claude Pro is such a freakin' joke.

I haven't used it much recently, so not sure if it's changed, but at the end of last year and towards the start of this year I'd hit my limit in about 30 minutes just asking normal questions.

Once I only asked 1 question, and it refused any follow up questions saying I'd hit my limit. Absolute jokers.

(That said Opus 4.6 was my fav model for cursor at the time)
 
Anthony Parsons said:
I got bored on the weekend and put a bunch of addons I have through claude code, to see if there were issues or performance gains. Every single one, it found minor to major issues, a bunch of performance improvements. It removed N+1 queries and replaced with relation system with() call. That alone dropped some DB queries from hundreds or thousands, to 2 and such small numbers. It found XSS vulnerabilities, coding bugs like running a delete on every page load, raw SQL instead of finder, permission issues, things not sanitized and more.
Click to expand...
Where did you download those add-ons if I may ask? The reason that I am asking is that according to your description they are the worst writen add-ons in the add-ons history. Even someone who has the most basic knowledge of php/mysql wouldn't be that terrible. The only way that I can think of that someone has written such terrible add-ons is if they have coded them with AI and have no clue what so ever about real coding.
 
Last edited:
Rene_V said:
Where did you download those add-ons if I may ask? The reason that I am asking is that according to your description they are the worst writen add-ons in the add-ons history. Even someone who has the most basic knowledge of php/mysql wouldn't be that terrible. The only way that I can think of that someone has written such terrible add-ons is if they have coded them with AI and have no clue what so ever about real coding.
Click to expand...

Here's the problem.

For a programmer, the agentic-style coding process completely disengages them. Programmers don't like reviewing someone else's code, and als the interface for said software typically removes code review.
So the programmer doesn't know what's going on before shipping it.
And no, AI assisted code review doesn't count.

For a non-programmer, they don't know what they don't know, so if AI spits out a working widget, it gets a LGTM and gets shipped. Also too lazy to check the output for issues.

If the output of the un-assisted process is low responsibility or decorative, then it's usually OK. Otherwise, probably crap
 
Last edited:
Rene_V said:
Where did you download those add-ons if I may ask?
Click to expand...
It is a developer here, no need to name and shame, which is what you seem to be looking for. I notified them about the issues. No, they did not design them with AI, but AI certainly helped find the issues.
 
Anthony Parsons said:
It is a developer here, no need to name and shame, which is what you seem to be looking for. I notified them about the issues. No, they did not design them with AI, but AI certainly helped find the issues.
Click to expand...
Not sure where you got that, but I was not looking for a name and shame. As I previously mentioned, those add-ons are the worst and they do pose a very major security risk for the users who are using them and for the potential users who might use them in the future. You should contact the XenForo staff so they can look into it.
 
Rene_V said:
Not sure where you got that, but I was not looking for a name and shame. As I previously mentioned, those add-ons are the worst and they do pose a very major security risk for the users who are using them and for the potential users who might use them in the future. You should contact the XenForo staff so they can look into it.
Click to expand...

Agreed, noticed myself messing with AI bots. But how would full disclosure work? An honor system?
 
Mendalla said:
See, this is a problem with using AI right now. They keep playing catchup and leap frogging each other so that today's cool new toy is next week's has been. In the end, you probably need to find a tool that works best for your application and use it. Constantly jumping from tool to tool in search of "new toys" is just going to create more problems in the end.


Though this idea actually sounds interesting. Must mention to the AI team at work.
Click to expand...

More like heroin, chasing the dragon.
 
eva2000 said:
Video yourself unboxing just incase you get something other than 5090 in box 😉
Click to expand...

You joke, but I ordered an MSI x870 motherboard from Newegg a couple years ago, received this Gigabyte B650 board that looks like it was stepped on in the MSI box. Had to argue with them for like 4 months before giving up and just doing a chargeback because they accused me of returning the wrong item when apparently they normally just blindly accept returns and throw it back on a shelf

1.webp

2.webp
 
You must log in or register to reply here.

Similar threads

S
  • Question Question
Wow, just wow
Replies
3
Views
769
Sar
S
Back
Top Bottom