They aren't. Even when (a rare 'if') they give the correct answer out of the box in coding, they rarely implement it in the same way. It's great to feed their answers to one another to get a quick and dirty first take on vulnerabilities.
August 19, 2025 at 12:39 PM
They aren't. Even when (a rare 'if') they give the correct answer out of the box in coding, they rarely implement it in the same way. It's great to feed their answers to one another to get a quick and dirty first take on vulnerabilities.
pastebin.com/q5wbaZWN yes , GPT-5 Thinking (and not base model) did a great job there combining relevant code already available on web for this
1. Hexagonal Grid & Cell Physicshttps://observablehq.com/@lloydrichardsintern/ - Pastebin.com
Pastebin.com is the number one paste tool since 2002. Pastebin is a website where you can store text online for a set period of time.
pastebin.com
August 16, 2025 at 11:32 AM
pastebin.com/q5wbaZWN yes , GPT-5 Thinking (and not base model) did a great job there combining relevant code already available on web for this
GPT-5's base model is weak, even worse than GPT-3.5. But the GPT-5 Thinking model is a powerhouse. It could likely create a Photoshop-style app in HTML if the code is on the web, and it's one of the best models for coding—though not at a PhD or even junior dev level
August 16, 2025 at 11:11 AM
GPT-5's base model is weak, even worse than GPT-3.5. But the GPT-5 Thinking model is a powerhouse. It could likely create a Photoshop-style app in HTML if the code is on the web, and it's one of the best models for coding—though not at a PhD or even junior dev level
I wouldn't want them to converge and all be wrong about the same things. At least right now, you can get a second (or third, or fourth or nth ) opinion from a different LLM on a coding-related question before running your own tests.
August 16, 2025 at 10:46 AM
I wouldn't want them to converge and all be wrong about the same things. At least right now, you can get a second (or third, or fourth or nth ) opinion from a different LLM on a coding-related question before running your own tests.
To be honest, I've had a lot of fun with local models in the past. When I say local I also include stage or beta servers. I don't believe any end-user will engage with an open-source model through an API unless we, as programmers, have a compelling reason to make it happen , to make money out of it
July 30, 2025 at 11:06 PM
To be honest, I've had a lot of fun with local models in the past. When I say local I also include stage or beta servers. I don't believe any end-user will engage with an open-source model through an API unless we, as programmers, have a compelling reason to make it happen , to make money out of it
Do LLMs really have everyday purposes in coding? Like what? Could you just chat with an open-source model about a coding problem you have, and it would give you good advice without costing more than a month's trip to an exotic island? Even state-of-the-art models are just advisors, and often wrong.
July 30, 2025 at 10:56 PM
Do LLMs really have everyday purposes in coding? Like what? Could you just chat with an open-source model about a coding problem you have, and it would give you good advice without costing more than a month's trip to an exotic island? Even state-of-the-art models are just advisors, and often wrong.
I don't have a Raspberry Pi, but even on my PC, which has many magnitudes more resources, those models are only for playing. They have no meaningful impact, unless you're completely new to a subject.
July 30, 2025 at 10:42 PM
I don't have a Raspberry Pi, but even on my PC, which has many magnitudes more resources, those models are only for playing. They have no meaningful impact, unless you're completely new to a subject.
If you can transfer that logical, programming-based approach to problem-solving to those kids, you'll have achieved a great deal.
July 30, 2025 at 10:33 PM
If you can transfer that logical, programming-based approach to problem-solving to those kids, you'll have achieved a great deal.
As someone with dyslexia, I often make similar errors like 'shouldn't there should' myself. But if you had the money to run a viable open-source coding LLM locally now , why would you even mention it, other than for bragging? Unless that's the same "what if" scenario all smoke companies sell.
July 30, 2025 at 10:21 PM
As someone with dyslexia, I often make similar errors like 'shouldn't there should' myself. But if you had the money to run a viable open-source coding LLM locally now , why would you even mention it, other than for bragging? Unless that's the same "what if" scenario all smoke companies sell.
And what have you produced in 7 months of vibe coding without manually altering the code?
July 30, 2025 at 10:07 PM
And what have you produced in 7 months of vibe coding without manually altering the code?
I am dyslexic , I do typos all the time , some times I even surprise myself e.g. I was looking yesterday why a complex SQL statement failed and it had "UDPATE" on it ;) I realized it only in the console while try to recreate it
June 16, 2025 at 11:03 PM
I am dyslexic , I do typos all the time , some times I even surprise myself e.g. I was looking yesterday why a complex SQL statement failed and it had "UDPATE" on it ;) I realized it only in the console while try to recreate it
No chance for a medium or small e-shop to be referenced in "AI Mode" especially for product searches. What will happen if it becomes the default option in Google search ? In such cases adding content in a way that "AI Mode" wants to digest is unpaid labor contributing solely to Google's revenue.
June 16, 2025 at 10:51 PM
No chance for a medium or small e-shop to be referenced in "AI Mode" especially for product searches. What will happen if it becomes the default option in Google search ? In such cases adding content in a way that "AI Mode" wants to digest is unpaid labor contributing solely to Google's revenue.
No chance for a medium or small e-shop to be referenced in "AI Mode" especially for product searches. What will happen if it becomes the default option in Google search ? In such cases adding content in a way that "AI Mode" wants to digest is unpaid labor contributing solely to Google's revenue.
June 16, 2025 at 10:50 PM
No chance for a medium or small e-shop to be referenced in "AI Mode" especially for product searches. What will happen if it becomes the default option in Google search ? In such cases adding content in a way that "AI Mode" wants to digest is unpaid labor contributing solely to Google's revenue.
Demis Hassabis didn't told this , this is the actual video '''https://www.youtube.com/watch?v=CRraHg4Ks_g&t=798s''' yes he is exaggerating but he is talking in a future tense and not in few years limit.
June 10, 2025 at 1:35 PM
Demis Hassabis didn't told this , this is the actual video '''https://www.youtube.com/watch?v=CRraHg4Ks_g&t=798s''' yes he is exaggerating but he is talking in a future tense and not in few years limit.
Which human? If you pick a programming novice, yes, LLMs are better. But AI is a great tool to learn new things at entry level, or offload common tasks like regex, Linux commands, XPath. (I still know if their response is good because I know the context.)
June 10, 2025 at 1:09 PM
Which human? If you pick a programming novice, yes, LLMs are better. But AI is a great tool to learn new things at entry level, or offload common tasks like regex, Linux commands, XPath. (I still know if their response is good because I know the context.)
Some more ideas for an article. Java vs Spring , Ruby vs Ruby on Rails , to be or not to be
June 10, 2025 at 12:56 PM
Some more ideas for an article. Java vs Spring , Ruby vs Ruby on Rails , to be or not to be
Most CEOs don't care for the long term, only for quarterly reports and revenues. They want/need to boast to the board that they saved $X by eliminating N junior programmer jobs. (Some junior programmers won't have real jobs to learn.) This is why this bubble is so dangerous, in my opinion.
June 10, 2025 at 12:22 PM
Most CEOs don't care for the long term, only for quarterly reports and revenues. They want/need to boast to the board that they saved $X by eliminating N junior programmer jobs. (Some junior programmers won't have real jobs to learn.) This is why this bubble is so dangerous, in my opinion.
I agree 100%. AI code autocomplete (even in cases where there isn't any hidden bug on it) more often ignores project specifics (frameworks, naming conventions, tools). These exist not just to ease our lives, but also to keep projects safe & maintainable
June 10, 2025 at 12:00 PM
I agree 100%. AI code autocomplete (even in cases where there isn't any hidden bug on it) more often ignores project specifics (frameworks, naming conventions, tools). These exist not just to ease our lives, but also to keep projects safe & maintainable
LLMs are extremely useful if you know how to use them , and extremely misleading if you don't. Either way, what they can actually do is far below what most of their creators claim they can already do. As for the future, we'll see. The claims of AI hype promoters are disproven every day.
April 27, 2025 at 4:24 PM
LLMs are extremely useful if you know how to use them , and extremely misleading if you don't. Either way, what they can actually do is far below what most of their creators claim they can already do. As for the future, we'll see. The claims of AI hype promoters are disproven every day.