New articles are saying chat gpt 5 not comimg out until 2025 this is way too long as 4 refuses at coding 10s of thousands of lines
1 Like
Claude 4 will kill chat gpt soon as claude 3.5 sonnet is already better than 4o
1 Like
Can you clarify what you mean by this? Do you mean that GPT4 isnt as useful for large coding projects as Claude 3.5 Sonnet?
Sonnet is already destroying chat gpt 4o and the CTO mira woman was interviewed and saying its going to take a year and a half bahhahah this is a joke as claude 4.0 will be outsooner to keep stealing chat gpt customers also chat gpt just lost the best mind italya as he is starting his new super intelligent ai company
1 Like
I didn’t mean it in a facetious way, I was genuinely asking. I figured you might know something about Sonnet’s capabilities that I did not.
Altman has been hinting in lots of interviews the next model won’t be called gpt-5.
There are more nuances than creating a slightly better coding model.
Why do you care which company / model “dies”? As end users we benefit from the best model.
Nah, The model used to train GPT 5 is so much bigger than anything out there right now. It is not even close. Also Claude is not available for more than half the world’s population. Claude does not allow brazil to use its services because of fear of BCC 927. So basically, even if they have a decent system, it is mute. Another thing is that claude refuses to answer a huge amount of questions that GPT4o does answer
The articles are misquoting Mira. She said that “systems available in 18 months will be at the Ph.D. level.” She did not say they would pause training, delay systems, or that GPT 5 was not coming out. Too many people are using AI to analyze news and not actually looking at the actual video interview. Calm down.
1 Like
Coding involves the iterative design process. You’re not going to have GPT-5 code thousands of lines of code which will just run off the bat without iterative design. I think we’re still quite a bit away from that level of intelligence from large language models. The knowledge and reasoning skills required would have to be exceptional and knowledge up to date with the versions of various libraries you’re using. An improvement in it’s coding abilities in terms of reasoning would be most useful and for better contextual memory and we will see this with GPT-5. It will likely be able to have longer outputs in some scenarios but it’s still going to be an iterative process. With tens of thousands of lines of code it can become too complex to complete in one output. Also, you should utilize the separation of concerns approach in your projects with good structure and documentation. With this the models should be capable of discerning which documents it needs to take into consideration each step of the way. There’s no need to have tens of thousands of lines of code taken into consideration in a single message. You need to break things down into more refined, simple, and maintainable components.
Nah the god like fallen angel chips are arriving next generation architecture and 10000 lines of code in one message is the god like power we need lower case g of course
I just signed up for an account, thanks for posting about this. I will probably cancel my paid ChatGPT account and switch over. Can Claude track code changes across an entire project and maintain dependencies and awareness when helping out with coding?