Coding with OpenAI o1

[q]

The problem is later, when the conversation continues and more requirements are introduced, it forgets the first requirements.

[gohkairen]

damm we have computer scientist putting computer scientist out of job before GTA 6

[palovalova]

Sir, I need you to stop. I need to feed my family.

[chromaflow]

Honestly, big win to OpenAI for this new model being available today. I really appreciate the “talk about it when we release and not before” approach.

[universemaster]

It would be great to finally release advanced voice mode too all paying customers, please!

[atharvparlikar]

Ok alright, I got scared for a second. we are safe for now

[apw-nzsg]

The calm soothing lofi like music they put in the background so that the programmers watching this won't panic

[Achievly]

Watching him work to run the code makes me love Claud artifacts that much more!!

[GamingSmithJohn]

This feels like more of a commercial video rather than a walkthrough. It's like this video is included in the OpenAi - Apple partnership deal so the latter can include the mac, the resemblance to Steve Jobs's iconic fit and of course we sprinkle a good actor that looks like your typical IT nerd to read some lines.

[scrollop]

"Which game should we ask for?"

"Let's ask for the game that is in many benchmarks and LLM tests." Uhuh

[naiqiangqian]

i think it is calling the agents underneath, the calling of each agents becomes the thinking steps...it makes the waiting less frustrating... very smart product design...

[NK-iwrq]

Altman sold his soul. OpenAi was supposed to be open source and accessible to all, and now its been turned into a for-profit company.

[m.f.mfazrin]

Claude Sonnet 3.5 already doing excellent job in this area

[dr_rd]

Basically, the model is no longer biased toward completing the shortest, most superficial task. Instead, it now prioritizes engaging in deeper “reasoning” and fully addressing all the initial multi-step requirements in a coding prompt. How many examples were needed to influence this shift in behavior? Has this behavior generalized to other areas, such as handling subsequent multi-step requirements? Or will the new model struggle when asked to integrate or expand new functionality while maintaining backward compatibility? How many additional examples would be required to influence these other behaviors?

[calvinhobbes]

It wouldn't have overlapped had he used Emacs.

[desireco]

To me this is a wrong way to go about this, you are expecting perfect answer to perfect prompt, instead of getting to desired result through interactive steps. It is cool but without human in the loop, it is made unneccesarily difficult for AI to produce result

[pmarreck]

Programmer skill, still matters.
Every time I guide it to, for example, write unit tests for its generated code (which it does NOT do by default), it codes better.
Every time I guide it to use TDD (which, again, it does NOT do by default), it codes better.
Every time I tell it to break down larger functions into simpler ones and test those, it codes better.
Every time I tell it to compute a value (using Python) instead of guessing it, it does better.
(And this only applies to languages it can run itself, by the way, which is currently only Python. I hope there will be more soon!)
The same things that help human developers, help LLM coding, but since the LLM has been trained on bad development practices and bad code, it will not do it by default and requires skilled guidance from an experienced developer for maximal utility.
And even then it will still screw up often, such as asking for a regex that replaces any combination of a newline with a space anywhere around it with a space, will always include newlines without spaces even though it shouldn't... encountered that one just yesterday.

[danielvichinyan]

ChatGPT is not a tool to rely on. You have to test it and proof-read it multiple times. In many cases, the code is inefficient, has obvious mistakes, missing validations.

Another great problem is that AI does not have the context of the project. Okay, you can do some simple pages and functions. But you could never build a whole system from scratch. Also, it’s impossible to maintain it long-term using just ChatGPT.

[ErnestoConfused]

"hoover"

lol, but in all seriousness, great work y'all!

[rupertcroune]

"Unlike the older models it is thinking" - I think you mean it is loading and showing on screen what it is doing in the background....