Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

ChatGPT and Codex Down

status.openai.com

Post reply on HN
Print view View original

11–15 of 15 posts

Re: ChatGPT and Codex Down

#11

Post by lrvick »

virgildotcodes wrote: post #7
lrvick wrote: post #5

Burn baby burn. Meanwhile, you can always buy hardware like a Strix Halo and have local LLMs that no third party can take away from you.

I really wish local models could compete with Codex, but they are miles apart for now. I'm not sure how they would ever not be, unless local models at some point in the future catch up to the current state of 5.4 high. Even then, the frontier models would likely have improved by an equivalent degree, so you'd again be faced with the same choice of deciding between a dramatically less effective local tool and a far mo…

I do linux kernel, compiler, and operating system dev with Qwen3.5 122b running locally on a Strix Halo 128G ad 35t/s. Pretty much the most complex software problems one can work on.

I think a lot of people just want to put in a credit card and press an easy button.

Re: ChatGPT and Codex Down

#12

Post by lrvick »

andyfilms1 wrote: post #9
lrvick wrote: post #5

Burn baby burn. Meanwhile, you can always buy hardware like a Strix Halo and have local LLMs that no third party can take away from you.

Sure, but unless you're training them yourself they can still be compromised with poisoning or bias. They're still black boxes even if you're running them locally.

Obviously, and that is no different than remote models. You do not and should not ever trust an LLM, but with proper handling they can still be super useful.

You give LLMs a dedicated OS to work in, let them do research or debugging and commit to branches, review and clean up those branches as you like from a trusted OS, then sign the commits and mark a PR as ready for review.

Re: ChatGPT and Codex Down

#13

Post by lrvick »

Archit3ch wrote: post #10
lrvick wrote: post #5

Burn baby burn. Meanwhile, you can always buy hardware like a Strix Halo and have local LLMs that no third party can take away from you.

What's the alternative to frontier models? Disk-streamed GLM 5.1? By the time you get a single response back, the API will be back up.

35t/s on Qwen3.5 122b on a Strix Halo. The local stuff works great now. Stop giving the corpo monopolists money.

Re: ChatGPT and Codex Down

#14

Post by virgildotcodes »

lrvick wrote: post #11
virgildotcodes wrote: post #7

Earlier quoted context omitted.

I really wish local models could compete with Codex, but they are miles apart for now. I'm not sure how they would ever not be, unless local models at some point in the future catch up to the current state of 5.4 high. Even then, the frontier models would likely have improved by an equivalent degree, so you'd again be faced with the same choice of deciding between a dramatically less effective local tool and a far mo…

I do linux kernel, compiler, and operating system dev with Qwen3.5 122b running locally on a Strix Halo 128G ad 35t/s. Pretty much the most complex software problems one can work on. I think a lot of people just want to put in a credit card and press an easy button.

Yeah the easy button, if translated to a more capable model that requires less hand holding, manual correction and consistently produces better quality code, is of course the point. You wouldn't want to go from Qwen3.5 122b back to GPT 3.5 for coding assistance.

People can definitely be productive with less powerful models. Supermaven or Cursor's tab autocomplete models from a year ago were already a huge boost over the pre-AI days. They just don't have the same capabilities as the leading models.

Curious if you've tried Gpt 5.4 High through Codex to compare for your use case?

Re: ChatGPT and Codex Down

#15

Post by lrvick »

virgildotcodes wrote: post #14
lrvick wrote: post #11

Earlier quoted context omitted.

I do linux kernel, compiler, and operating system dev with Qwen3.5 122b running locally on a Strix Halo 128G ad 35t/s. Pretty much the most complex software problems one can work on. I think a lot of people just want to put in a credit card and press an easy button.

Yeah the easy button, if translated to a more capable model that requires less hand holding, manual correction and consistently produces better quality code, is of course the point. You wouldn't want to go from Qwen3.5 122b back to GPT 3.5 for coding assistance. People can definitely be productive with less powerful models. Supermaven or Cursor's tab autocomplete models from a year ago were already a huge boost over…

My use cases requires sovereignty with all data on my property, offline use, and the ability to customize models to let me do security research without refusals.

Anthropic and OpenAI have nothing to offer people like me that require freedom and privacy.

Post reply on HN