“our team was blown away by how much Codex was able to accelerate its own development.”
At what point will LLMs be autonomously self creating new versions of themselves?
GPT-5.3-Codex
571–580 of 634 posts
Re: GPT-5.3-Codex
#572Page me when codex can run the right version of node. Are we all changing the system node version to match the current project again? [shell_environment_policy] inherit = "all" experimental_use_profile = true [shell_environment_policy.set] NVM_DIR = "[redacted]" PATH = "[redacted]"
It worked for me after I configured mise. I needed the mise setup in both `.zprofile` and `.zshrc` for Codex to pick it up. I think mise sets up itself in one of those by default, but Codex uses the other. I expect the same problem would present itself with nvm.
I.e. `eval "$(/Users/max/.local/bin/mise activate zsh)"` in `.zprofile` and `.zshrc`
Then Codex will respect whatever node you've set as default, e.g.:
mise install node@24
mise use -g node@24
Re: GPT-5.3-Codex
#573Whats interesting to me is that these gpt-5.3 and opus-4.6 are diverging philosophically and really in the same way that actual engineers and orgs have diverged philosophically With Codex (5.3), the framing is an interactive collaborator: you steer it mid-execution, stay in the loop, course-correct as it works. With Opus 4.6, the emphasis is the opposite: a more autonomous, agentic, thoughtful system that plans deepl…
I'm personally 100% convinced (assuming prices stay reasonable) that the Codex approach is here to stay. Having a human in the loop eliminates all the problems that LLMs have and continously reviewing small'ish chunks of code works really well from my experience. It saves so much time having Codex do all the plumbing so you can focus on the actual "core" part of a feature. LLMs still (and I doubt that changes) can't…
I've been using codex for one week and I have been the most productive I have ever been. Small prs, tight rules, I get almost exactly what I want. Things tend to go sideways when scope creeps into my request. But I just close the PR instead of fighting with the agent. In one week: 28 prs, 26 merged. Absolutely unreal.
Re: GPT-5.3-Codex
#574Earlier quoted context omitted.
I didnt say one shot it, coding agents have been out for more than couple years and yet we cant point to single Good piece of software built by it.
"Good" is obviously subjective but this mentality is so interesting because at my big tech company most of our software today is written by agents. From my perspective, comments like these read as people having their head stuck in the sand (no offense, I might be missing something.)
Show me whats been built by agents
Re: GPT-5.3-Codex
#575Earlier quoted context omitted.
The free market has simply decided these consumers are not as relevant as the others.
Not really. Investors with hundreds of billions of dollars have decided it. The process by which capital has been allocated the way it has isn't some mathematically natural or optimal thing. Our market is far from free.
Saying "investors with hundreds of billions decided it" makes it sound like a few people just chose the outcome, when in reality prices and capital move because millions of consumers, companies, workers, and smaller investors keep making choices every day. Big investors only make money if their decisions match what people actually want; they can't just command success. If they guess wrong, others profit by allocating money better, so having influence isn't the same as having control.
The system isn't mathematically perfect, but that doesn't make it arbitrary. It works through an evolutionary process: bad bets lose money, better ones gain more resources.
Any claim that the outcome is suboptimal only really means something if the claimant can point to a specific alternative that would reliably do better under the same conditions. Otherwise critics are mostly just expressing personal frustration with the outcome.
Re: GPT-5.3-Codex
#576It is absurd to release 5.3-Codex before first releasing 5.3. Also, there is no reason for OpenAI and Anthropic to be trying to one-up each other's releases on the same day. It is hell for the reader.
Let them fight, that's how we as users get more tokens for less money. If only all markets were so competitive...
Sure. At this point, what matters more to me imho is in effect the Plan stage where I refine my crude specification, iteratively and repeatedly, until I work out all the flaws and minimally necessary details in it. This is hard to do and uses up a lot of tokens, and it is where a lot of my initial effort goes. I have literally repeatedly exhausted my token quota just in this stage alone. I can take then take this refined specification to even a dumb agent from a year ago, and it would have no trouble producing decent code for it. This refined spec is what is equally important to me as the code. Tests, both unit and integration, also go hand in hand with the spec, although they're less important to me if I am carefully reviewing every line of code, and more important when vibe coding instead.
Re: GPT-5.3-Codex
#577Earlier quoted context omitted.
That is where the money is.
This. I think software development is the best usecase for AI yet. I use it almost daily at work and it's a huge help. Enterprise customers will happily pay even 100$/mo subscriptions and it has a clear value proposition that can be decently verified.
Revenue should not be confused with profit. The large AI companies must easily be spending more on compute than they're making from a $20-200/mo subscription. In the best case it might break even for the AI companies. There is no way that they're actually earning a profit from these subscriptions at this time.
Re: GPT-5.3-Codex
#578I have found GPT 5.3-Codex to do exceedingly well when working with graphics rendering pipelines. They must have better training data or RL approaches than Antropic as I have given the same prompt and config to Opus 4.6 and it seems to have added unwanted rendering artifacts. This may be just an issue specific to my use case, but wonder since OpenAI is partners with MSFT, which makes lots of games, that this may be an area they heavily invested in
Re: GPT-5.3-Codex
#579Whats interesting to me is that these gpt-5.3 and opus-4.6 are diverging philosophically and really in the same way that actual engineers and orgs have diverged philosophically With Codex (5.3), the framing is an interactive collaborator: you steer it mid-execution, stay in the loop, course-correct as it works. With Opus 4.6, the emphasis is the opposite: a more autonomous, agentic, thoughtful system that plans deepl…
I'm personally 100% convinced (assuming prices stay reasonable) that the Codex approach is here to stay. Having a human in the loop eliminates all the problems that LLMs have and continously reviewing small'ish chunks of code works really well from my experience. It saves so much time having Codex do all the plumbing so you can focus on the actual "core" part of a feature. LLMs still (and I doubt that changes) can't…
I will personally never consider using an agent that can't be easily pushed toward working on its own for long periods (hours) at a time. It's a total waste of time for me to babysit the LLM.
Re: GPT-5.3-Codex
#580Why does OpenAI have a separate model for coding (Codex) but Anthropic uses the same model for chatbots and coding?
They seem to be slowly moving away from having a separate coding model. With this release, they're calling the model Codex but expressly mention that it's also supposed to be more suitable than GPT 5.2 for general use.
I remember a paper coming out a while back that said training the models to code made them much better at normal tasks. It improved their logic etc.