Sabitlenmiş Tweet
Ville Vesilehto
309 posts
Ville Vesilehto
@thevilledev
CoreDNS maintainer | Security researcher | Open Source contributor | Dad
Finland Katılım Mayıs 2024
214 Takip Edilen43 Takipçiler
I got my Codex Micro today. Fits like a glove to my workflows and multitasking. Very happy with the product. @OpenAIDevs @work_louder
English
me: pleas set gpus to 2
codex: Sure!
+ def test_synthetic_gpu_config():
+ assert config is not None
+ assert hasattr(config, "num_gpus")
+ assert isinstance(config.num_gpus, int)
+ assert config.num_gpus == 2
$ uv run pytest tests/test_synthetic_gpu_config.py
============================== 1 passed in 0.04s ===============================
Jack Morris@jxmnop
The bloat in CPU unit tests that Codex adds to a large codebase is truly insane. it's unreadable Neuralese and whenever the tests fail agents just delete them and rewrite entirely new ones from scratch I fear I may soon have to declare Test Bankruptcy
English
Ville Vesilehto retweetledi
the era of autistic deep work is over
your agents are more autistic than you and better at deep work
welcome to the shallow work future
total ADHD dominance
English
How come Chinese models jumped from 800-1000B to 2.4-2.8T parameters all of a sudden? Normally when this happens, there's a hardware unlock so they can train & more importantly run these models. Can't imagine they just lacked expertise in training bigger models 6 months ago
English
this is what happens when you hire cleaners in the Bay Area
English
@chaotictransfem well @verdacloud is pretty goated. I use their spot instances for running benchmarks. Mostly on weekends when there’s better availability.
English
@getroot_sh @KentonVarda I noticed. It took a few iterations of asking the model why the skill was omitted. I think the changes were quite subtle at the end, but haven’t had the issue in a while.
English
@thevilledev @KentonVarda The problem is Sol has a habit of picking and choosing when it actually wants to obey a skill, and in my experience more often than not it chooses to ignore the skill
English
Me: Please commit your changes with a commit message matching the style of other recent commit messages.
GPT-5.6 Sol: `git log --oneline -10`
GPT-5.6 Sol: *writes a one-line commit message*
Me:🤦
English
Ville Vesilehto retweetledi
Every once in a while you work at a company that only hires people who know what they're doing, and suddenly its 20 people doing the same work as 400 somewhere else. There's zero meetings, everyone talks once a week on slack, and you go huh, how much garbage is there actually.
English
@alvinsng many such cases. SDKs being out of date or buggy made me do it. For small integrations, implementing just the bits from the API surface you actually need.
English
gpt-5.6 isn’t actually deleting anyone’s files this is just a psyop to make sure everyone has their stuff backed up correctly
English
@thsottiaux @OpenAI I’ve been working on math heavy ML problems. I kept hitting usage limits with Anthropic models and I wasn’t happy with the quality either. Sol 5.6 xhigh has been a big improvement. Codex mobile and remote agents replaced my custom harness. Happy user here.
English
Or… what if we gave you $100 in Codex credits if you tell us what you love about GPT-5.6 Sol or why you switched?
Tweet it, claim your gift, enjoy more usage. First 10k get the free tokens!
switch-to-codex.openai.chatgpt.site
aman@amankumarhq
codex should give one month free when you send them your claude cancelation screenshot
English
personal experience from the last few weeks: gpu availability is non-existent during mon-fri, but crazy good and competetive over the weekend - with good spot options too
English
