Omar Costilla Reyes, PhD

@Yuchenj_UW Capitalism. its not about AI OSS labs, the entire society revolves around the idea of individualism above anything else.

It's weird that the US still doesn’t have a truly competitive open-source model lab. It’s clearly not a money problem. Several neolabs have raised billions. It’s not a compute problem. US labs have easier access to B200s/B300s than Chinese labs. So what is the issue?

Please stop submitting AI slop code pull requests to RPCS3. We will start banning those who do without disclosing. There are plenty of resources online to learn how to debug and code instead of generating slop that you don't understand and that doesn't work.

@Romy_Holland this is exactly how i remember it before the pandemic. i hope tourism has not yet destroyed it.

it’s actually crazy that tokyo is both the largest city in the world and is plainly the most civilized. there is no crime, it’s perfectly clean, transit is incredible, i’ve yet to see a single person doing anything remotely disorderly. most world class cities are western and it’s really cool to get a radically different aesthetic while still getting top tier amenities. i’m kind of obsessed with this place. hard to explain, but the feeling of being here is incredibly distinctive and kind of dreamlike.

@trq212 @bcherny So the newest anthropic innovation is not to use MD files but html? Have you guys ran out of ideas? There is one simple rule whatever wastes context window is uncessary, html is. You are optimizing for human understanding. Machines don't care about it.

@sheriyuo I am not sure in which world you live in. Everyone is using AI to write, the difference is between those who read the generation and those who don't. Back in the day we needed years to write a PhD thesis. Most students are done in a month today, without learning how to write.

If my coauthor ever uses an AI agent to write our paper again, I’m actually going to kill him. First glance: every section looks legit, the LaTeX formulas are chef’s kiss beautiful. Closer look: the equations are pretty, but the writing is straight-up garbage — like an undergrad feeding a 2022 LLM and calling it research. Now I get to rewrite the entire thing into a final draft. This garbage isn’t even as good as just letting DeepSeek Chat do it from the start. 😑

@thsottiaux @nicdunz not everything in life is about money tibo.

@nicdunz How much would you pay for this

codex should do this

@flaczip Welcome to the age of AI

neurips paper fully baked and ready to submit 3 days before the deadline what have i become

@Linahuaa Did chatgpt wrote everything you just said? Sound like it.

The average Korean isn't anywhere near this. Even the average Korean with plastic surgery isn't anywhere near this. Even the top 10% Korean WITH top tier plastic surgery isn't anywhere near this. Half of young Koreans do plastic surgery- and if it was this easy for random modded Korean girls to literally stunlock a professional sports commentator with their mere existence, then every fucking man would be learning Korean right now and strive to become a successful sexpat. Guys have become turbo porn-brained. They say women have delusional standards, but they are even more fucking delulu

@ajambrosino I'd advice you to do 2 versions of the app. 1 that has essential minimum features just as in early days and 2 for people looking to see and use all features. I really don't care about any if your new shiny features I can do with the basics. What you have added have slow me down.

what should we do in the next 3 months?

@Yampeleg The real question is why codex folks have not yet released and official Linux app.

Just told GPT-5.5 to port and compile the codex app to Linux. 1. WTF. It actually worked. 2. WTF. That thing is incredible.

@amasad Here is the big difference: they see an opportunity. You see a problem. I've never heard a founder before complaining about growth. You are the first one.

It's honestly impressive that GitHub kept the service up at all, given this kind of growth. I predicted this years ago: Free services will become untenable with the advent of human-level bots. Worth exploring micro-payments: Even cents per git push might be enough to reduce spam and make this sustainable. Maybe powered by Bitcoin to keep this open and accessible (as opposed to KYCing users).

@aiandcloud Ollama has been bad since always. It's the easier to install but the worst to work with long term. @lmstudio it's the only reliable serious solution. The other thing that a lot of people don't know is that Ollama came out from YC... So it tells a lot of the folks runing it.

Ollama 是怎么样从开发者最喜欢的本地工具，一下子变得众人唾弃的呢 🤣

folks at @Google have a lot to work on to catch up.

this is how @antigravity look like today when working with gemini 3.1 pro. Not sure if the agent will ever finish to marge some basic work to main. This while folks at the @OpenAIDevs codex team are cooking. Is it really hard to understand user experience and make this useful?

@fchollet google did the same. gemini 3.1 its useless for me, the model vastly under-performs against the so called benchmarks they tested it with, we need an independent org to test these models in a realistic way so we stop these miss leading benchmarks.

The new model from Meta is already looking like a disappointment: overoptimized for public benchmark numbers at the detriment of everything else. Knowing how to evaluate models in a way that correlates with actual usefulness is a core competency for AI labs, and any new lab is unlikely to be successful without first figuring that out.

hey @thsottiaux, have you investigated the issue on codex where the agent does some work, then the codespace is different to git main due to those changes, but the agent refuses to continue coding due to a "dirty workspace"? this is a consistent, weird behavior i am getting.

I love the term "AI slop" it summarizes our criticism, resentment and current understanding of present AI. At this rate of advancement, we will soon be leaving this adjective. My worry is concentration of AI capabilities. We need much more competition before we regret it.

@ilyasergey @andrew_n_carr By AI, unfortunately.

Are paper rejections really that bad? My papers get read by ~3 people on average. Each rejection means a resubmission, which means 3 more readers. After 4 rejections, that's double-digit readership.

@kimmonismus Can you confirm this is true @steipete and not yet another cherry picked benchmark? @openclaw gives a warning for models below 300b including this one so not sure who to listen, seems contradictory, given that @NVIDIAAIDev I assumed helped the oc team to harden sec.

Nemotron remains completely underrepresented. NVIDIA has cooked up a storm, and most people are still ignoring it.