Joady

@duborges Agent deck does this for me

1) i need a MacOS open-source software 2) that I can easily work on 8 terminals 3) have a files tree with reliable editor - vs code: not good for many terminals - cmux: not good for editing files - tmux: no time to learn on yt videos what am i missing?

@almonk I think that's mostly because a lot of people are selling a dream that isn't quite real. If you used to using a claude code and Codex you're not gonna be blown away by GLM 5.2, no matter the harness used. Good for quick jobs with an exact prompt, but not long or hard tasks

What harness are people using GLM 5.2 in because my results so far are pretty lacklustre

Honestly GPT 5.5 low and possibly Opus 4.8 low are significantly cheaper than GLM 5.2 high. Considering that you can't run it on consumer hardware at anything but the most quantized size, which is going to heavily degrade the performance compared to the benchmarks that are floating around, I don't see any reason to use this over those other models.

GLM-5.2 scores pretty well with max reasoning actually beating GPT-5.5-low and Opus 4.8 low on DeepSWE but it still needs to work on reasoning efficiency

An awful lot of the code quality comes down to the custom instructions and the layout of your project. I tend to build very modular codebases and use very strict instructions and so I get quite good code, not always perfect but usually quite good. Getting Codex to review Claude Code's work and vice versa also helps with quality

if you think models are outputting good code you’re bad at coding

Totally agree, I use them both. I prefer planning with Claude code and then if you do you're planning well enough you could actually use a lot of different models for the implementation. Especially if it's really well planned. But Codex is great Claud code is great use, whatever you like

claude vs codex truly is the new apple vs android it's completely retarded because both are fine both can learn from each other but holy fuck do people love to pick sides

@femalebodybuil6 Honestly I'd be willing to pay the $700,000

$700,000 or a date with Kim Kardashian? 🤔😅.

@philippilk @GiorgiaMeloni By "Finished" do you mean everybody in Italy is rallying around behind her and that includes even supporters of her political opponents? Because I'm in Italy right now and that's exactly what's happening. Trump has made her more popular than ever here

The destruction of @GiorgiaMeloni was a controlled destruction. The Europeans realised that America were not going to help with their war and so she became an annoyance. They wound her up like a toy doll to do her ridiculous little dance and now she’s finished. 🤷‍♂️

Gemini is months old and no longer SOTA, so you could argue that GLM 5.2 is in the ballpark to compete with it, but Opus 4.8, it's not remotely close to. Opus is another level, doesn't get stuck in loops, doesn't forget what to do, understands context and loosely worded prompts. It's a year ahead by my reckoning

@Joadys @itsolelehmann However they got there, GLM 5.2 beats Gemini, Opus 4.8 by a mile. Agreed they're hampered by their domestic chips, but a year ago, they had nothing. Chinese are incredibly resourceful.

this might be the best news you’ll get out of the whole US-China AI race Elon just guessed that Chinese open-source catches US frontier models by Q1 next year. then the founder of Zai (the lab behind GLM) jumped in and said it won't even take that long. he thinks they hit the frontier this year. 2026. GLM 5.2 is *already* the 3rd best model in the world right now, behind only fable and GPT 5.5. it's also the number one open source model on the planet. and it already beats GPT 5.5 on coding benchmarks at 1/6th the cost per token. this is amazing because every company faces the same tradeoff right now: frontier models are smart enough to trust with real work, but too expensive to run on every user, every task, all day. (Uber literally burned through its entire 2026 AI budget in the first 4 months this year, then had to cap every employee at $1,500/month) the result: you're forced to give the cheaper models all the high-volume jobs. but you settle for worse output to keep the bill sane. well, it looks like that tradeoff is about to disappear before the year ends. the cheap frontier is coming.

@Selene_Mariposa Tell us you've never been to Europe without telling us, you've never been to Europe

The World Cup has turned America into a discovery channel for the rest of the world. And they are not handling it well. In the best possible way. Here is what they are discovering: Free public restrooms. Europeans pay every time. Free water at every restaurant. Just appears. Free refills. Coffee. Sodas. Iced tea. Unlimited. Free chips and salsa before you even order. Free warm bread with dinner. Ice in drinks like civilized people. Air conditioning everywhere. Not a moral debate. A fact. Parking lots attached to the actual place you are going. Drive throughs where the food comes to the car while you sit in it. Ranch dressing by the gallon. Tex-Mex that cannot be explained only experienced. Dental care that actually works. Buccee’s. There are no words for Buccee’s. Then they found the grocery stores. Five of them within one mile. Each one the size of an aircraft hangar. Burgers. Steaks. Brisket. Ribs. Pulled pork. Lamb. Veal. Every cut of every animal ever domesticated by human civilization available in one refrigerated aisle at ten in the morning on a Tuesday. The Germans stood in the meat section for forty five minutes. In silence. Processing. They finally understand why we do not have trains. We have roads wide enough for the cars we actually drive. Parking lots the size of small European countries. Airports in every city worth visiting. Why would we need trains. The Germans are taking ranch home by the bottle. The Dutch found queso and briefly lost the ability to speak. The Japanese are photographing HEB like it is the Louvre. The Czechs are weeping in West, Texas. Welcome to America. Everything is free, enormous, air conditioned, comes with chips, and has five grocery stores within a mile that will sell you any cut of any animal you have ever imagined. Write that down. 🦋

@teortaxesTex It's so close to Opus that when asked it often claims to be Claude. I wonder how that could have happened?

GLM blows a big hole in my thesis that the Chinese have a catastrophic disadvantage in high-quality data. It's too close to Opus. They did this without spending billions. I don't know how. Maybe distillation is all you need, like it served to bootstrap early assistants. Wild.

"Stabbed him in the back over Iran" Translation Trump went to war half way accross the world for no reason other than Bibi convinced him to and said it would be over in 3 days. Trump has been desperately trying to get out of the war he was tricked into ever since. And the leader of an independant country not joining in this ridiculous and disasterous war is somehow betrayal... Such a beta take

Giorgia Meloni is all butt-hurt now that Trump publicly criticized her after she stabbed him in the back over Iran. She was the one who betrayed Trump, not the other way around.

@nullctl @Amank1412 metricool.com/what-is-engage…

@Joadys @Amank1412 wtf is this retard saying? Do you have any idea what engagement farming is?

If GLM 5.2 is truly near Opus 4.8 parity then does that mean frontier Chinese labs are only months behind an open source Mythos class model?

@themagaking Stupid war no-one but Bibi wanted, she chose wisely. What are you whinging about?

Italy sat out the entire Iran war. Refused U.S. bombers landing rights. Refused to send ships to Hormuz. So forgive me if I don’t lose sleep over what Prime Minister Meloni thinks about a photo‑op with Trump.

Hungary has no sea borders. Italy has the most in the entire Mediterranean. German NGOs drop people off constantly. This is a thing Hungary doesn't have to deal with. In addition the multi-party system in Italy means that, although she is the leader in Italy, she doesn't have a majority of either parliamentarians or the popular vote. She has a grouping of different potatoes to keep together and keep cohesive. Orbán didn't have to do that.

I’ve held back on Giorgia Meloni for long enough. She’s the leader of Italy and has had almost four years to do something to change it. She’s done nothing. She acts powerless. How come Orban could figure out how stop mass migration into Hungary and Meloni can’t? She is too busy up Zelenskyy’s ass all the time. Giorgia Meloni was elected by the right-wing on a right-wing agenda. She’s governing as a centrist with leftist tendencies. Nothing has been done about the problem of mass migration. The numbers still suck. They’ve gone down somewhat but not nearly enough. Where’s the naval blockade? Now she’s busy attacking Trump so she can eventually become President of the EU. She’s the only logical option because the Left is worse but she’s far from a good option to lead Italy. I am not sure she’ll have my vote next time. I vote in Italy because I’m a citizen and even when you don’t live there you’re entitled to a vote. Giorgia has lost mine.

@bankof_amERICA Of all the things that never happened, this never happened the most.

I let Claude take control of my computer and it just tried buying a random GoDaddy domain for $18k when i just simply asked it to fill out an excel sheet We have not reached AGI

@Kristinartz Butter and sour cream (and flat leaf parsley or chives too if possible)

I need to settle a debate: Milk or butter in mashed potatoes? 🍠

@Dr_Gingerballs You land right in the middle with this take. Of course SpaceX, Google and Anthropic are wrong and Ginger-anon is right on the money...

Dear moron, It will cost 30x more to do AI in space vs. on land. We are also building more datacenters than ever before. The primary difference is that land datacenters are real and take time, and space datacenters are imaginary grift. Thanks.

@jun_song Lets circle back in July. I expect a big leap from them with their next (1.5T) model and a bigger one for the even bigger models currently in training for release after that. Grok 4.3 is a very old pre-train and the last of their old architecture.

I highly value xAI as a vertical infrastructure, but one thing is for sure: Grok performs worse than even GLM 4.7, let alone GLM 5.2.

@babayagatwt Spacex. Datacentres in the cloud will be the only ones to avoid the anti AI sentiment and power limitations in the west. Terrafab builds the chips, Spacex builds the satellites and launches them XAI and competitors run in space with free electricity and no permitting

What's one tech company making the most important long-term bet right now?

@jamonholmgren Maybe try being specific. "Answer in under 100 words"

@Joadys Pretty sure 90% of my sessions include "be concise" tbh

I need my AI to reply to me like it’s tweeting. I can’t deal with this 5,000 word answer to a question