Eddi

@0xSero @RayFernando1337 Well, I found out tha quantizing if you only have one RTX Pro 6000 needs RAM as well

@RayFernando1337 Graphic cards nothing else matter

Local AI can be so good, but you’d need about 12k USD to get it. Then it’s not so great Here’s a Q4 of Qwen3.5-262B-REAP Weights 131GB KVCache 50GB 256,000 context 350 tokens/s prefill warmup 4,000 tokens/s prefill cache 36 tokens/s generation Vision enabled REAP is good

@badlogicgames That’s invasive. Crazy shit

that's hilarious notes.zachmanson.com/copilot-edited…

@enjojoyy So much shit has been killed one might wonder what is even alive at this point.

claude computer killed openclaw macos killed linux ios killed grapheneOS aws killed home servers visa killed ethereum telegram killed signal cursor killed vscodium arc killed chrome what else?

@badlogicgames but the graph looks pretty!

I have one of these too. libGDX was part of many products that made a ton of money and still has a sizeable community. It's 16 years old. It's hundreds of thousands of lines of code, works on desktop, mobile, web. pi is 6 months old and trivial in comparison. Stars mean nothing.

@badlogicgames Maybe time for pi-git?🤔

github api rate limits are miserable if you want to automate large scale unclankering.

@0xSero Put the donation link into the video description;)

Tons of people followed me in the last 3 days, here's my most important video for learning how to work like me. youtu.be/VgR66ybAtdg

Let it run on Friday and burned quite some tokens (90 bucks of GPT-5.4-mini). My error was, that I had a multistep process, where one half needed to be done first and then the other half needed the loop. But, as the auto-research feature is designed, it looped through both steps. So, yeah, might be something to be aware of.

going to try pi-autoresearch to see if we can optimize startup time and memory usage in pi that way. excited to burn some tokens for something worthwhile! github.com/davebcn87/pi-a…

@RhysSullivan Building a validation layer that the model can check against during the refactor. I think this will help, but results won't come in till second week of April. Then I think I will write an article about it.

are any of the models actually good at doing large refactors? i have to spend so much time fighting with them to not take shortcuts and actually make large changes to code

@badlogicgames Love that. Is it a plain refactor or are you planning on reworking the event handlers?

People of pi. I'm going to break the extension API hard. Specifically, business logic (event handlers, custom tools/compaction/etc.) needs to be split off from the ui layer. it will likely not be a massive amount of work to migrate an existing extension, but it will hurt a little.

Ich bin grundsätzlich auch skeptisch mit Rückholaktionen, aber du brauchst einen Spieler wie Sancho und die Qualität die er hat, auch jetzt noch, kostet locker nochmal 20 Millionen. Die kann man sich sparen, daher wäre das, vorausgesetzt stark leistungsbezogene Bezüge, schlicht schlau.

@_Eddited_ @Sky_Marlon89 Es soll also wieder eine Rückholaktion werden. Dieser Verein wird nie aus den Fehlern lernen, vorallem ruiniert man komplett die Reputation dieses Vereins indem man eine Lachnummer die bei drei Vereinen gefloppt ist zum dritten Mal holt.

Meinung: Wenn der #BVB Jadon #Sancho ablösefrei bekommen kann und dieser sich wirklich mit „nur“ 5-6 Mio Gehalt zufrieden gibt, dann ist das in meinen Augen immer noch ein Topdeal ohne großes Risiko. Du weißt, was du mit ihm bekommst. Der Spieler fühlt sich in Dortmund wohl, kennt jeden Stein und die Fans lieben ihn. Ich habe ihn gestern Abend für Aston Villa in der Europa League gesehen und fand ihn stark. Bei seiner letzten Leihe zum BVB waren die Statistiken vielleicht nicht unfassbar berauschend, aber sein Wert für das Dortmunder Spiel auf dem Weg ins Champions League-Finale war enorm groß, wenn man die Spiele alle gesehen und genau hingeschaut hat. Wenn du ihn ans Leistungslimit bekommst, kann er für mich immer noch einer der besten Tempodribbbler Europas sein. Außerdem würde ein Transfer - zumindest beim Großteil der Fans - mal wieder so etwas wie Euphorie entfachen. Let‘s see ⏳⚫️🟡

@dinhojr144 @Sky_Marlon89 Gut, irgendwas macht er scheinbar richtig, wenn er zum UEL POTM gewählt wird. Und ich sehe ihn auch eher als Zehner oder so wie Kovac auch gerade die Stürmer hinter Guirassy spielen lässt. Aber wir werden sehen.

@_Eddited_ @Sky_Marlon89 Genau das Profil was er nicht hat🤣🤦🏾‍♂️ Er ist mit der langsamste Flügelspieler in der Premier league und gewinnt kaum ein 1vs1 noch dazu hat er nur 1 Scorer und es ist März

@adahstwt control

I'm a vibe coder, scare me with one word.

@ivanfioravanti How are you fans doing? :D But nice to see, just looking forward to an M5 Ultra Mac Studio to avoid the noise^^

Video in normal speed of Pi coding agent running on M5 Max with MLX Qwen3.5-35B-A3B-8bit model. Faster prefill is game changer here! 🔥 Let me try Qwen-Next-Coder-5bit now!

@0xSero have a new workstation with one RTX Pro 6000 I am not using properly atm due to lack of time. I don't know what is possible, but maybe you could somehow use the compute partially. Just don't have that much system RAM (64GB) due to RDIMM RAM being impossible to buy. Just DM me

Putting out a wish to the universe. I need more compute, if I can get more I will make sure every machine from a small phone to a bootstrapped RTX 3090 node can run frontier intelligence fast with minimal intelligence loss. I have hit page 2 of huggingface, released 3 model family compressions and got GLM-4.7 on a MacBook huggingface.co/0xsero My beast just isn’t enough and I already spent 2k usd on renting GPUs on top of credits provided by Prime intellect and Hotaisle. ——— If you believe in what I do help me get this to Nvidia, maybe they will bless me with the pewter to keep making local AI more accessible 🙏

this guy has 29 models on huggingface at page 2 ranking. no lab behind him. no sponsorship. $2,000 from his own pocket on GPU rentals. he compressed GLM-4.7 to run on a MacBook and quantized Nemotron Super the week it dropped. all public. all free. nvidia is a trillion dollar company with hundreds of teams but they are not the ones quantizing models middle of the night and pushing them out before sunrise. if nvidia stopped tomorrow their employees stop working. people like @0xSero would not. that is the difference between a paycheck and a mission. @NVIDIAAI you talk about making AI accessible. the people actually doing it are right here. 29 models deep burning their own compute with no ask except more hardware to keep going. you do not need to build another program. just look at who is already building for you. one GPU to this man would produce more public value than a hundred internal sprints. i am not asking for charity. i am asking you to invest in someone who already proved it.

@VadimYuryev Don’t you think it might be overheating? Seen one video where they literally just put a thermal pad on the chip and got 14% more performance

Can 8GB of Unified Memory on MacBook Neo match 16GB on Windows Laptops? NO! Did a RAM stress test by exporting photos in Lightroom which is VERY RAM-hungry and opened browser tabs. The slow SSD speed sabotages the swap memory performance. Full video: youtu.be/f9fhCMBIbis

every time I try to challenge any AI to build complex projects, I find appreciation for what we humans do just naturally in our profession. And how crazy impressive the brain is.

@tomfgoodwin @MatthewParrott If the context part makes sense to you, then skills should as well. All goes into context, even if it is in a subsequent discovery step.

@MatthewParrott The context window makes sense. The skill part does not. And yes. One wonders with the effort to bring the strands together and be coherent , if much is gained except for spending tokens

I’m surely being stupid. But if AI is rather unconstrained by expertise or capacity or to some extent speed Why do we need to divide tasks or departments to 9 agents ( the marketing agent, the optimization agent etc ) to each do one thing. And then another agent to manage the swarm. Cant one agent just be doing it all you know. It seems very skeuomorphic. Will we have HR agents to make sure the agent agents are being looked after ? A office canteen manager agent to feed the agents ? Seems daft

@badlogicgames I only try to inject state and instructions mid session. Working on something like a "subconsciousness" loop with a small model that retrieves learnings that apply to the agent output to interrupt bad behavior and maybe steer it a bit better :D

this is immensely illegal