bifkn

I want to build weird, useful AI stuff around the things I actually care about. local models agents memory games training health experiments. notes. dashboards. failures. receipts. The goal: turn tinkering into experiments, notes, dashboards, failures, and receipts. 🧪 this is my lab.

@cwolferesearch In what domain? So far 27B handles all of my coding tasks significantly better than Gemma 4

Gemma-4 has received more attention than prior generations of Gemma, but I still somehow feel like these models are so underrated. I don't think people fully realize how good Gemma-4 models are, especially for their size.

@tawer1O Qwen’s game looks like much better game play. That’s been my experience too. Gemma has faster results, but the quality isn’t there. bifkn.vercel.app/lab/octopus-ar…

Comparison of Gemma 4 and Qwen 3.6 in the same coding task Same hardware, same prompt, comparable model size > gemma 4 31b: 27 tok/s, 3m 51s, 6,209 tokens, stronger game logic > qwen 3.6 27b: 32 tok/s, 18m 04s, 33,946 tokens, better visuals Gemma 4 with lower tok/s finished 14 minutes faster because it used 5.5x fewer tokens to reach a complete answer

@AlexFromAtomic @kimmonismus @atomic_chat_hq Now you're talkin! Also, here are all the models/variants I've ran bifkn.vercel.app/lab/octopus-ar…

@therealbifkn @kimmonismus @atomic_chat_hq I like Qwen version ;) Let's host it and make our own B2B AI SaaS with 10k MRR))

/1 Gemma 4 31B just crushed Qwen 3.6 27B in a local LLM gamedev contest inside @atomic_chat_hq (prompt is below) Device: MacBook Pro M5 Max, 64GB RAM Results: Qwen 3.6 27B: 32 tokens/sec · 18m 04s · 33,946 tokens Gemma 4 31B: 27 tokens/sec · 3m 51s · 6,209 tokens So what is more important: tokens per second, or the quality of the final answer? Qwen made a very long response and showed more creativity and visual style. But Gemma gave a shorter, clearer, and more logical answer in much less time. In this one-shot Pac-Man gamedev contest, Gemma 4 31B was the clear winner. Its game logic was stronger: click reactions were smoother, and it handled interactions with elements like walls, ghosts, and particle effects better. But this was only one test. Maybe Qwen 3.6 27B can show better results with better settings. Open the comments, try our prompt, and share your result below.

Yes - the site is not great - but minibif is working on it 😂 I made a thing. bifkn.vercel.app/lab/octopus-ar…

@tszzl You can just change the power settings?

people are walking around with their laptops slightly ajar to keep their agents running

@outsource_ I have begun the journey!

Guys trust me build in public

@LottoLabs Will this work for Spark and Macs too?

Bros localmaxxing is going to have Rentals It’s a area where you can expose your local rig to the site and people can test your system. You get paid out weekly by users, set your price in dollars/thousand tokens. If your rig is idle you may as well let a moot test drive it

Ready for a model that pushes boundaries? Meet Qwen3.6-27B-AEON-Ultimate-Uncensored-BF16-mlx-2Bit, a text generation beast that processes both images and text. It's abliterated and uncensored, giving you raw, unfiltered AI power.

@witcheer No time for security

when running local AI in your machine, what security harnesses are you using?

@leftcurvedev_ @OpenRouter Can't deny how insanely good 27B is, but also props to 397B, very pretty! We need 3.6 397B and 122B!!!

🥊 Time for a new fight Qwen3.5 397B A17B vs Qwen3.6 27B 🌸 "Cherry Blossom" (↓ prompt below) Using @OpenRouter for 397B Running 27B locally on a single RTX 5080 Wow! 🤯 I'm sure this one is going to divide opinions. 397B feels cinematic with the flare and shadows it added, while 27B is just beautiful; it has a completely different color palette and the leaves are falling from the branches as asked. It's really tough to decide which model is the best. I was about to call it a tie, but then I remembered that 397B needs 180GB VRAM to run locally 😅 Don't get me wrong the model is amazing, but right now the VRAM it requires doesn't feel justified. It's simply not 10× better than 27B. We'll have to dig deeper with more prompts to be sure, let's see what each one has in store for us. What do you think?

@loktar00 From wife's point of view: "Please don't spend the rest of the night working on small demos with Qwen 3.6.... Please don't spend the rest of the night working on small demos with Qwen 3.6.... Please don't spend the rest of the night working on small demos with Qwen 3.6...."

I will not spend the rest of the night working on small demos with Qwen 3.6.... I will not spend the rest of the night working on small demos with Qwen 3.6... I will not spend the rest of the night working on small demos with Qwen 3.6...

@LuizaJarovsky Prove to me you are conscious

Unpopular opinion: AI is NOT conscious.

@astnkennedy Cognitive overload

I'm 22 years old and Claude Code is deteriorating my brain. Every single day for the last 6 months I've had 6 to 8 Claude Code terminals open, waiting for a response just so I can hit 'enter' 75% of the time. And it's doing something to me. In convos with a couple of friends, it's been a point that's been brought up pretty frequently. None of us feel as sharp as we used to. I don't know if it's just us, or others in their 20s are feeling the same thing, but it's something I've been thinking about a lot. P.S. I know this is a problem with my reliability/usage of it, not Claude Code itself, but the effects are real nonetheless

@TwisstedToast @loktar00 It's just a really really dirty goblin

@loktar00 Why does gpt 5.5 make you a black sphere

gpt 5.5 won this, but 27b wasn't far behind. Opus 4.7 not sure what happened there. Autonomous Infinite miner that gains upgrades as it collects gold and crystals.

@loktar00 Nice comparison!

@AlexFromAtomic @kimmonismus @atomic_chat_hq Qwen 2.7B vs Gemma 4 31B: Both are UD-Q8_K_XL

@therealbifkn @kimmonismus @atomic_chat_hq great idea! could you share the results here??

Fair enough! I'm in the same boat as you with the wife and kids, my wife literally just walked into my office and said, "I'm not digging the creepy eyes", "It actually looks really real, but it's weird". My daughter actually enjoys coming into my office to talk to me while I'm working with them on haha. I gotta say that after using them for a bit now, I do not think Apple should abandon this tech at all. It's very very good and there are tons of use-cases for them. I don't see us getting this experience via 'Smart Glasses' anytime soon. It's really next level imo.

@therealbifkn @ADTCoach I know sorry lol I was messing with you there. I guess I should probably stop now. I can understand how some people get value out of it. I just think that from Apple’s perspective, it makes sense to stop making new versions of Vision Pro. M5 should be the last model, imo.

Honestly having a hard time understanding what there is to do on Vision Pro every single day. You couldn’t pay me to use Vision Pro every day. Can someone please educate me? I’m being completely serious. No sarcasm at all. I just don’t understand it.

@AlexFromAtomic @kimmonismus @atomic_chat_hq Absolutely!