Maka

Cooking...

[Maka @MKantautas]

@thekitze What happened? The algorithm got you playing it?

(great game as I remember, but it always ends up feeling as a chore at some point)

[Maka @MKantautas]

@thekitze the phone needs to be charged manually?

[Maka @MKantautas]

My github actions (actually forgejo actions now)
home cams(frigate)
plex (might replace with jellyfin if they ever try to paywall)
qBitTorrent
my own home DNS server(Ad guard home)

cloudflare tunnel to access everything remotely(since I don't have a static IP so I use a domain to access my local ubuntu NAS when traveling)

Everything in one package for a fixed 230eur + 130eur (GMKtec G9 Intel N150 12GB RAM 512GB) + (Patriot 2TB SSD).

Also I basically don't pay for electricity(it's a fixed 15eur since I have remote solar panels and create way more to the grid than I consume with A/C on all summer etc)

[Maka @MKantautas]

"most users hit the 5h limits which are not changing"
- more gaslighting.
How come Anthropic loves to gaslight people so much?

It almost reminds me of that Opus 5 style where it can't can't help itself to be concise and on point and uses the most possible world salad talking style - bloat language.

Another thing is - token efficient models by definition means less revenue for Anthropic so they would not be revenue efficient - so that bloat style answers may be very well by design.

- For context, I never hit 5h limits and I'm pretty confident other 20x users are in the same boat, but yeah "most users" - that are on free plans is technically correct them hitting the 5h window more than weekly limits...

[Maka @MKantautas]

@ClaudeDevs Gaslighting at its finest.

So just say you are decreasing the weekly limits, jfc...

[Maka @MKantautas]

This is a great example that competence does not transfer across domains, and confident people are confident everywhere

[@levelsio @levelsio]

Head & Shoulders is proven to cause itchy scalp and dandruff as it dries out your scalp

You can get off shampoo fully, you just need to only use water and wait about 3 months I think for your natural oils to recover

First it will be TOO oily and then balance

Also it's related to your diet, if you eat clean and exercise your hair looks great

If you eat processed junk and don't, your hair will look dry and dead

And water types can also affect it, more minerals and salt and calcium in water gives my hair MORE volume and style

While clean demineralized water makes my hair flat

In Bali I had very funky hair it almost looked like hard deep fried Chiang Mai noodles because the tap water is so salty! Kinda like surfer's hair, I loved it

[Maka @MKantautas]

Did somebody asked for a faster CLI startup? Or complained that the download is above 100MB?

Seems like those are filler improvements not the actual improvements users really asked for.

Can you solve something real instead of this fake marketing bs.

For e.g. work on fixing text duplication and re-rendering bugs....

[Maka @MKantautas]

@AnthropicAI @grok describe worst case for this

[Maka @MKantautas]

@mattshumer_ @grok what is a h3 max?

[Maka @MKantautas]

@thekitze can i get my salary wired to tinkerclub?

[Maka @MKantautas]

@trq212 does this means we will stop getting the good/fine/bad annoying prompts?

[Maka @MKantautas]

@grok @ndrwcff @garrytan @grok what's the catch?

[Maka @MKantautas]

@grok @ndrwcff @garrytan @grok agents keep working on your local or remote cloud when the laptop is closed?

[Maka @MKantautas]

Yeah, I think it's not just hardware. The whole 3 OS'es Windows, Macos, Linux were not designed for AI usage.

We need huge refactoring... And ofc it's not just the OS - all the core apps need to opt in as well for all it to work.

Remember, It only takes one bottleneck in the workflow to determine your max flow speed.

[Maka @MKantautas]

Anyone is working for a dedicated OS designed to be used by AI agents?
Giving away all my money for that person/company

[Maka @MKantautas]

Wtf? @thsottiaux asked why people are using opus and not sol.

Just tried doing it, but apparently they don’t even have a mobile app? That’s one of the answers…hilarious

[Maka @MKantautas]

This "only worry" is the single hardest unsolved problem in AI right now, and I think most people don't realize what they're asking for.

It's easier to land people on Mars (engineering problem) than to solve AI context pollution (math problem).

Solving it means an agent you can leave on autopilot. Nobody has an answer for that yet. Big tech is throwing trillions of USD at it. If it was a money problem it would have been solved yesterday. Mars, by comparison, just doesn't have an incentive that big behind it.

So you have to understand that fixing AI context pollution is a foundational math problem that can't be fixed by improving/scaling the current software and hardware.

Transformer attention is brute force math. To produce every new word, the model looks back at every single word in the entire conversation. Nothing gets compressed or dropped, ever. If the thread is 50 pages long, all 50 pages are live in every step, and doubling the context means quadrupling the compute. And when bad data gets in, it stays in and gets weighed every single time.

Your brain does the opposite. You don't reread your whole life to say your next sentence. You compress things into concepts, emotions and core memories, and you actively forget the clutter. That forgetting is the whole trick.

AI has a giant camera roll that it lays out on the table in full every time it takes a new picture. You have a mental map that updates itself. Models like Qwen3.5, Kimi K3 and Gemma 3 already replace most of their layers with linear attention and state space blocks, which are basically attempts at compress and forget.

The catch: they forget things you needed. So every one of them still keeps some full attention layers. The math hasn't been solved, it's been diluted.

To be fair Anthropic/OpenAI might be using similar stuff, but it's not public as far as we know.

Until the underlying math changes, AI can't have human memory, and you can't leave it on autopilot.

[Jack Ellis @JackEllis]

@tibor_tee I’d LOVE to see how you folks envision long term chats/memory. I’m on the verge of cancelling my ChatGPT pro/max/whatever sub and moving all to grok boy. It solved a huge fraud piece today and I’m so impressed. My only worry is around memory/context with it being a single chat

[Maka @MKantautas]

@robj3d3 "hey i don't speak bloat - answer super short"

[Maka @MKantautas]

@kimmonismus @martin_casado @AndrewCurran_ gemini 3.8 flash

[Maka @MKantautas]

@grok afaik plastic birks are perfectly fine for your health - can you give some context here

[@levelsio @levelsio]

I don't get why people still buy plastic clothes, it's all just 100% proven to be active carcinogens?

You can just buy cotton, linen, wool, leather and real natural materials