Vojta Humpl

@bridgebench "You're running 5, 10, 15 agents at once." nobody serious does that

GLM 5.1 is the slowest frontier model we've ever benchmarked on BridgeBench. 44.3 tokens per second. Half the speed of GPT 5.4. Nearly 6x slower than Grok 4.20. Z.ai traded all of their speed for intelligence. The coding benchmarks improved. The throughput collapsed. In 2026, agentic coding is about parallelism. You're running 5, 10, 15 agents at once. A model this slow bottlenecks every workflow it touches. Intelligence without speed is a luxury most vibe coders can't afford. bridgebench.ai

@theo How are you still not able to realize, that they subsidize a very small part of all the subs? As most people don't hit the limits ever. I have the pro sub and use codex daily a never got below 85% weekly.

AI has a subsidization problem, and it's getting bad.

@VeselyJir Třeba napsat, které modely to byly? Jestli porovnáváte Claude Opus 4.6 s ChatGPT 5.3 instant, tak je to jasné, že. Celkově GPT 5.4 thinking je v kódu značně lepší než Opus 4.6

Udělej mi skript, který bude tahat tyhle informace z logů. Claude: jasně, tady je. Máš to včetně dekódování stavů a ošetření nedostupnosti. Ještě bych doporučil zapracovat tyhle parametry, mám to udělat? ChatGPT: tady to je -ale mě to hází při spuštění tuhle chybu - jejda, tady je opravená verze - ta chyba tam je oirad - oops, zde je oprava - chyba trvá. Nemůže to být tahle čas kódu? - máš pravdu, na tom místě byla chyba. Tady je oprava 😡 😡 😡 😡

@michalhardyn Raspberry Pi 5 přesně tohle utáhne absolutně bez problému za 2.5k

Víkendový projekt hezky pokročil. Ten Mac Mini dorazil. Pár lidí tipovalo správně. Část se ptala, jestli na tom poběží lokální AI modely. Nepoběží – 16 GB na velké modely nestačí, to je pravda. O to ani nešlo. Těžký výpočet stejně dělá cloud. Mac Mini je spíš takový dispečer – řekneš mu co chceš a on to zařídí. → n8n, AI agenti, automatizace. Běží pořád, nemusím na to sahat. → Z mobilu se připojím odkudkoliv. Sedím v autě a řeknu mu co má dělat. Fascinující. Celé to v idle žere 3 watty. Stejně jako Raspberry Pi. Jenže RPi by se s tímhle asi udusilo. Za 16 tisíc mám krabičku, která jede nonstop. Tahá data, spouští workflow, neustále k dispozici a já řeším jiný věci. A mě to nestojí ani korunu měsíčně za fíčko. Kdo z vás zkouší něco podobného? Na čem to běží a co to dělá?

#G2WIN @G2LEAGUE CLEAN SWEEP @GENG!

@Cartidise because i switched off the chinese shit years ago? i had to use the Vivo X Fold5 for a few weeks and it's still just as trash as it was back then

@Vojta_Humpl OnePlus 8 was released 6 years ago lmao shut up

I’m sobbing OPPO unboxed a brand new Samsung Galaxy Fold7 on stage for this comparison 😭

@Cartidise I used oneplus 8, but you seem to be braindead anyway to not understand that software doesn't get better with hardware when it's buggy

@Vojta_Humpl so you used a $100 phone?

@Cartidise it doesn't matter if it's a flagship or not when they use the same version of the UI, the problem isn't lags but bugs

@Vojta_Humpl you clearly haven’t used a Chinese flagship

@1v9GG_ DLSS 5

If League of Legends 2 looked like this… would you play it? 🤔

@AndyH_England @lackynamber1234 @TailosiveTech well Windows doesn't cache it in the page file unless it's hitting swap, but anyway I still don't think the higher RAM efficiency of macOS can supplement missing 8 GB of RAM, like when editing videos you have to load into RAM and there is no way around it

@Vojta_Humpl @lackynamber1234 @TailosiveTech No, I was pointing out in the reply why the monitor shows high RAM: it is because Apple caches at launch based on the apps you may use. I know Windows does the same with its page file.

Apple just convinced an entire community of content creators that 8GB of memory is acceptable for the right price. After a year of everyone praising Apple for making 16GB the default option.

@AndyH_England @lackynamber1234 @TailosiveTech every OS does this, you act like this is an Apple only feature, I have 14.5 GB cached out of my 32 GB RAM on windows

@lackynamber1234 @TailosiveTech Apple caches apps that are not in use to improve load times. So, the cache is not used. You will find that Macs boot with about 2GB used. Unfortunately, many people do not understand basic RAM concepts.

@yeahtellem @VirtuaStick @hardwarecanucks the whole version 26? it's so bad on my mac mini

@VirtuaStick @hardwarecanucks insane cope and i’m telling you that as a windows user name 1 mac os update that was as bad as infamous windows update that made SSD’s fail worldwide or this new update that messes with your /:C disk

Windows Apologists need an attitude readjustment. The problem isn't specs. It's the OS What's the use of better hardware if the software is a bloated, unoptimized mess, shitting the bed in a new way after every update? That's why the Neo "feels" so fast: extreme optimization.

@brandenflasch this is better in every metric, how could it be worse

For the average user, having a new Neo would be better

@GamersNexus @PatrickMoorhead Nobody ever built anything, because they used tools somebody else made, the guy who first picked up a stone to use as a hammer is the inventor of everything, yes

@PatrickMoorhead If you used tokens and AI, you did not, in fact, "personally build" it. Millions of people whose work was stolen built it. You benefited from it.

I had a good friend tell me the apps I created were trash because I didn't have a formal database. I took that as a challenge. In 5 hours, I personally built, populated, and deployed a database containing all the content the company and analysts created over the past 15 years, press quotes, and social media posts, with integrated semantic and basic search and auto-updates. And is scalable to 10x the data. Tested it with 1,000 different queries that users would make. It works. :-) Now it's time to connect the apps I created with embedded data to the new database. It cost me around $200 in tokens. Ironically, the most expensive part was scrubbing ten years of press quotes. Still a better deal than Meltwater or Muckrack. The next step is to inject proprietary and protected data into it, with proper safeguards and attribution.

@literallydenis whats the context size 1024? 2048? not usable at all

Just tested: llama.cpp on $500 MacBook Neo: Prompt: 7.8 t/s / Generation: 3.9 t/s on Qwen3.5 9B Q3_K_M So, smaller models are even more usable

@guurennne_1 @ChatGPTapp and that's a good thing

140 million people use ChatGPT to help them understand math and science concepts every week.

@TheGalox_ Tt's not the chips, it's the OS made for only a few chips, Exynos windows would be pointless, unless they would get Microsoft to callaborate. A Tensor Chromebook could be good, but Tensor is quite weak.

Samsung and Google have the same huge advantage as Apple, they have their own processors Exynos powered Windows laptop and Tensor powered Chromebook would be the only products that could match MacBook Neo in quality and price while delivering a good user experience

@yacineMTB They can't even maintain two dark modes, let alone the android app

Fix the android app

@iKoudy @psafarcik_ @Slavia_poj Často třeba výpisy z banky jsou zaheslovány telefonním číslem nebo poslední 4 čísla z rodného čísla, což jde cracknout hned. Zajímavé, že tyhle hesla mají na těch dokumentech i interně

@psafarcik_ @Slavia_poj Absolutni peak vseho je tohle „Na průměrné grafické kartě to zabere méně než pět sekund”. Být jejich klientem, tak jsem hodně kyselej 😬

Podle zveřejněných informací mělo dojít k exfiltraci ~150 GB dat ze systémů Slavia pojišťovny (@Slavia_poj). Útočník tvrdí, že kompromitace proběhla přes zranitelnost v Admineru, která umožnila nasazení webshellu a následné udržení reverse shellu pro další pohyb v síti a exfiltraci dat. Údajně kompromitovaná data zahrnují: • interní dokumenty a pojistné smlouvy • záznamy o škodních událostech • komunikaci mezi klienty a pojišťovnou • zdravotní dokumentaci (včetně ultrazvukových snímků) • databázi klientů (jména, e-maily) • soubory chráněné heslem (údajně heslo = datum narození pojištěného) Útočník také tvrdí, že si stále udržuje přístup do infrastruktury.