Sol4ra

@mattcreed23 @WongUpdates Literally the most dogshit ending, the fuck are you talking about. Literally anything would be better.

@WongUpdates Would have been crazy but def not a better ending then him getting what he deserves

Antony Starr reveals that they shot an alternate ending for ‘THE BOYS’ where Homelander wins. “He fucking kills everyone and scorches the entire planet. Oh well, it will never see the light of day.” (via: variety.com/2026/tv/news/t…)

@NancyJo1961 @depressionlesss Why are you so retarded

@depressionlesss Please, please stop anthropomorphizing animals!

Cat immediately apologises to dog after realising that she was just playing with her baby, not harming him

@winterlilies333 @Awk20000

@Awk20000 I always found him so insufferable and very quick to shame women, people of color, and the poor. Won’t be surprised when he comes out as republican or something else shitty.

Caleb Hammer on those paying $20 per lunch to get fast food “Pack a sandwich”

@nicksortor How many fucking fires will this shitty state go through before it actually puts in solid infrastructure to combat it lmao

🚨 BREAKING: EVACUATIONS ORDERS have been issued in Simi Valley, California, in the Greater Los Angeles area as a brush fire threaten entire neighborhoods Structures are reportedly engulfed — possibly homes. Pray for anyone in the path of this fast-moving blaze 🙏🏻 🎥 @DanielJMcGreevy

@VoodooDE_Gaming @edison_goff You didnt answer the question.

@edison_goff Steam Frame is not for Mixed Reality

@Ch1bi_ @AB_type_aruaru1

アメリカの男性カップルがクラファンで"代理出産"の資金調達 ↓ 資金が集まり無事男児の親権をゲット カップルで仲睦まじく男児にキスするほのぼの動画をup ↓ 動画右側の男性が過去に...

@yxngaltpolo @peter_lingua_ Africans =/ African Americans

@peter_lingua_ Bold as hell letting three random niggas in her car

After their village is burned down during the civil war, a group of siblings flee for survival. But tragedy strikes when one sibling is left behind while the rest begin a new life in the United States. 🎬 Title: The Good Lie

@danielhanchen Not paroquant in 2026? 😴

We released experimental MTP Qwen3.6 Unsloth GGUFs! Qwen3.6 27B MTP now runs at 140 tokens/s. Qwen3.6 35B-A3B MTP gets 220 tokens/s generation on a single GPU. Qwen3.6 27B and 35B-A3B have >1.4x speed-up over the original GGUFs without any change in accuracy. Guide + GGUFs + Benchmarks: #mtp-guide" target="_blank" rel="nofollow noopener">unsloth.ai/docs/models/qw… In terms of average speedup, we see a 1.4x for dense models at draft tokens = 2 and for the MoE around 1.15 to 1.2x. We do not recommend more than 2 draft tokens because the acceptance rate drops precipitously from 83% to 50% with 4 draft tokens, and the forward passes for MTP become less beneficial. Use `--spec-type mtp --spec-draft-n-max 2` Thanks to Aman for github.com/ggml-org/llama…!

Young men in my dms please stop flirting with me i am 38 years old

@Montreal Never been prouder to be an American

You're welcome here. 🏳️‍🌈🏳️‍⚧️

@bnjmn_marie Please test paroquant! Its supposed to be better than FP8 while being the same size as Int4! huggingface.co/z-lab/Qwen3.6-…

Evaluations of Quantized Qwen3.6 27B: FP8, NVFP4, and INT4 The model is robust to quantization. But have to be careful with the linear attention, not all submodules can be safely quantized to 4-bit, and NVFP4 underperforms (evaluated autoround and llm compressor). Intel’s INT4 model is particularly strong, and I also found it’s the fastest, and the smallest I’ve evaluated, with vLLM. One issue: these 4-bit models tend to generate more tokens, +10~30% compared with the BF16 model Full analysis: Latency, accuracy, MTP compatibility, memory consumption, here: kaitchup.substack.com/p/qwen36-27b-q…

@fw_keitth @PoroRaal @veizau Black men are actually just as likely to shoot up a school as white men based off the demographics of prior shooters.

@PoroRaal @veizau Just like whites shoot up schools and kill multiple innocent children and ruin multiple innocent families ?

Joe Bartolozzi GOES OFF on people who say Black people are more likely to commit crimes than white people “If you are going to say statistically they do, that is only because of impoverished areas and the lives they’ve been given due to the implications of systemic racism.”

@ArtificialAnlys Need to see where Qwen 27b stands and which harness is best for it. You already know its a beast…

Announcing the Artificial Analysis Coding Agent Index! Our new coding agent benchmarks measure how combinations of agent harnesses and models perform on 3 leading benchmarks, token usage, cost and more When developers use AI to code they’re choosing a model, but also pairing it with a specific harness. It makes sense to benchmark that combination to understand and compare performance. The Artificial Analysis Coding Agent Index includes 3 leading benchmarks that represent a broad spectrum of coding agent use: ➤ SWE-Bench-Pro-Hard-AA, 150 realistic coding tasks that frontier models struggle with, sampled from Scale AI’s SWE-Bench Pro ➤ Terminal-Bench v2, 84 agentic terminal tasks from the Laude Institute and that range from system administration and cryptography to machine learning. 5 tasks were filtered due to environment incompatibility ➤ SWE-Atlas-QnA, 124 technical questions developed by Scale AI about how code behaves, root causes of issues, and more, requiring agents to explore codebases and give text answers Analysis of results: ➤ Opus 4.7 and GPT-5.5 lead the Index: Opus 4.7 in Cursor CLI scores 61, followed closely by GPT-5.5 in Codex and Opus 4.7 in Claude Code at 60. GPT-5.5 in Cursor CLI follows at 58. ➤ Open weights models are competitive, but still trail the leaders: GLM-5.1 in Claude Code is the top open-weight result at 53, followed by Kimi K2.6 and DeepSeek V4 Pro in Claude Code at 50. These are strong results, but still meaningfully behind the top proprietary models. ➤ Gemini 3.1 Pro in Gemini CLI underperforms: Gemini 3.1 Pro in Gemini CLI scores 43, well below where Gemini 3.1 Pro sits on our Intelligence Index, highlighting that Gemini’s performance in Gemini CLI remains a relative weak spot for Google’s offering. ➤ Cost per task (API token pricing) varies >30x: Composer 2 in Cursor CLI is cheapest at $0.07/task, followed by DeepSeek V4 Pro in Claude Code at $0.35/task and Kimi K2.6 in Claude Code at $0.76/task. At the high end, GPT-5.5 in Codex costs $2.21/task, while GLM-5.1 in Claude Code costs $2.26/task. For both models this was contributed to by high token usage, and in GPT-5.5’s case by a relatively higher per token cost. ➤ Token usage varies >3x: GLM-5.1 in Claude Code uses the most tokens at 4.8M/task, followed by Kimi K2.6 at 3.7M/task and DeepSeek V4 Pro at 3.5M/task. GPT-5.5 in Codex uses 2.8M tokens/task, substantially more than Opus 4.7 in Claude Code at 1.7M/task. In GLM-5.1’s case, higher token usage, cost and execution time were partly driven by the model entering loops on some tasks. ➤ Cache hit rates remain high but vary materially: Cache hit rates range from 80% to 96% across combinations. Provider routing, harness prompt structure and cache behavior can materially change the economics of running the same model given cached inputs are typically <50% the API price of regular input tokens. ➤ Time per task varies >7x: Opus 4.7 in Claude Code is fastest at ~6 minutes/task, while Kimi K2.6 in Claude Code is slowest at ~40 minutes/task. This is contributed to by differences in average turns per task, token usage and API serving speed. Opus 4.7 had materially lower amount of turns to complete a task than all other models while Kimi K2.6 had the most. ➤ Cursor made real progress with Composer 2: Composer 2 in Cursor CLI scores 48, near the leading open-weight model results, while being the cheapest combination measured at $0.07/task. Cursor has stated Composer 2 is built from Kimi K2.5, showcasing they have made substantial post-training gains. This is just the start. We are planning to add additional agents (both harnesses and models). Let us know what you would like to see added next.

@rpcs3 Lol would love to see the little bitch behind these shitty posts. You do nothing but deface the real RPCS3 team.

Our guidelines for submitting AI-generated code are now up in our repository! As for all the AI bros seething on our socials, we're simply blocking you. Learn how to debug, code, and leave behind something useful to humanity when you're gone, instead of peddling slop.

@taco2p @animeupdates Youre a retard.

@animeupdates There’s no way this many people think “data center has been burned” means it actually lit on fire.

Notorious Piracy Streaming site AnimeKai will be shutting down, with developer ending the project following the data center catching on fire "Sorry, our data center has been burned :( We're no longer able to provide the file hosting service."

@rpcs3 Lol, I have no issue calling out shitty half assed PRs but to require disclosing the use of AI and telling people to go learn how to debug and code the old fashioned way is dumb asf.

Please stop submitting AI slop code pull requests to RPCS3. We will start banning those who do without disclosing. There are plenty of resources online to learn how to debug and code instead of generating slop that you don't understand and that doesn't work.

@TheoryLemur @RealestMemes_ You know you can delete games off your steam library right?

@RealestMemes_ Sometimes the idea that a game I'll never play again after it gets old is gonna sit on my games list is the main thing that will stop me from buying it in the first place no matter how cheap or many people I know like it.

@bnjmn_marie Please compare this new 4 bit paroquant with the others, apparantly its supposed to be better than FP8, I wouldnt believe it but it comes from Z-Labs! huggingface.co/z-lab/Qwen3.6-…

4-bit Qwen3.6 models, even with some layers kept in 16-bit, are more prone to endless thinking loops. FP8 behaves well. The worst case is the NVFP4 version with quantized linear attention, although the behavior curve is interesting: when it does not loop, responses tend to be shorter.

@sudoingX @GrizzledTexan Sure bud, or its because you dont use this for literally anything worth a damn. All just smoke and mirrors. What useless software.

@GrizzledTexan wouldn't you like to know. some things stay between me and the machine.

i named my dgx spark "spark." it runs hermes agent /goal overnight. brain is qwen 3.6 27B Q8, 262K context, i set a goal before bed and wake up to results. no rate limits. no token costs. just local inference grinding while i sleep. this thing never stops.

@theodormarcu Working on a Claude/ chatgpt ecosystem equivalent but that is model/ API agnostic. This compute would go far as my claude pro and gpt pro subscriptions are both at their limits 😭

Have a few more Max subscriptions to give out to folks if you missed this! Just reply and DM me :)