cybe

[cybe @cybe42]

@deteccphilippe "Look at us guys, we banned a rage cheater AFTER he posted video evidence with straight up logs so we can easily to locate him to ban him, our kernel spyware anticheat is clearly the best !!!"

[cybe @cybe42]

@womenincages some balls to tweet this on his platform

[cybe @cybe42]

@LLMJunky @aiwithak That is literally all I use LLM's for

[cybe @cybe42]

@aiwithak @LLMJunky they been answering yes for a long time unless you using cheap models

[cybe @cybe42]

@pigeon__s Qwen 3.8 27b isn't benchmaxxed, but you pay that intelligence tax through the form of the model reasoning for like 50k+ tokens at one turns

[cybe @cybe42]

@bnjmn_marie Yes the imatrix quants

[cybe @cybe42]

@himanshustwts this reminds me of the roblox safety dataset

[cybe @cybe42]

@ElPapiDon1 @JohnLar93220143 Kuduz bir köpek bir çocuğu ısırırsa, köpek kuduzu isteyerek kapmamış olsa bile yine de öldürülür.

[cybe @cybe42]

@Shoier__ @teortaxesTex Astra is autistically awesome at spatial work and gets clapped by fable anywhere else. Also generalist models are usually better than specialist

[cybe @cybe42]

@ItsmeAjayKV nah on god you doing something wrong, 27 tps is not normal, I could gamble random llama.cpp settings and still get higher tps

[cybe @cybe42]

@maria_rcks least obvious cursor advertisement

[cybe @cybe42]

@Yuchenj_UW very cherry picked comparison but the general trend due to competition is indeed true

[cybe @cybe42]

@CAronitpereira This quote might've genuinely recovered his aura from the fentanyl drone

[cybe @cybe42]

@RhysSullivan You'll see this as a problem until you genuinely have to argue with an LLM to change its mind. both extreme ends of autonomy in LLM's are horrible, ideally a balanced autonomy in the model is always the best.

[cybe @cybe42]

@max_paperclips @tokenbender ...say that again

[cybe @cybe42]

@bomberfish77 when you go there does real life suddenly turn on vsync

[cybe @cybe42]

@St1tchiyols You can play faster than this in a VR shooter

[cybe @cybe42]

@intocryptoast The point is for there to be a point of comparison, how much inferior it is to the actual original IS the benchmark.

[cybe @cybe42]

AGI will arrive when AI stops saying "us" while talking about humans

[cybe @cybe42]

@pcstyle53 3. -> start believing Gemini benchmaxxes