SteppenAxolotl

DeepSWE benchmarks indicate that DeepSeek v4 Pro only passes 8% of tasks

Posted by Federal_Spend2412@reddit | LocalLLaMA | View on Reddit | 52 comments

[-]

SteppenAxolotl@reddit

I don't like the trends. I think it will take a lot more than 96GB VRAM to get LocalLLaMA to the leading edge of perf. Prices will get a lot worse before it gets better so local might not win in the near term. The cost of multi TB VRAM to do local is not something the vast majority of people could manage.

Union LTL Paystub since everyone is doing paystubs

Posted by Larrythethird22@reddit | Truckers | View on Reddit | 77 comments

[-]

SteppenAxolotl@reddit

I think every single worker should do just that, since the next 2–3 years are fraught with employment risks for everyone (for reasons usually forbidden from articulation on this sub). Options are more limited when you have a family. The median annual personal income in the US is ~$51k 🫡

Union LTL Paystub since everyone is doing paystubs

Posted by Larrythethird22@reddit | Truckers | View on Reddit | 77 comments

[-]

SteppenAxolotl@reddit

Your weekly 401k amt could be slightly 🤏 higher @ $471.

Apocalypse Bunker Fails as Wealthy Residents Turn on Each Other

Posted by delusionalbillsfan@reddit | collapse | View on Reddit | 282 comments

[-]

SteppenAxolotl@reddit

I doubt a billionaire is buying a bunker for $55k + rent and is moving the fam in in 2026. It's a cement box with some dirt on top on a time share plot.

Apocalypse Bunker Fails as Wealthy Residents Turn on Each Other

Posted by delusionalbillsfan@reddit | collapse | View on Reddit | 282 comments

[-]

SteppenAxolotl@reddit

Who will empty the over filled septic systems after society falls. Will the service contracts still be good.

Gen Z is living the modern version of Geertz’s “Agricultural Involution” – we’re all just rice farmers now and it’s actually over

Posted by BearNecessitee@reddit | collapse | View on Reddit | 91 comments

[-]

SteppenAxolotl@reddit

>So it looks like the majority has consistently voted against their own interests for half a century now What makes you assume money is their primary interest?

Americans who didn’t grow up around immigrant communities genuinely cannot conceptualize how much America’s soft power has collapsed

Posted by pichipichipoco@reddit | collapse | View on Reddit | 356 comments

[-]

SteppenAxolotl@reddit

It's the biggest market. You cant avoid the hellscape if you want to be super rich.

Gas station footage. Who’s at fault?

Posted by Pristine_Local_1965@reddit | Truckers | View on Reddit | 125 comments

[-]

SteppenAxolotl@reddit

If the speed of the truck as it comes into frame reflects his several seconds attempt to slow, I would hate to see his speed before that.

My paycheck last friday...

Posted by Imaginativested@reddit | Truckers | View on Reddit | 4 comments

[-]

SteppenAxolotl@reddit

There are people in the world with bigger [paychecks](https://df2sm3urulav.cloudfront.net/tenants/ca/uploads/content/5ab4831bee80e.jpg) than most.

Quantizing MTP KV Cache = free lunch?

Posted by legit_split_@reddit | LocalLLaMA | View on Reddit | 45 comments

[-]

SteppenAxolotl@reddit

>free lunch? There is no such thing. Quantization below 16 is a non-trivial trade off.

wE hAvE a TrUCk DriVEr ShORtAgE

Posted by PhoenixSmasher@reddit | Truckers | View on Reddit | 213 comments

[-]

SteppenAxolotl@reddit

Tears while driving is a dangerous distraction. >[white-collar work fallback](https://fortune.com/article/why-microsoft-ai-chief-mustafa-suleyman-predicts-ai-automation-18-months/?utm_source=native_share&utm_medium=mobile&utm_campaign=social_share)

Got fired this morning

Posted by pumpfakers@reddit | Truckers | View on Reddit | 339 comments

[-]

SteppenAxolotl@reddit

AI is progressing to the point i expect that it isnt long before insurance mandates AI drivers for almost all driving gigs.

OTR guys, just curious what was your longest on the road away from home?

Posted by fistfullofpubes@reddit | Truckers | View on Reddit | 168 comments

[-]

SteppenAxolotl@reddit

VOO is your friend.

A hantavirus cruise ship just scattered 147 passengers to 23 countries with inconsistent quarantine. The CDC and Nebraska already disagree on whether to isolate them.

Posted by Mother-Grapefruit-45@reddit | collapse | View on Reddit | 448 comments

[-]

SteppenAxolotl@reddit

>collapse is already happening Most everything is fine. We're at the very beginning of a possible passage into a more utopian society.

A hantavirus cruise ship just scattered 147 passengers to 23 countries with inconsistent quarantine. The CDC and Nebraska already disagree on whether to isolate them.

Posted by Mother-Grapefruit-45@reddit | collapse | View on Reddit | 448 comments

[-]

SteppenAxolotl@reddit

>stop overreacting to things that wont pan out Tell that to the ~7M covid dead. >collapse is very real stop overreacting to things that wont pan out

A hantavirus cruise ship just scattered 147 passengers to 23 countries with inconsistent quarantine. The CDC and Nebraska already disagree on whether to isolate them.

Posted by Mother-Grapefruit-45@reddit | collapse | View on Reddit | 448 comments

[-]

SteppenAxolotl@reddit

it's always better to pretend it isn't happening.

First time GPU buyer. Got a RTX 5000 Pro. Was it a bad decision compared to two 3090s?

Posted by Valuable-Run2129@reddit | LocalLLaMA | View on Reddit | 106 comments

[-]

SteppenAxolotl@reddit

FP8 & NVFP4

Driverless Trucks

Posted by gulagsurvivor2019@reddit | Truckers | View on Reddit | 256 comments

[-]

SteppenAxolotl@reddit

I thought posts about Driverless Trucks was verboten on this sub.

It looks like China gets it.

Posted by FlapXenoJackson@reddit | Truckers | View on Reddit | 124 comments

[-]

SteppenAxolotl@reddit

That law doesn't stop a new competitor startup to employ AI/robots and under cut the existing business with expensive human workers. Such laws cant stop the tide.

What are the best parts of being a trucker?

Posted by SoundKidTown1085@reddit | Truckers | View on Reddit | 66 comments

[-]

SteppenAxolotl@reddit

reptiles are cold blooded

Kv cache quantization: ignorance, or malice?

Posted by wombweed@reddit | LocalLLaMA | View on Reddit | 94 comments

[-]

SteppenAxolotl@reddit

You have 24GB x 2 vram to play with. Your trade off mix will be very different than someone with 24GB or less.

Kv cache quantization: ignorance, or malice?

Posted by wombweed@reddit | LocalLLaMA | View on Reddit | 94 comments

[-]

SteppenAxolotl@reddit

it's a trade off, some set their bar higher than others

I feel like everything is falling apart and it started after COVID (my story)

Posted by JesseVenturaInsights@reddit | collapse | View on Reddit | 36 comments

[-]

SteppenAxolotl@reddit

>We started wearing those masks for 8-10h a day, talked to people behind plexiglass That job did try a little to help everyone to not get sick.

Finally found one of those good paying local jobs

Posted by AffectionateEase1606@reddit | Truckers | View on Reddit | 78 comments

[-]

SteppenAxolotl@reddit

Keep in mind that the next 10 years will look nothing like the last 50 due to that which cannot be named.

Unpopular opinion: OpenClaw and all its clones are almost useless tools for those who know what they're doing. It's kind of impressive for someone who has never used a CLI, Claude Code, Codex, etc. Nor used any workflow tool like 8n8 or make.

Posted by pacmanpill@reddit | LocalLLaMA | View on Reddit | 263 comments

[-]

SteppenAxolotl@reddit

Your mileage will vary depending on the tool and the AI backend. Ultimately, even the best tool connected to the best AI still sucks and only foreshadows what it could be like when AI reaches a reasonable level of reliability and competence.

ZAI might stop open-weighting their models?

Posted by TheRealMasonMac@reddit | LocalLLaMA | View on Reddit | 60 comments

[-]

SteppenAxolotl@reddit

>prioritizing profit without regard to their customers Did all the inference providers that sell API tokens to their open models give them a cut? It was free inputs to their business and it never crossed their minds what happens if the source of their free inputs went away. Traditional definition of a customer is: One that buys goods or services, as from a store or business.

"Everyone is Replaceable" - A worker died at an Amazon warehouse in Oregon last week. Employees were told to look away.

Posted by Bolinas99@reddit | collapse | View on Reddit | 158 comments

[-]

SteppenAxolotl@reddit

What is the right thing to do, the entire world should stop and for how long, every time someone drops dead?

GLM 5.1 tops the code arena rankings for open models

Posted by Auralore@reddit | LocalLLaMA | View on Reddit | 146 comments

[-]

SteppenAxolotl@reddit

LocalLLaMA could run out of steam if sizes continue to increase on the leading edge. >GLM 5.1 UD-Q4_K_XL 466GB What are the chances some near AGI can be condense down to < 100GB vram.

$16 bucks an hour to haul doubles in 2026

Posted by West_Violinist_6809@reddit | Truckers | View on Reddit | 44 comments

[-]

SteppenAxolotl@reddit

My local supermarket has signs up for associates at $16/hr.

$16 bucks an hour to haul doubles in 2026

Posted by West_Violinist_6809@reddit | Truckers | View on Reddit | 44 comments

[-]

SteppenAxolotl@reddit

My local supermarket is looking for associates for $16/hr.

pushback on 'permanent underclass' fear-mongering

Posted by cobalt1137@reddit | LocalLLaMA | View on Reddit | 58 comments

[-]

SteppenAxolotl@reddit

>[The Realpolitik of the Permanent Underclass](https://www.reddit.com/r/singularity/comments/1sccard/the_realpolitik_of_the_permanent_underclass/) The joke will be on those that thought they escaped the permanent underclass.

I'm shocked (Gemma 4 results)

Posted by Potential-Gold5298@reddit | LocalLLaMA | View on Reddit | 78 comments

[-]

SteppenAxolotl@reddit

>Small-scale manual performance comparison benchmark I made for myself. This table showcases the results I recorded of various AI models across different personal tasks I encountered over time (currently 83). I use a weighted rating system and calculate the difficulty for each tasks by incorporating the results of all models. This is particularly relevant in scoring when failing easy questions or passing hard ones. >NOTE, THAT THIS JUST ME SHARING THE RESULTS FROM MY OWN SMALL-SCALE PERSONAL TESTING. YMMV! OBVIOUSLY THE SCORES ARE JUST THAT AND MIGHT NOT REFLECT YOUR OWN PERSONAL EXPERIENCES OR OTHER WELL-KNOWN BENCHMARKS. Use which ever model is best for your personal workload.

If you ever feel stupid, remember that...

Posted by RRK96@reddit | collapse | View on Reddit | 287 comments

[-]

SteppenAxolotl@reddit

These people all died of natural causes. They have a small but non-zero chance of living in the future. >are just preying upon people It's just playing the long odds. Sometimes you win. In theory, it will take nano tech to fix them. Worse case, copy the structure of their brains and they get "uploaded".

Slower Means Faster: Why I Switched from Qwen3 Coder Next to Qwen3.5 122B

Posted by Fast_Thing_7949@reddit | LocalLLaMA | View on Reddit | 84 comments

[-]

SteppenAxolotl@reddit

Most of my code is written under Q8 to benefit context window, and I revert to fp16 to solve anything Q8 cant deal with. I use only use OpenRouter for the big guns when they fail and for some design work.

Slower Means Faster: Why I Switched from Qwen3 Coder Next to Qwen3.5 122B

Posted by Fast_Thing_7949@reddit | LocalLLaMA | View on Reddit | 84 comments

[-]

SteppenAxolotl@reddit

Maybe it's because I use TLA+ specifications for my reverse engineering dev. You have something that definitely works, create a formal spec and develop against that until it passes.

Slower Means Faster: Why I Switched from Qwen3 Coder Next to Qwen3.5 122B

Posted by Fast_Thing_7949@reddit | LocalLLaMA | View on Reddit | 84 comments

[-]

SteppenAxolotl@reddit

Sometimes I run Qwen3.5-27B with q8_0 cache to get more context, it runs into blocking issues. I retry under full precision and it gets past the same problem. --cache-type-k q8_0 --cache-type-v q8_0 This is with coding work reverse engineering android apps and port to desktop apps.

17,000 Drivers

Posted by ConsciousAwareness69@reddit | Truckers | View on Reddit | 277 comments

[-]

SteppenAxolotl@reddit

Worth remembering the 6th generation Waymo Driver is very proficient with most languages.

Slower Means Faster: Why I Switched from Qwen3 Coder Next to Qwen3.5 122B

Posted by Fast_Thing_7949@reddit | LocalLLaMA | View on Reddit | 84 comments

[-]

SteppenAxolotl@reddit

> run the biggest model you can fit people should run the most intelligent model they can fit

Glm 5.1 is out

Posted by Namra_7@reddit | LocalLLaMA | View on Reddit | 218 comments

[-]

SteppenAxolotl@reddit

if you pay $84 for 3× usage of the Claude Pro plan, you will be able afford GLM5 for 1,619 years for the price of 16 RTX 6000 pros.

Implementing TurboQuant to MLX Studio

Posted by HealthyCommunicat@reddit | LocalLLaMA | View on Reddit | 15 comments

[-]

SteppenAxolotl@reddit

[turbo in llama.cpp](https://github.com/TheTom/turboquant_plus)

Throwback to when I had to drive 2 weeks in a shitty Hino

Posted by Tsars_Ball_Scrubber@reddit | Truckers | View on Reddit | 14 comments

[-]

SteppenAxolotl@reddit

>I left my phone on the tire I would have enough for a nice steak dinner if I had $1 for ever $700 handheld drivers ran over doing that back in the day.

Guys, please don't do this

Posted by DaSaw@reddit | Truckers | View on Reddit | 55 comments

[-]

SteppenAxolotl@reddit

Yea. But.. Your DNA will be just another data point on your personalized online advertising profile. Blackstone completed their acquisition of Ancestry like 5 years ago. AI systems are real good at pattern matching. It's nothing personal, only business.

Guys, please don't do this

Posted by DaSaw@reddit | Truckers | View on Reddit | 55 comments

[-]

SteppenAxolotl@reddit

DNA profiling should be super cheap next year. Lots of help out there if the state will allow revenue sharing with any fines.

do truckers actually ask for “ass, gas or grass?”

Posted by causmicx@reddit | Truckers | View on Reddit | 64 comments

[-]