vvector

@TheAhmadOsman 2x'd gg! Honest question don't you think this kind of post may attract speculators and keep away GPU from humble learners ?

People that followed me last summer scored some sweet deals btw - $2,000 RTX GeForce 5090 - $6,000 RTX PRO 6000

My only regret is actually not telling people to Buy a GPU earlier than last summer Given how many people called me crazy so close to the demand surge, I am not sure how people would've reacted to me saying that in 2024 for example BTW, you should acquire some compute ASAP

@pupposandro Seller told me he won this one, comes with a "known" defect, will be fun to tweak (no issue 200w long prompt)

On a mission to collect every different version of the 3090 ever produced

@OrganicGPT yes don't get them they so bad it's incredible ;)

If you wanna run AI models locally, the best option is an RTX 6000 Pro. DON'T get a 5090/4090. And DON'T listen to people who hype the 3090; those cards are beat at this point. Get the RTX with education discount through Nvidia. These used to be $8000, now they're +$9000.

@justaguy_net I try not to work on phone for mental sanity + i do bad prompt on phone (lazy), still have an orchestrator plugged on tg

@computedvv Yeah, thanks. What SSH phone app you'd recommended?

Guys, how do you access your Hermes agent remotely? I somewhat don't like Slack, never did to be honest. From some reason it feels cluttered to me... I decided to give it a try again whilst using with my Hermes agent instead of Telegram and I still have a same issue with it. Question is if migrating my Hermes agent to Discord will help because I haven't used Discord much either. Then there is another issue and that's a privacy reason. I don't think I'd trust any of these mentioned either.

@oliverhenry I use the hermes and i wish i was a bot

I have still yet to meet someone actually using hermes. Whenever I mention it, i just get bots in my replies. Openclaw 4 lyf

@OrganicGPT @theo Electricity bill

@theo what costs to you have that can't be covered by this much revenue?

Another $6000 payout 👀 X revenue share is officially paying me more than YouTube Adsense

Hope i fixed u and u'll fix my lack of cognitive abilities

@pupposandro What motherboard you guys use for multi-gpu. I heard my doesnt have the right thingies

Didn't see this coming at all. What a crazy market.

Skkrrrt skkkkrrrt, more gpu

I guess i cant do local llm without a proper rack. Brb

@Teknium @unusual_whales I here i so powerfull they hesitated to release it fearing agi

@unusual_whales Or use hermes and self improve better

BREAKING: Anthropic has released a feature called "dreaming" which allows AI agents to self-improve

@pupposandro already bought 2 x 3090. First time doing local inference. Day one learned a loooooot. Need more compute

it's time to get all the compute you need before it's too late

Ok this is fun

@sudoingX 一 Socrate -450 BC

i am retarded in many ways but at least i know i don’t know enough

@vxunderground 5) isnt against rule 34 ?

Here's the thing no one wants to tell you about AI: 1. It's the worlds largest Python script 2. It runs on Linux, Windows Defender slows it down 3. It uses lots of numbers (nobody knows why) 4. It requires a bunch of GPUs (for gaming) 5. No, you can't have sex with it

I think i got nerd sniped, bought GPU's Let's see if im smurt enuf to do something with them @0xSero @sudoingX @TheAhmadOsman

@0xSero The untold story of having megawatt machinery at home.

I have experienced the most stressful 2 hours of my life. Someone we hired to fix our lights tripped the circuit. My life flashed before me. I regretted ever existing. Still recovering

@bcherny @icanvardar Looks like you have an answer for everything but your models became somehow less than what they was

👋 1h prompt cache is nuanced actually. It costs more for cache writes, and less for cache reads. Whether you benefit from cheaper cache reads depends on your usage pattern -- context window size, whether the query is the main agent or subagent, etc. We have been testing a number of heuristics to give subscribers better prompt cache hit rates, which means lower token usage and lower latency, when it works. But this effect is far from uniform due to the nuance above. Say you use 1h cache for an agent, but only used the agent to make a single query -- in this case the 1h cache would be wasted and you'd be overcharged. At this point we have rolled out 1h prompt cache by default in a number of places for subscribers to optimize cache duration based on real usage patterns, but we actually keep it at 5m for many queries also (eg. subagents, which are rarely resumed so you'd be paying for them even though they do not benefit from 1h). We also are not defaulting API customers to 1h yet -- this needs more testing to make sure it's a net improvement on average. Separately, when we do this kind of experimentation, we use experiment gates that are cached client-side. When you turn off telemetry we also disable experiment gates -- we do not call home when telemetry is off -- so Claude reads the default value, which is 5m. We will soon be changing the client side default to 1h for a few queries, since we now feel good that it is a small token savings on average for those queries. We will also give you env vars to force 1h and 5m. In any case, the token savings is nowhere near 12x unfortunately. It is a small win though, that we have been in the process of rolling out to everyone. Hope the explanation helps. More here: #pricing" target="_blank" rel="nofollow noopener">platform.claude.com/docs/en/build-…

wait… claude code literally punishes you for turning off telemetry?? if you disable it, anthropic drops your cache from 1 hour to 5 minutes so in claude code, anthropic basically becomes an evil corp where privacy costs you 12x performance… am i reading this right?

@0xSero Got the big promo they did before glm-5. 180 year for x5 no weekly limits. Best value ever, glm never let me down

Droid - GLM-5.1 I asked it to install Ghidra Pointed at compiled typescript for a popular product "Reverse engineer this and provide me code that compiles to it then make it so I can use the app from my own dev build" 20 million tokens 1 shot later Full application, compiles down to the same thing, usable, can rip out any BS Making a bet, ZAI will have the first non-western #1 model in almost all benchmarks soon. Don't discount the value of training data, people who use ZAI are especially from the west have a higher technical cieling. Model is excellent. 432$ a quarter well spent.