Prof Adebayo
4.9K posts
Prof Adebayo
@ProfAdebay
CTO @Academicnight | i literally want ai to replace you.
you can't reject light forever انضم Mayıs 2023
903 يتبع1.2K المتابعون
@ProfAdebay I pulled the server out of a scrap bin. The RAM, 3060, and the CPU were purchased Q1 2025. Everything else was this year.
English
I have no clue how this monstrosity even booted.
RTX 3090 FE 24GB
2x RTX 5060TI 16GB (32GB)
Quadro RTX 5000 16GB
72GB total VRAM
256GB system RAM
22 core 44 thread intel xeon processor
All in cost ~$3000
#opensource #localllama #qwen #ai #nvidia
English
@0xSero @AmnonTamsut You shouldn't say this in public post @0xSero
it's a coding secret reserved fr senior devs.
English
@RealTjDunham @DSBatten that will be good. troll it at me when it's out
English
Bitcoin grew out of the Internet, but has now become more resilient than the Internet. A $6 feature phone, a 2G network, even mesh radio (like a low-tech walkie-talkie) can send money anywhere in the world.
Fee-free. Bank-free. Middleman-free. Delay-free
Bitshala@bitshala_org
No internet? No problem. Bala just dropped the ultimate cypherpunk demo at the BOSS Summit: From Off-Grid to On-Chain He literally broadcasted a live Bitcoin transaction using Mesh Radio. No ISPs, no Wi-Fi, no cellular data. Just pure radio waves bypassing the traditional internet layer to hit the mempool. When we say we are building unconfiscatable money for uncertain times, this is exactly what we mean. Mind blown. Here is the repo used to connect meshtastic to bitcoin core:github.com/BTCtoolshed/Me…
English
This is Farzapedia.
I had an LLM take 2,500 entries from my diary, Apple Notes, and some iMessage convos to create a personal Wikipedia for me.
It made 400 detailed articles for my friends, my startups, research areas, and even my favorite animes and their impact on me complete with backlinks.
But, this Wiki was not built for me! I built it for my agent!
The structure of the wiki files and how it's all backlinked is very easily crawlable by any agent + makes it a truly useful knowledge base.
I can spin up Claude Code on the wiki and starting at index.md (a catalog of all my articles) the agent does a really good job at drilling into the specific pages on my wiki it needs context on when I have a query.
For example, when trying to cook up a new landing page I may ask:
"I'm trying to design this landing page for a new idea I have. Please look into the images and films that inspired me recently and give me ideas for new copy and aesthetics".
In my diary I kept track of everything from: learnings, people, inspo, interesting links, images.
So the agent reads my wiki and pulls up my "Philosophy" articles from notes on a Studio Ghibli documentary, "Competitor" articles with YC companies whose landing pages I screenshotted, and pics of 1970s Beatles merch I saved years ago. And it delivers a great answer.
I built a similar system to this a year ago with RAG but it was ass.
A knowledge base that lets an agent find what it needs via a file system it actually understands just works better.
The most magical thing now is as I add new things to my wiki (articles, images of inspo, meeting notes) the system will likely update 2-3 different articles where it feels that context belongs, or, just creates a new article.
It's like this super genius librarian for your brain that's always filing stuff for your perfectly and also let's you easily query the knowledge for tasks useful to you (ex. design, product, writing, etc) and it never gets tired.
I might spend next week productizing this, if that's of interest to you DM me + tell me your usecase!
Andrej Karpathy@karpathy
Wow, this tweet went very viral! I wanted share a possibly slightly improved version of the tweet in an "idea file". The idea of the idea file is that in this era of LLM agents, there is less of a point/need of sharing the specific code/app, you just share the idea, then the other person's agent customizes & builds it for your specific needs. So here's the idea in a gist format: gist.github.com/karpathy/442a6… You can give this to your agent and it can build you your own LLM wiki and guide you on how to use it etc. It's intentionally kept a little bit abstract/vague because there are so many directions to take this in. And ofc, people can adjust the idea or contribute their own in the Discussion which is cool.
English
@FarzaTV honest question: why didn't you used RAG for it?
Though this method of fetching data seems faster but i think it might not be effective for large files
English
English
@Alibaba_Qwen @OpenRouter The popularity would explain why I'm continually seeing rate limit exceeded errors on @kilocode
English
Qwen3.6-Plus ranks # 1 on @OpenRouter , and the first model on OpenRouter to break 1 Trillion tokens processed in a single day！!🥇🔥
We are thrilled to see Qwen3.6-Plus topping the charts so quickly. This milestone wouldn't be possible without our amazing developers. ❤️Thank you！！
OpenRouter@OpenRouter
Qwen 3.6 Plus from @Alibaba_Qwen is officially the first model on OpenRouter to break 1 Trillion tokens processed in a single day! At ~1,400,000,000,000 tokens, it’s the strongest full day performance of any new model dropped this year. Congrats to the Qwen team!
English
That’s the way forward, hopefully it gives their users more compute headroom to breathe. Anthropic is essentially prioritizing how its resources are allocated, even if that means turning down certain user demands and directing them to other models. It’s a calculated move, they understand their positioning in the market and are managing access in a way that maintains performance and exclusivity. One thing is certain: a large portion of the users they’re redirecting will likely keep their subscriptions, even if they can’t fully useit with Openclaw.
English
Claude is cutting off apps like OpenClaw from using Claude subscriptions.
I read somewhere that a $200/month Claude plan can use up to $5,000 in compute, so it’s heavily subsidized. Given Claude’s uptime issues, this might be the right move under current Anthropic GPU constraints.
Codex is the more generous one for 3rd-party apps (OpenAI has more GPUs). It’ll be interesting to see how this strategy difference plays out.
English
@steipete @abustamante @davemorin can you share with everyone best replacement models to replace anthropic?
English
woke up and my mentions are full of these
Both me and @davemorin tried to talk sense into Anthropic, best we managed was delaying this for a week.
Funny how timings match up, first they copy some popular features into their closed harness, then they lock out open source.
mvpr@marinatedvapor
English
@kamellperry_ @TheAhmadOsman for codex is solid (esp the xhigh is approx Opus level somehow)
unless you're working on frontend
English
@TheAhmadOsman I use most of these and claude code still handles long running agentic tasks better than anything else I've tried. codex is solid but it's a different workflow entirely
English
@ProfAdebay thats my one complaint about cursor is you get locked out for weeks if you hit your limits fast.
English
I did it. Found a way to greatly reduce usage rates in Claude.
step 1:
navigate to your .claude folder and open settings.json
make the following changes:
json
{ "model": "claude-sonnet-4-5", "compactThreshold": 200000, "subagentModel": "claude-haiku-4-5-20251001"
}
save and close the file
step 2:
back in your terminal, run claude to open a new session
- once you're inside, type `/effort medium`
- press enter to save
step 3:
now let's make sure we're not using unnecessary additional tokens in outputs.
type /config
scroll down to "verbose output"
press space bar and set to "false"
step 4:
good. now for the most important step
press ctrl+c once to pause the session
you'll see a prompt asking if you want to exit, press ctrl+c again to confirm
keep your terminal open
step 5:
type the following:
`npm i -g @openai/codex && codex`
Lydia Hallie ✨@lydiahallie
Thank you to everyone who spent time sending us feedback and reports. We've investigated and we're sorry this has been a bad experience. Here's what we found:
English
@LLMJunky Yeah… I can’t risk a $200 plan and then hit the weekly limit within 24 hours (and get lockeddown for 7 days). I’d rather keep subscribing with a new account once I hit the weekly limit.
on 4th acc today...
Codex is more budget friendly for heavy usage... than Claude
English
@LLMJunky I'm currently on my 4th codex subscriptions (account). all 3accs on hit weekly limit.
I just subscribed 4th today.
English
@ProfAdebay @Every_Other_Odd It wants Qwen3.6-9B. Qwen3.5-9B and 4B absolutely destroys Gemma 4 E4B.
English
We are planning to open-source the Qwen3.6 models (particularly medium-sized versions) to facilitate local deployment and customization for developers. Please vote for the model size you are **most** anticipating—the community’s voice is vital to us!
English
@vaiduakhu @Every_Other_Odd @ChujieZheng feed it with E2B; it will run it with speed of light.. I believe you will enjoy it 😊
English
English
Folks, pls vote Qwen3.6 27B
Chujie Zheng@ChujieZheng
We are planning to open-source the Qwen3.6 models (particularly medium-sized versions) to facilitate local deployment and customization for developers. Please vote for the model size you are **most** anticipating—the community’s voice is vital to us!
English
@MichaelAntonF @ChujieZheng MoE still wins on raw scale, and this newer generation might be better than older generation
English
@Every_Other_Odd @ChujieZheng ..that's because it wants Gemma 4.
English
