HOARK

I open-sourced Wave-Density Attention, a new causal attention mechanism that replaces dot-product similarity with wave interference. Instead of QKᵀ, attention emerges from superposed sinusoidal masks whose interference patterns form dynamic density fields. A sparse Mixture-of-Masks (MOM) router activates only a small subset per token. Result: a 130M model trained from scratch hits ~18 perplexity on instruction data — on a single GPU. This isn’t an optimization of transformers — it’s a different primitive for attention. Code: github.com/H0ARK/wave-den… Model: huggingface.co/H0ARK/wave-den… Try it. Break it. Improve it. #MachineLearning #DeepLearning #OpenSource @karpathy @huggingface @arankomatsuzaki @tri_dao @jimmyba

@zuess05 at senior level that made me laugh.

I am actually curious. If junior devs are using Claude to write code at a senior level... And senior devs are using Claude to write code at a senior level... What exactly is the difference between the two roles right now, other than a $100k gap in salary?

@rxhit05 5.4 was better at agentic task completion and less trying to get done regardless of how complete it is

@rxhit05 for smarts yes but 5.5 is really anoying where i have to baby it to keep going

codex 5.5 >>>> opus 4.7 who else feels claude has fallen off?

@thegenioo @OpenAIDevs @thsottiaux 5.5 uses 2x the usage because on small tasks its the same as 5.4 but you have to remember it also costs 2x over 5.4 so easy tasks you should use 5.4

GPT-5.5 (low) working for 3 minutes on a very small repo of a very simple scheduling Electron app burns 8% of my 5h limit. I was in a new chat, so no extended context. And it was a very simple task like extremely simple. I have attached the diff screenshot too. Am I going crazy or is this normal?

@ScriptedAlchemy i would do 5.1 with turbo quant

I have a B200 cluster. Looking to fine tune a model. What’s the best OSS model that’s near GPT 5.4 or opus 4.7

@mumtor99 @zachlloydtweets yes lol thats the first thing im doing right now.

@zachlloydtweets Does this mean I can use the Codex models via the ChatGPT subscription with Warp? Can I submit a PR for this change?

@ImKingGinger "if everyone is super no one is"

If everyone gets a million dollars. No one has a million dollars. I'm not sure why no one understands this.

@Yuchenj_UW my model does this i trained mine on a single rtx 3090 took abount 2 hours

OpenAI just dropped a training challenge: Train a <16MB language model in 10 minutes on 8×H100s and minimize held-out loss on a fixed FineWeb dataset. Basically NanoGPT Speedrun. They’re sponsoring $1M in compute. I can summon my autoresearch army to win it… if I have time.

@jawwwn_ @elonmusk @PeterDiamandis @elonmusk I feel grok4.20 is just a fraction of what you really have locked up cooking in the oven. Don't get me wrong its awsome but its how you talk makes me think there is. also when grok code 4.20?

.@elonmusk: “We are in the singularity.” “I think we’ll hit AGI in 2026.” “You’re at the top of the rollercoaster about to go down.” “Don’t worry about squirreling money away for retirement. It won’t matter.” “I don’t just have courtside seats— I’m on the court. It still blows my mind multiple times a week.” Via @PeterDiamandis

As an engineer I’ll spill the way it happens. People think user<->ai and that’s not the case there are many filters load balancing and content moderation that is in between the user and the chat even though it’s encrypted machines can read the content and if it is flagged as “dangerous” open ai has the right to look at it (tos). You’re really talking to 4-5 models before you get to big daddy gpt.

@KatieMiller I just thought it will be encrypted or at least anonymized 😭

Canadian trans shooter was flagged via internal systems at Open AI for writings about real-world violence, including gun violence. Over a dozen Open AI employees debated telling law enforcement. OpenAI leaders decided not to inform authorities about a potential mass murder.

@neural_avb Add 1 kv and long context

LLM research in 2026 is basically all about 3 things: - scaling Asynchronous RL at a system design level - harnesses/recipes to train independent (clawdbot style) agentic systems - make low-level architectural choices that achieve this cheaply

@mziizm I would love to work with people in bleeding edge ai research. I have been working on some very interesting research ideas.

We're hiring a Research Engineer who understands models at a deep technical level and excited to take responsibility across the full lifecycle. If you're excited to join a small team driving research with real-world impact, we'd love to hear from you. shorturl.at/EJxxr

@wtfshyraaa Hey would love to chat

this ai startup is hiring remote across engineering AND marketing. they're building the best web scraping infra on the internet. devs mass love their product. 150k-250k$ (depending on the location) no return to office. no bs. fully remote. roles won't stay open long. DM me and I'll send you the company name + links. 💸 #Hiring

@ezpedersen_ I would work with you hit me up.

Hiring engineers at Virio > $250k-500k > Senior SWE, Founding Engineers, Product Engineers We are looking for cracked grinders. Cracked = knows how to do shit Grinder = will actually do shit no matter what If you want to own your life and own this company, DM me

@christophersaum i got you how about the digital card collection marketplace? real playing cards digitized for rapid trading. rippzz.com

Please don’t pitch me your digital twin startup - I’ve already seen 5 today. If you’re an AI-native founder building a non-obvious startup & not the 100th version of the same thing then I'd love to chat. We’re backing at least one new team every month with $100K to $1M checks

@BernieSanders by bringing back the oldschool family structure where single income families make what they used to back in the day.

Amazon says that it’s going to replace 600,000 workers with robots. Other companies are moving in the same direction. How will working families feed their kids and pay their bills? We’ve got to fight back.

@aingtalks Full stack engineer here lets chat. rippzz.com

looking for 2 founding engineers (equity only) 1) data systems engineer 2) frontend engineer (react) early stage. no salary yet. if you’re serious about building something ambitious, hit a text.

the only real good way to solve this that does not include slapping a fat model on top is to sanitize visual data into precise actionable data before feeding in that data back into the main (fast) model. i would do something like use a masking agent that would mask the screen and the ui components and only feed thoes components into the agent (a cleaned dom with out any styling or fluff). the hard part is intermediate state change so interupting the model with the new page state would have to happen and if you can make it so the model can respond very quick. screen shots are not the way its way to slow. this is something i have been planning on trying to solve myself as a project.

Solve this in under 5 minutes and I’ll offer you $500k/year in cash plus several million in equity I'm building a Computer-Use team, goal is to use computers better than humans No experience or PhD needed Instructions: 1. Solve all 30 challenges on this website in under 5 minutes: serene-frangipane-7fd25b.netlify.app 2. Feel free to use any tools or vibe code it. Provide us a zip folder with instructions on how to run the agent and reproduce your results, as well your run statistics 3. The agent should be able to solve all the challenges, use browser, and provide overall metrics around time taken, token usage and token cost. Your agent must solve this challenge in under 5 minutes Email your response: agents@brettadcock.com If you have any questions about this challenge, feel free to email us

@akbuilds_ sup

I gained 196 followers in one day. Still feels unreal. Growth happens when you start conversations. If you're in Tech, AI, Startups, Indie Hacking, say Hi 👋

@trikcode @trikcode been working on a few things myself lets connect.