I think OpenAI is more like openrouter because at 2 am it's auto switching to Chinese models and I get Chinese in my output. Wtf

DeepSeek v4.1 Flash Fully Uncensored for 2x DGX Spark users - No damage to knowledge at all - VL working I think I’ve perfected figuring out where refusal behavior related to “ethics” and “morals” live - with my method and pipeline for dsv4.1flash, the only areas in MMLU logit mode that were hit are literally only within the “ethics” topic, resulting in less than a -1% difference in scores when comparing to the base (12% difference when considering the drop in ethics score) I put the recipe on how to run and use this model within the readme. huggingface.co/dealignai/Dee…

oregano oil killed every single bacterial strain it was tested against in a Georgetown University study including strains that were resistant to pharmaceutical antibiotics. every strain. zero survivors that sentence should have made international headlines. a plant extract outperforming prescription antibiotics against resistant bacteria is the kind of finding that changes medicine. it didn't make headlines because a $12 bottle of oregano oil can't fund a pharmaceutical marketing campaign the mechanism is carvacrol and thymol. two compounds that destroy bacterial cell membranes on contact. the bacteria cannot develop resistance to this mechanism because the compounds attack the physical structure of the cell rather than a specific metabolic pathway. antibiotics target pathways. bacteria evolve around pathways. oregano oil targets the wall itself. you can't evolve around having your wall destroyed • killed staphylococcus aureus, E. coli, and pseudomonas aeruginosa in laboratory conditions at concentrations achievable through oral supplementation • eliminated H. pylori (the bacteria behind most stomach ulcers) in clinical use when pharmaceutical triple therapy failed. a backup treatment that outperformed the first line • antifungal activity against candida albicans comparable to nystatin (a prescription antifungal) in direct comparison studies • antiviral properties demonstrated against norovirus and herpes simplex virus in laboratory settings. one compound active against bacteria, fungi, AND viruses simultaneously • reduces SIBO (small intestinal bacterial overgrowth) symptoms as effectively as rifaximin ($1,500 per course) in a Johns Hopkins affiliated study. a $12 oil matching a $1,500 antibiotic • the antibacterial mechanism does not destroy beneficial gut bacteria at normal supplemental doses because the beneficial strains have naturally thicker cell walls. the selective toxicity is built into the chemistry the pharmaceutical industry generates $46 billion annually from antibiotics. a $12 bottle of oregano oil matching or exceeding multiple prescription antibiotics in laboratory and clinical settings is an existential threat to that revenue. the research exists. the FDA classification prevents health claims. the gap between what the science shows and what the label is allowed to say is where the profit is protected 2 to 4 drops of emulsified oregano oil in water 2 to 3 times daily for acute infections (limit to 2 weeks max). for maintenance: 1 drop daily or every other day. must be emulsified or enteric coated to prevent esophageal irritation. north american herb and spice (Oreganol P73) is the most researched brand. $18 for 2 months a plant that has been used as medicine for 4,000 years killed every antibiotic resistant strain thrown at it in a Georgetown University laboratory. the pharmaceutical industry responded by making sure you never heard about the study.

DeepSeek V4.1 Flash being this good makes Anthropic’s $2T IPO sound crazy stupid That’s why they hate Opensource AI and wanna ban it

Today I've been putting Hermes Agent to the ultimate test of dramatically cleaning up the Hermes Agent repo. One /goal and it's been working for ~15hrs now and has simplified, unified, optimized, removed, or in some way cleaned up the repo enough to shed 375,000 lines of code. it's had waves of ~120 subagents, 3x sets of 15 subagents with all of those subagents able to spawn their own subagents (yes, recursive subagents all the way down) for almost the entire time - and my single desktop machine has handled it like a champ. Pretty epic test of the limits of both the latest gen models and Hermes Agent Stay tuned for more!

GLM-5.3 FP8 made for Offensive Pentesting and CyberSecurity. We’ve all heard about how GLM-5.2 helped huggingface throughout the OpenAI attacks; I wanted to make a version specifically in which has as highest of quality possible while the ablation is targeted mainly towards allowing users to be able to have no refusals for somethings that may be considered “blackhat”. Made by dealignai & @jordanschenck huggingface.co/dealignai/GLM…

Walmart is installing digital price tags across all 4,611 US stores by end of 2026. Kroger already has them in 25% of locations. The FTC has confirmed personal data is being used to set individual prices. Here is what to do:👇

We’ve built exactly this :) AO lets you manage Claude Code, Codex, Cursor, and 26 coding-agent harnesses from one desktop app. aoagents.dev github.com/Untrivial-ai/agen…

A Bloomberg Terminal costs $30,000 a year. Here's how to build 90% of it for free. Wall Street doesn't advertise this. But every function that matters has a free alternative in Python. Here's the DIY version (with Python):

IVERMECTIN & MEBENDAZOLE should be sold over-the-counter and available to everyone. New study shows 84.4% Clinical Cancer Benefit — Nearly 50% Report Cancer Disappearance. Dr. John Campbell... "We could be CURING your wife, your husband, your parents & your children."

It will be common knowledge in 100 years. It remains unknown today by most. We will have to wait perhaps 100 years for the surrounding society and professionals to mature to accept real empirical science. Dr. Jerry Tennant’s voltage insight on cancer is one of the most elegant and empowering reframes in modern health. He shows that healthy cells run at roughly –20 to –25 mV. Healing and new cell creation require –50 mV. When voltage collapses past zero and reaches +30 mV, polarity flips, oxygen delivery fails, and the body is forced into the only emergency response left: cancer. In other words, cancer is not a mysterious genetic rebellion—it is the predictable result of a drained cellular battery. Restore the voltage, reverse the polarity, and the terrain that allows cancer simply ceases to exist. That single observation turns a terrifying diagnosis into a solvable energy problem. And it has very strong duplicatable evidence for treatments. Bookmark this for 2126.

BOOM! For the first time, more than 50% of internet traffic is now non-human driven by AI agents and bots. Cloudflare data shows automated traffic has overtaken humans (hitting ~57% in key metrics earlier this year). CEO Matthew Prince called it out in June, noting it arrived way ahead of the 2027 prediction. The agentic internet is here. Publishers and sites now face new realities around content access, controls, and monetization. What does this mean for the open web?

People have no idea how much faster Local AI actually is Thousands of tokens per second on the fly Once we get that Kimi K3 quality on a single RTX PRO 6000, I am not sure I will keep any of my subscriptions

BREAKTHROUGH: A full, unmodified 2.78-trillion-parameter Kimi K3 on a consumer laptop by streaming only the activated experts from NVMe. YOU CAN’T RUN KIMI K3 “ON THAT” THEY DECLARED. There are many paths to do it. This is one: Marco Bambini Just Gave Us the Full Kimi K3 on a Laptop Meet Marco Bambini he did something that felt impossible only a day ago. He built WASTE Weight-Aware Streaming Tensor Engine a clean, dependency-free C inference engine that runs the complete, unmodified 2.78-trillion-parameter Kimi K3 model by streaming only the activated experts straight from NVMe. No distillation. No pruning. No cloud. The full open-weight model. We have it running in the lab right now. What Marco Actually Built Kimi K3 is a sparse Mixture-of-Experts system. Only about 4 % of its weights fire on any given token. Marco’s insight was simple and ruthless: the idle experts do not need to live in RAM. They only need to be reachable in time. WASTE keeps the model’s “trunk” (attention, shared components, embeddings) resident in memory — roughly 27 GB on the converted container. The 82,000+ routed experts stay on disk as tightly packed residual vector-quantized records. When the router selects its 16 experts per layer, the engine issues direct, cache-bypassing reads from the internal NVMe and feeds them into a bounded expert cache. The rest of the machine’s RAM becomes working space for that cache. On a 64 GB MacBook Pro with the container on the internal SSD, we are measuring 0.32–0.34 tokens per second at a comfortable memory budget. Prefill sits a little higher. The vision tower works. Logits match the reference implementation to within a few parts in a million. It is the real model. The container itself is 982 GiB after conversion from the original 1.42 TB MXFP4 weights. Minimum RAM floor is just over 29 GB for short context. Push the budget higher and the expert cache hit rate climbs; push too high and you start paging and the speed collapses. The sweet spot on current consumer hardware is clear and measurable. How We Are Testing It We converted the official weights, verified the container, and began systematic runs the same day the engine stabilized. First we confirmed numerical fidelity against the PyTorch reference on short prompts. Then we moved to longer generation, vision inputs, and multi-turn chat using Kimi’s native XTML format. We are measuring wall-clock decode, expert I/O versus compute split, cache hit rates at different RAM budgets, and thermal behavior under sustained load. We are also exercising the OpenAI-compatible server that sits on top of the same C library so we can drop the model into existing agent loops without rewriting anything. Early observations: •Expert I/O dominates the timeline, as expected. On a fast internal NVMe the engine is already near the practical ceiling of the storage subsystem. •The architecture’s sparsity is the entire enabler. A dense model of this size would be dead on arrival for local use. •Context length is currently limited by RAM more than by the model itself. Practical working contexts sit comfortably in the tens of thousands of tokens on 64 GB hardware; the full million-token window will need more memory or smarter KV management. •Thinking tokens are expensive at this speed. Long internal monologues turn into multi-hour runs. For agent work we are already experimenting with tighter control over when full reasoning is requested. We are treating this as a research instrument, not a finished product. Every run teaches us something about expert locality, prefetch opportunities, and how far pure software streaming can push trillion-scale inference on ordinary machines. 1 of 2

unsloth already done with their Kimi K3 quant

Google just released free 1-hour course on building agentic knowledge Graphs from 0% to 100%: 10% → 4:01 - how to build a GraphRAG agent 30% → 15:00 - Graph Engineering explanation 55% → 30:00 - Agentic search Engineering 80% → 35:48 - Graph Engineering practice 100% → 47:06 - self-improving agents in graphs this free Google course mass replaces a $500 graph engineering bootcamp - learn it in 60 min to 100% watch it today - then read the full graph playbook in the article below ↓

We've open-sourced AgentENV in collaboration with kvcache-ai. AgentENV is a distributed system for running agent environments at scale. Its components power agentic RL training for Kimi K3, with fast snapshot, resume, and fork support for large-scale parallel agent workflows. Explore on GitHub: github.com/kvcache-ai/AgentE…

Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Model weights: huggingface.co/moonshotai/Ki… Tech report: github.com/MoonshotAI/Kimi-K… Tech blog: kimi.com/blog/kimi-k3

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-W…

Kimi's CEO Zhilin Yang: "everyone's trying to build one smarter agent many agent hits a wall fast, so instead of making it smarter we just made more - one boss, a thousand workers." in a 39-minute talk he explains why one agent won't get you to real work. agent swarms + long context + RL on every sub-agent - that's the fix. bookmark this ↓