Knowledge catalogue
arXiv:2510.21285v4 Announce Type: replace Abstract: Large Reasoning Models (LRMs) achieve strong performance on complex multi-step reasoning, yet they still exhibit severe safety failures such as harm
arXiv:2604.21993v1 Announce Type: new Abstract: We study the detection of transient liquidity erosion ('crumbling quotes') in electronic limit order books, where observable quote deterioration may ref
When violence is categorized accurately, this unequivocally true. You know who the bad guys are, because they’re the ones who want censorship of free speech. We have a left-wing violence problem in Am
Devin, an AI coding assistant, enables developers to hand off work from their local laptop to cloud-based processing via CLI commands, allowing tasks to continue running in the cloud environment witho
arXiv:2604.22127v1 Announce Type: new Abstract: Hybrid language models that interleave attention with recurrent components are increasingly competitive with pure Transformers, yet standard LoRA practi
arXiv:2604.22096v1 Announce Type: cross Abstract: In enterprise fraud detection, model accuracy alone is insufficient when insiders can tamper with audit logs or bypass approval workflows. Real-world
arXiv:2604.22102v1 Announce Type: cross Abstract: Many robotic tasks are unforgiving; a single mistake in a dynamic throw can lead to unacceptable delays or unrecoverable failure. To mitigate this, we
NASA's SpaceX Crew-13 mission, comprising astronauts from the
Xiaomi MiMo-V2.5 is now officially open-sourced！ MIT License, supporting commercial deployment, continued training, and fine-tuning - no additional authorization required. Two models, both supporting
yep, that’s what I do. I am increasingly concerned that people in China are soaking up my wisdom, while a lot of people in the US try to dismiss me. i don’t think that is going to end well. 10. @GaryM
❤️ @ying11231, @BanghuaZ and @lmsysorg @sgl_project @radixark Let's go DeepSeek v4 Pro! The DeepSeek V4 garbled output bug in open source inference engine is fixed in SGLang. To everyone affected over
You can look at the history of media through the cost of producing a credible frame. I call this the cost of a frame. Not a good story. Not a good movie. Not a meaningful work of art. Just the frame i
Without trusted data as the foundation, even the most sophisticated models will produce outcomes that enterprises can’t act on with confidence. As a matter of fact, enterprises are paying a steep pric
As enterprises navigate the complexities of scaling AI initiatives, the agentic AI blueprint for success lies in combining deep process intelligence with powerful cloud platforms, enabling organizatio
arXiv:2604.22723v1 Announce Type: cross Abstract: We present a method for discovering morphological features in low-resource Bantu languages by combining cross-lingual transfer learning with unsupervi
25% of very liberal people say political violence can “sometimes be justified.” Among very conservative people it’s just 3%. In other words, the far left is 8x more likely to support political violenc
3 hr notice I had over 50 people come to the coffee shop incl a guy who brought his GPU for me to sign. Crazy! Media If anyone wants to meet in New York City today, I'll be at Dark Matter Coffee, NYC,
7) Multi-agent design. I loved the design of Cove (they were the acquired by Microsoft). It was more of a whiteboard than tabs or chat threads. I don’t think we’ve cracked the right UI for managing ag
This post demonstrates a 7-second animation created by combining multiple AI models within ComfyUI and Adobe After Effects, utilizing tools including Wan Animate, Wan 2.2 I2V (image-to-video), SCAIL,
I cannot provide a summary for this entry because the source URL and content appear to be fictional or corrupted (evidenced by the strikethrough text and impossible X post ID). Without verifiable info
Benjamin Wallace / New York Times: A profile of Dwarkesh Patel, whose podcast has become mandatory listening in the AI community, with guests like Jensen Huang, Elon Musk, and Mark Zuckerberg — Dwarke
A thing I see missing from AI job debates is that as jobs are disrupted by AI, professions will compete over new boundaries Abbot's System of Professions is a good read on what happens historically. H
This post announces a 'Proactive Agent Build Day' event hosted by AGI House in collaboration with Even Realities, with registration available through their events platform. The event appears to be foc
“AI: it feels most impressive when you don’t know much about the subject, don’t care or don’t have a clear idea of what the you want… Once you know what something should feel or look like, it becomes
All your jobs are belong to us Then surely Anthropic is making a big mistake by hiring all those people huh? Strange since they would be expected to know the most about the next frontier. With OpenAI
Clem Delangue argues that the United States should increase investment and policy support for open source AI models to maintain competitive advantage and reduce dependence on proprietary systems contr
Stephen Totilo / Game File: An interview with Xbox CEO Asha Sharma and EVP Matt Booty on the “Return of Xbox” memo, making Xbox Series X and S the “first-class experience again”, and more — In an excl
Anthropic’s CEO says software engineering is dying. Anthropic’s job listing has 70 open positions in software engineering. 🙄 @GaryMarcus https://www.anthropic.com/careers/jobs 70 opened position in SE
April saw significant activity in large language model releases, with five major models introduced including Gemma 4, GLM-5.1, Qwen 3.6, Kimi K2.6, and DeepSeek V4. These releases have been added to a
AUTOMATIC1111 remains the most popular choice for most users , though ComfyUI offers node-based workflows with more control and faster performance, while AUTOMATIC1111 provides a traditional interface
Kim Mackrael / Wall Street Journal: ASML says it plans to make at least 60 of its standard EUV machines this year, 36% more than it sold in 2025, as it races to meet demand for making AI chips — ASML
Release b8934 of llama.cpp was released on April 26, 2026, and includes improvements to hexagon hardware support, specifically guarding HMX clock requests for v75+ platforms. The release provides bina
This release adds OpenCL GPU acceleration support for the IQ4_NL quantization format in llama.cpp, enabling more efficient inference of quantized language models on compatible hardware. IQ4_NL is a 4-
llama.cpp release b8936 changes the default quantization parameter type from LLAMA_FTYPE_MOSTLY_Q5_1 to LLAMA_FTYPE_MOSTLY_Q8_0, providing a more reliable default for external programs using the quant
b8937 is a llama.cpp release that re-enables fast gelu_quick_f16 support and changes the default quantization parameter from LLAMA_FTYPE_MOSTLY_Q5_1 to LLAMA_FTYPE_MOSTLY_Q8_0 . The release provides b
b8940 is a build release of llama.cpp, an open-source C/C++ library for LLM inference that enables running large language models locally with minimal setup and high performance. This intermediate buil
@badlogicgames Yup: https://x.com/antirez/status/2048344770234249588 The GGUF tool calling template is wrong but I'm uploading a new GGUF file. Otherwise there is the right template file in one of the
🚨BREAKING: I found Cole Allen's archived tweets. He predicted 'Kamala wins all swing states,' compared Trump's win to “Nazis getting elected,” and moved to Bluesky. The scary part: he retweeted every
ChatGPT's GPT Image 2 is actually incredible. Full visual breakdown of how to fix my desk setup. I refuse to get rid of my second monitor (I need 3 for live zoom client workshops), but this was the ki
Gary Marcus, a prominent AI researcher and critic, posted a message on X (formerly Twitter) addressing software engineers and coders, likely discussing technical aspects of AI development, programming
Chris Metinko / Axios: Collov Labs, whose visual interface lets users feed images and camera input into a model that AI agents can reason over and act on, raised a 23M Series A — Collov Labs, which tu
Continued weak spots of AI, from the point of view of a business professional and not a PhD biochemist: 1) SVGs. The ability to 'illustrate' and have that thing be infinitely scalable. See below image
🔥DeepSeek Input Cache Price Drop! Effective immediately, the price for input cache hits across the ENTIRE DeepSeek API series is reduced to just 1/10th of the original price! Build more efficiently fo
DeepSeek v4 Flash with *local inference* after 24h of playing with that: even with the 2 bit selective quantization GGUF, iti is the FIRST time I feel I have a frontier model running on my computer. T
SpaceX successfully deployed 25 Starlink satellites into orbit, as confirmed by Elon Musk on X (formerly Twitter). This deployment is part of SpaceX's ongoing effort to expand the Starlink satellite i
Describe Seattle: I want to live near the sea and I want to live near the forest and I want to live near the mountains and I want to live near a café and I want to live near a library and I want to li
disconcerting. A European Commission proposal could create one of Europe’s largest privacy and national-security risks in decades. https://techletters.substack.com/p/the-european-commission-is-turning
Doomers have made the AI industry much of what it is today, by hyping the narrative. Chamath captured the current state of AI perfectly in his 2025 annual letter 'In testing the most effective messagi
Stephen Morris / Financial Times: Epoch AI: Google controls ~25% of global AI compute, with ~3.8M TPUs and 1.3M GPUs; Google Cloud CEO Thomas Kurian says demand and revenue justify the spend — Thomas
Existential risk mongers are a small, very vocal cult with a lot of very clever online astroturfing skills. Politicians never waste a good fake crisis, which is why they're perfect for Bernie to try t
extremely happy that we are in q2 2026, and engineers i look up to are plowing a path for the local model future. yes, we are still beholden to some lab publishing weights. but i take that over compan
Sam Altman suggests the need for fundamental redesign of operating systems, user interfaces, and internet architecture, implying current designs may be outdated or inadequate for modern needs. The pos
Fieri Iussit https://x.com/emollick/status/1969993499157446943?s=20 My modest proposal for how to acknowledge AI work is to use the Latin phase 'Fieri Iussit,' which means 'commanded to be made,' a co
FSD v14.3.2 has been driving me all over LA tonight. This build is f-ing incredible, smooth as butter and it’s fascinating how well it can handle complex traffic scenarios, road closures and tight roa
FYI Claude Code is mostly a vibe-coded product (as they say, 100% written by Claude) It's the worst harness for Opus 4.6 among ANY harness on Terminal-Bench 2 I feel sorry for Claude Code I know they'
Gary Marcus reported receiving a high volume of phishing emails through X (formerly Twitter) direct messages, suggesting either widespread account compromises or a potential security vulnerability all
🚨GUYS QWEN3.6 27B/35B @unslothai MLX QUANTS @Brooooook_lyn cooked 🔥 Apple Silicon users, the exact models y’all have been asking for LANDED. Unsloth mixed-precision quants in native MLX format. >Fast.
HealthBench Professional, our new evaluation of clinician chat tasks, is now available on HuggingFace for easy access. Each example was written, reviewed, and adjudicated by three or more physicians.
Here is a very common problem when building complex agents. Long-horizon agents (in particular) fail in two ways: the decision-maker can't decompose well, or the skill library goes stale. This new res
Hermes Agent tip of the day: There are 4 ways to deal with the model while its running, - Message it, by default, it will interrupt the agent loop, stopping it and making it respond to your new messag