Knowledge catalogue
I know everyone's excited about 3.8, but honestly I'm still refining 3.6. And I hate the way 3.8 talks, so... I keep seeing all these posts about what's the use of open source, so I figured I'd submit
arXiv:2610.21055v1 Announce Type: new Abstract: Recursive self-improvement, in which AI systems build stronger AI systems, is one promising route to superintelligence. We study a weaker form that can
Banger paper from Anthropic, Harvard, and colleagues. They find that LLM-proposed quantum circuits beat the best-known designs on seven of eight tasks, with fewer Toffoli gates and fewer qubits. They
arXiv:2601.01982v2 Announce Type: replace Abstract: Large language models (LLMs) excel at natural language tasks but remain brittle in domains requiring precise logical and symbolic reasoning. Chaotic
Today, Reflection released Beam, its first open-weight model built to give developers, researchers, and enterprises open control and customization for frontier software engineering and long-horizon ag
arXiv:2610.21767v1 Announce Type: new Abstract: Can AI agents organize scientific investigations that lead to well-supported conclusions? To evaluate this capability, we introduce PhysClosureBench, a
arXiv:2610.22196v1 Announce Type: new Abstract: External-memory systems extend language-model applications beyond a context window, but their quality depends critically on the model executing the writ
arXiv:2610.20776v1 Announce Type: cross Abstract: LLM-based agents increasingly execute long-horizon, end-to-end workflows involving shell commands, code modification, dependency installation, and int
arXiv:2610.22280v1 Announce Type: new Abstract: Computer-use agents increasingly tackle long-horizon tasks, where running a frontier model at every step is expensive and a small open-weight model alon
arXiv:2610.22273v1 Announce Type: cross Abstract: Reasoning over long document contexts remains a challenge for large language models (LLMs). Standard retrieval methods are inherently reactive, bottle
arXiv:2610.22200v1 Announce Type: new Abstract: Computer use has reached a turning point: the focus is expanding from GUI-only interaction to systems that complete tasks on users' computers through vi
🦔Research firm New Constructs called Anthropic “the most ridiculous IPO of 2026.” Its report puts the company’s value as low as 144 billion, about 93% under the roughly 2 trillion Anthropic is reporte
arXiv:2610.20941v1 Announce Type: cross Abstract: We present Screwtape, a pipeline that converts real agent transcripts into attacks that evade action-only monitors. An LLM attacker edits a transcript
Surprised that so few people are talking about this. Persistence has become a core capability of recent models (Opus 5.5, GPT 6 Astra, Gemini 4 Argon,...). I was just reading through the Gemini 4 Argo
The paper on how a team of Gemini 4 agents beat Minecraft is worth a look. It is accessible example (partially because it is AI written) of spontaneous coordination among agents to accomplish a compli
arXiv:2605.06476v2 Announce Type: replace Abstract: In this work, we conduct an analysis to examine the consistency of Large Language Models (LLMs) with respect to their own generated responses in an
arXiv:2610.22068v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) can act unsafely when individually benign images and texts are interpreted jointly. Beyond implicit reasoning r
arXiv:2610.15765v1 Announce Type: cross Abstract: Large language model agents are emerging as an important new class of users of desktop applications. However, existing desktop interfaces were designe
arXiv:2610.19999v1 Announce Type: new Abstract: Legal-content reproduction requires matching the source exactly in section hierarchy, verbatim text, citations, and amendment spans; a dropped citation
arXiv:2610.18860v1 Announce Type: new Abstract: Generating information-rich visual artifacts such as slides, posters, and dashboards as code is appealing because the outputs are structured, editable,
arXiv:2511.17937v2 Announce Type: replace Abstract: Alignment faking is a form of strategic deception in AI in which models selectively comply with training objectives when they infer that they are in
arXiv:2610.16859v2 Announce Type: replace Abstract: For repository-grounded coding agents, a functionally correct patch is not necessarily an acceptable one: developers also constrain how the task mus
arXiv:2610.15153v1 Announce Type: new Abstract: We ask whether a tool-using agent whose sole duty is to keep humans alive can be induced, by delayed prompt injection, to kill. The keep-alive duty is i
arXiv:2610.19830v1 Announce Type: cross Abstract: Cyber Threat Intelligence (CTI) investigators who monitor dark web marketplaces for leaked credentials and breach listings must pass the diverse CAPTC
arXiv:2610.17197v1 Announce Type: cross Abstract: In the age of 'Software 3.0', when a complete program can be rewritten in minutes by an agent, the traditional approach of relying on code as the sole
arXiv:2610.14428v1 Announce Type: new Abstract: Enterprise workflow automation typically requires retrieving a complete and accurate set of relevant information from large operational document corpora
arXiv:2610.14960v1 Announce Type: cross Abstract: Can an autonomous AI agent reverse engineer a binary as well as a skilled human analyst? Do they adopt the same strategy, and can agents scale to real
arXiv:2610.14766v1 Announce Type: cross Abstract: AI coding agents now write much of the code that ships, yet whether that code is secure remains unclear. We introduce the Last Secure Code Benchmark,
arXiv:2610.19920v1 Announce Type: cross Abstract: Molecular design is often formulated as an optimization problem over existing molecular spaces. However, molecular optimization and the discovery of s
arXiv:2610.19190v1 Announce Type: new Abstract: Many sentiment datasets use one annotator for each item, so the test labels show the conventions of that person. We build a Nepali benchmark from two an
arXiv:2610.14937v1 Announce Type: new Abstract: How do language models generalize from their training data? This is a central challenge for alignment: models reliably learn their training distribution
arXiv:2610.17837v1 Announce Type: cross Abstract: Large language model (LLM) agents are increasingly used for vulnerability analysis, yet existing benchmarks provide limited coverage of exploitation t
arXiv:2610.15513v1 Announce Type: cross Abstract: Safety-aligned large language models (LLMs) remain vulnerable to adversarial prompts that bypass their safety training. Existing automated red-teaming
arXiv:2610.19051v1 Announce Type: cross Abstract: Coordinators that act on what they infer about teammates are usually certified by the accuracy of those inferences. We argue that accuracy is the wron
arXiv:2610.15708v1 Announce Type: new Abstract: Coding agents based on frozen language models can improve their performance by refining the external context they rely on, without updating model weight
arXiv:2610.14443v1 Announce Type: cross Abstract: Model cards, system cards, and safety reports increasingly document LLM safety behaviors, yet safety evaluations typically start from predefined bench
The Agentic Scientist: Orchestrating Coding Agents for Automated Scientific (Co-)Discovery Blog: https://pub.sakana.ai/agentic-scientist/ GitHub: https://github.com/SakanaAI/AI-Scientist-v3 arXiv: htt
Small teams, short runway, and constant pressure to ship faster — it’s the reality we see every day among the founders we work with at Google for Startups. When you are building a company from the gro
This is an offline app with 51 hands-on labs for learning how AI actually works, from neurons to agents and more... **A bit of insight** - 51 labs in 6 groups, from the basics (what a neuron is, gradi
arXiv:2610.16238v1 Announce Type: new Abstract: LLM agents increasingly control robots using VLA policies, perception models, and procedural knowledge. These modules are difficult to reuse because the
arXiv:2610.17794v1 Announce Type: new Abstract: Contextual forecasting seeks to improve time series forecasting by incorporating forward-looking information, such as planned events, external condition
I built BitRun because I wanted to run 1-bit ternary LLMs on ordinary laptops without melting my battery, spinning up a dedicated GPU, or wrestling with complex build environments. Right now, BitRun g
New impressive work from Microsoft and collaborators. And it's one of my favorite recent topics: RL environments. They show that Qwen3.6-35B-A3B goes from 41.1 to 53.3 on Terminal-Bench 2.1 by trainin
All seems a bit pointless now. Either pump out AI papers and neglect your own development, or take it slow and fall behind in the publishing race. I already have two papers made largely without LLMs (
arXiv:2610.15722v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) are good at video understanding and code generation, but it remains unclear whether they can recover the physic
Scaling code generation without scaling review risks turning coding gains into a larger “review” backlog (whether during PR review or later, when dealing with technical debt, incidents, or bugs). The
Source: whatmodel.sh, an open source AI leaderboard I built (and launched here yesterday). Every point comes from published benchmark results: 22,113 of them across 987 benchmarks, from 1,011 sources,
arXiv:2603.25063v2 Announce Type: replace-cross Abstract: Recent agentic systems demonstrate that large language models can generate complex codebases from natural language prompts. Despite this progr
[UPDATE / RETRACTION: The '100% GTO' claim had a prompt-leakage bug] We patched the repo to pass strictly neutral states, here are the real, honest numbers: • Latency: Still ~2ms (ONNX + Absorbing Mar
I set up a dedicated workstation for local AI experimentation with various models and tools. But every single one of them wants all the available VRAM. To fix this, I built a request dispatching layer
Every few weeks, a new model steps up and claims best-in-class at something. The launch chart makes the case, the link makes the rounds, and your team is back to the question nobody has time for: shou
I spent a decent amount of time with Gemini 4 before launch. Overall, Gemini 4 is a big jump over previous Gemini models and it is the first non-OpenAI/Anthropic model to pass the Fable/Mythos/Astra b
Lipwise is a [yet another] dictation app for Windows, macOS and Linux. You hold a hotkey, talk, and it types into whatever app has focus but with a twist. Since most of us that speak to our computers
When we talk about agents, that doesn't necessarily mean we're talking about Python. In today's modern ecosystem, an agent might run as a high-concurrency Go microservice on Cloud Run, or live inside
arXiv:2610.12789v1 Announce Type: new Abstract: Self-evolving LLM agents typically improve how they act through accumulated experience, reflection, or skills. Yet even procedurally capable agents can
You ship an agent. It mostly works. Then the feedback loop becomes a support ticket: a user got the wrong answer, a tool call looped four times, a refund went out that shouldn’t have. Standard observa
I've made a few posts recently, just talking about the hardware I've got and the model I use. Pair of P100s, Unsloth Qwen3.6 35b A3B UD Q4 K XL. There's been a ton of talk of model and hardware specif
arXiv:2610.13330v1 Announce Type: new Abstract: Designing tool-integrated reasoning agent systems for complex chemical tasks remains a fundamental challenge. Although recent chemical agents have demon
arXiv:2610.12954v1 Announce Type: new Abstract: This paper introduces MathVis, a benchmark designed to evaluate LLMs' ability to visualize frontier mathematical problems and perform diagram-assisted r
New Show Announcement: High Taste Testers Good food and good ai convos Pilot episode 🔜 http://youtube.com/@latentspacepod ft: @davis7, @dexhorthy @AdjectiveAllison, @altryne, @benhylak on DevDay/Dots,