Search: “reasoning”

Long-running models can solve hard open-ended problems, but their persistence can create safety risks that shorter-horizon evaluations miss. We’re sharing what we learned from studying a long-running

Extracting information from millions of documents at scale used to take an insane number of human hours. Even with recent OCR + document AI tech, humans would still have to spend a lot of time careful

Grok 4.5 is worth trying Grok 4.5 just ranked #2 on the FrontierSWE leaderboard, outperforming Claude Opus 4.8 and GPT-5.5 It delivers top-tier coding performance with exceptional speed, strong token

Note the current expectations are still around test time compute/more tokens for a given task This is not the case Tokens per task will now drop even as quality improves Cost per intelligence equivale

I want to have this in writing so I can refer back to this tweet before it inevitably becomes a consensus view on twitter and say 'I told you so!' - The mass majority of researchers and academics in d

The Anglo-Scottish Enlightenment – the real antidote to Rousseau and Voltaire The French Enlightenment and the Anglo-Scottish Enlightenment happened simultaneously, in the same century, reading the sa

GPT-5.6 Sol achieved a breakthrough by becoming the first verified frontier AI model to surpass performance on ARC-AGI-3, scoring 7.8% and setting a new state-of-the-art benchmark. ARC-AGI (Abstractio

I wonder how this could be used to study creative thinking and the generation of new ideas. If I understand correctly, the J-space is some sort of internal workspace where the model holds concepts to

Check out all the amazing work from our @SimonsFdn Collaboration on the Physics of Learning and Neural Computation (https://www.physicsoflearning.org/) presented at the main meeting of @ICMLconf #ICML

The take that frontier models aren't ready for use in medicine is dead wrong. I've been using Opus 4.x in my clinical workflow every day for 6 months. I'm a deep sub-sub-specialist in dermatology - th

In this episode, @OpenAI Chief Research Officer @markchen90 joins @allenpark to flambé shrimp, cook Korean stew, and chat about being at the frontier of AI research: why scaling laws and pre-training

New research from Meta. Building synthetic training data has stayed a fixed pipeline that you hand-tune and then freeze. Autodata casts an AI agent as a data scientist that builds training and evaluat

have been thinking a bunch about model routing and related things current thoughts here, would love feedback: 1/ there is a difference between 'model routing' and 'model council' 'model routing' = rou

The next generation of inference needs purpose-built infrastructure. Together AI and 5C are deploying NVIDIA GB300 NVL72 systems with high-density compute, advanced cooling, and AI-optimized storage f

Token costs are why there will be no saas apocalypse / good dev tools are cached intelligence for agents! The popular theory goes: agents can write code, so they'll just rebuild every tool from scratc

We've made a breakthrough in self-evolving AI scientists moving from 'search' to 'principled discovery': Scientific discovery requires that the search space itself changes, and an AI scientist must pe

Super excited to announce seven new world-class MAI models today. They represent what we consider a new era in AI designed to keep you in control and on the frontier. First is our text foundation mode

Just saving this here to document a story and as a self reflection on whether AI is really making me more productive Yesterday morning I found a way to complete the new HVM approach, that is much fast

I’m going to let you in on a secret. I made a mistake once.* In August 2024.* It’s true. I actually got something wrong. I predicted that there would be a collapse of the AI bubble (which in my judgem

this is so funny, training opus 4.7 on business skills makes it misaligned and dishonest 😭 Learnings from testing Claude Opus 4.8: > Much worse than Opus 4.7 and GPT 5.5 on Vending Bench > More aligne

If we had done everything I suggested in my 2020 arXiv article “The Next Decade in AI”, we might actually have reached AGI by now. In the last three years, after a detour driven by the false promise o

// Language Models Need Sleep // Let your agents 'sleep', folks. On a serious note, this is a fascinating paper on getting the most from long-horizon agents. Here is the problem with agents today: Att

Ralph Loops are powerful, but wrapping a naive loop in a shell script is a total token burner in production. Pinecone Principal Engineer Jen Hamon breaks down why standard loops collapse: ❌ The Bug: P

Often overlooked is the Frankfurt School’s central role, alongside French postmodernists, in forging Critical Theory and exporting it to America, where it evolved into one of the most influential (des

arXiv:2605.09852v1 Announce Type: new Abstract: Machine learning algorithms are being used in high-stakes decisions, including those in criminal justice, healthcare, credit, and employment. The resear

// LLMs Improving LLMs // Interesting progress the past of couple of weeks around self-improving AI agents. If autoresearch was interesting, you will like this read. (bookmark it) We've been hand-tuni

// The Memory Curse in LLM Agents // (bookmark it) Long histories apparently degrades agents as they become increasingly history-following and risk-minimizing. Across 7 LLMs and 4 social dilemma games

This post likely discusses a cognitive or computational principle where one should primarily focus on horizontal (broad, general, or foundational) aspects of a problem or system, then use inference or

Saw this and thought 'yes! ChatGPT voice mode is going to stop acting like a two-year-model' but that upgrade hasn't shipped just yet Introducing GPT-Realtime-2 in the API: our most intelligent voice

// HeavySkill // One of the cleaner takes on agentic harness design I've read. They argue that what actually drives agent harness performance is not the orchestration code. It's a single inner skill:

NEW paper from Microsoft Research. Nice study on long-horizon agent generalization. (bookmark it) The team runs a study where the only variable is task horizon length. They use the same decision rules

These folks are trying to ban open source. They're looking to take away your freedom to choose. They're also looking to take away the rights of businesses like Cursor to fine tune and make their produ

We’re excited to introduce KAME: Tandem Architecture for Enhancing Knowledge in Real-Time Speech-to-Speech Conversational AI, accepted at #ICASSP2026! 🐢 Blog https://pub.sakana.ai/kame/ Paper https://

// From Skill Text to Skill Structure // One of the more practical skill papers I've seen this month. SKILL.md files entangle invocation interface, execution flow, and tool/resource side effects in on

LOCAL AI MODELS ARE CATCHING UP TO FRONTIER MODELS WAY FASTER THAN ANYONE EXPECTED this guy ran qwen 3.6 27B locally on a base macbook pro M4 with 24GB of memory quantized and stripped of safety guard

Should we care about AI happiness? In our new research, we find evidence of functional AI wellbeing across several independent measures. We find which AI models are happiest, how to make them happier,

Gary Marcus critiques the marketing of large language model autocomplete systems as 'artificial intelligence,' arguing this terminology misrepresents their actual capabilities. He extends this critici

OpenAI's core principles emphasize democratizing access to AI technology, empowering individuals and organizations through AI tools, promoting universal prosperity across society, building resilient A

Great paper on improving proactive agents. (bookmark it) Proactive agents act before you do. But how do you evaluate something that's supposed to anticipate needs you haven't expressed? This work intr

Confidently wrong is worse than just wrong. When an AI searches one giant knowledge base for everything, it finds something—and serves it up like it's the answer. That's not a model problem. That's a

there are early signs of 5.5 being a competent ai research partner. several researchers let 5.5 run variations of experiments overnight given only a high level algorithmic idea, wake up to find comple

Unplugging completely! No WiFi and zero notifications. A great way to get deep focus on a project. Here is a walkthrough showing how to run Gemma 4 (26B A4B) fully offline with LM Studio & OpenCode to

For real agentic workloads (North), short-context calibration wasn't enough. We calibrated AWQ on long internal agentic traces (up to 64k tokens) and added token masking in llm-compressor to exclude r

Let me say this clearly: LLMs cannot feel emotions. Emotions are evolutionary mechanisms. They push us to avoid danger or approach what is beneficial. We experience emotions because we are alive, and

That's from this ChatGPT conversation - https://chatgpt.com/share/69e90ba5-7760-83e8-9e09-f06e25a90930 - my prompt was 'Do a where's Waldo style image but it's where is the raccoon holding a ham radio

This post highlights the Qwen 3.6-27B language model's image generation capabilities, noting that despite its relatively compact 16.8GB file size, it produces high-quality creative outputs like the ex