actually I love this idea, would make it very easy for ai agents to know how to build better experiments.

Your 1 million token context window is lying to you. The bigger you make the prompt, the worse the model gets at using it. It reads the start, reads the end, and skims the middle. A longer prompt is not a better prompt. You pay more, wait longer, and usually get a worse answer. This is why every serious AI agent in 2026 runs context compaction. I broke down 11 of these techniques and the exact order I run them in. #short

What do you call it when someone pours their heart out without knowing they’re talking to a bot? Business as usual. AI is making it easier to imitate the messy little signals we use to decide whether someone is real, safe, caring, charming, or in trouble. Best case? A disappointing date with someone whose messages had a little too much help from ChatGPT de Bergerac. Worst case? A panicked call from someone you love, except the voice isn’t theirs. In this episode of AI Advice From Anywhere, Cassie Kozyrkov walks through three relationship-based AI scams worth knowing about: from awkward to criminal to the kind every family should prepare for before anything goes wrong. 📍 AI Advice From Anywhere: Watch to the end and see if you can guess the city before the reveal. Post the number of seconds it took you in the comments. (No spoilers.) #AAfA #DecisionIntelligence #AILeadership #FutureOfWork #ExecutiveDecisionMaking #AIAdoption ********************************** Invite Cassie to speak at your event: https://makecassietalk.com Learn directly from Cassie: https://decisiongptcourse.com/ Newsletter: https://decision.substack.com Intro 1:1s: https://intro.co/CassieKozyrkov

Everyone is talking about prompt engineering. But Rebecca Evanhoe thinks the real advantage is somewhere else: 👉 Data. As AI teams move away from traditional scripting, they're spending less time tweaking individual words and more time evaluating outcomes at scale. The challenge? You can't evaluate AI without quality data. The future isn't prompt-first. It's evaluation-first. And evaluation runs on data. #AI #LLM #PromptEngineering #EnterpriseAI #DataScience

A talk from AI Engineer Melbourne https://webdirections.org/ai-engineer/ .

As you know, Anthropic became the first AI company blacklisted by the US government. The story everyone shared was simple. They said no to surveillance and weapons, got cut off, and OpenAI swooped in for the deal. Here's the part that got skipped. Anthropic wasn't refusing to work with the government. They'd been inside classified systems since 2024. So had OpenAI. The actual fight was narrow. The Pentagon wanted two safeguards pulled from Claude: mass domestic surveillance and fully autonomous weapons. Anthropic drew the line right there, loudly and in public. And OpenAI never signed the exact deal Anthropic walked away from. A few days later they quietly tightened their own wording, much closer to where Anthropic had been pushing the whole time. So it was never one clean company saying no and another saying yes. Both were already in. They just handled the line very differently once it got explicit. There's a lot more to it. The lawsuits, the supply chain risk label, the sympathy wave that shot Claude to number one. I covered the full timeline in the video 👇 #short

kind of true that usually teacher even for human is to do the thing well. learning on your own is where you figure out the issues and how to recover.

today we are taking a deeper look at of the secret behind LLM pretraining which is synthetic data pipelines with Joel Niklaus machine learning engineer at Hugging Face! joel an his team ran 90 controlled experiments and burned over a trillion tokens to figure out what actually makes good pretraining data and in huggingface fashion provided all of their findings/artifacts openly! We’ll check the different structured format they used, the multiple ablations they ran and the counter intuitive outcomes they got out of the result! Will be a fun one (recorded too no worries about it)

today we are taking a deeper look at of the secret behind LLM pretraining which is synthetic data pipelines with Joel Niklaus machine learning engineer at Hugging Face! joel an his team ran 90 controlled experiments and burned over a trillion tokens to figure out what actually makes good pretraining data and in huggingface fashion provided all of their findings/artifacts openly! We’ll check the different structured format they used, the multiple ablations they ran and the counter intuitive outcomes they got out of the result! Will be a fun one (recorded too no worries about it)