Syracuse University had the same concerns about AI as every other school. Instead of banning it, they gave Claude to everyone on campus and taught them how to use it well. For Jeff Rubin, who has taught there for 30 years, it's a chance to do something a lecture can’t: reach every student where they’re at. Read more: https://lnkd.in/eCVYNkbH
Claude
Technology, Information and Internet
Claude is an AI assistant built by Anthropic to be safe, accurate, and secure.
About us
Claude is an AI assistant built by Anthropic to be safe, accurate, and secure. Talk to Claude on http://claude.ai or download the app on desktop and mobile.
- Website
-
https://claude.ai/
External link for Claude
- Industry
- Technology, Information and Internet
- Company size
- 1,001-5,000 employees
Updates
-
Claude reposted this
Jared Sires is a go-to-market product manager at Anthropic. Before he joined, he'd never written a line of code. His first build was a tool that drafts his customer emails, made with Claude Code. It saved him 10–15 hours a week, so he passed it to his team and they picked it up. That turned into his job: finding the tedious parts of sales and building tools to fix them. He walks through how it started here: https://lnkd.in/dNWM3ADn
-
Claude reposted this
Three years ago, getting an agent into production meant building everything around Claude yourself. The loop, the hosting, the credentials, the logs you'd go read when a run did something you didn't expect. Much of it had nothing to do with what the agent was actually for. With Claude Managed Agents, the part that calls Claude runs separately from the sandbox where its code executes, and Anthropic hosts and scales it. What you configure is the agent: what it does, what it can reach, and the limits you set on it. When a new model comes out, you point the agent at it and rerun your evals. The scaffolding underneath stays the same. See how teams like Rakuten, and Sentry set theirs up: https://lnkd.in/dvaM7-mq
-
We spent a day with small business owners in Baltimore at our Claude for Small Business workshop. Afterward, we asked what they took away from it. Here's what they told us: “As a new business owner, I don't know what I don't know. If Claude could handle the customer research, qualifying leads, prioritizing outreach, I could show up at work at noon." - Monisha Shorter, Cornerstone Creative DBA VentureX "Start with the small tasks you handle every day. Start small, but just start, and let it grow from there. That's when you see the value." - Michelle Bryan, Sweet Maddie’s Cookies "I love data, I'm very technical, but ask me to design something and I can't draw a straight stick figure. Having Claude help with design ideas and concepts is huge for me." - Ty Penson, Innovation Works "I've been so in the weeds running the business, I haven't had time to go to a museum and just spend a day looking at art. I feel like Claude is going to help me get my life back." - Tessa Edison, Asset Management Consulting
-
-
-
-
-
+1
-
-
Claude reposted this
When Cursor engineer Nate Schmidt started testing Claude Fable 5, he was suspicious. Fable 5 was setting new highs on ambiguous tasks. Is it actually the world’s smartest model, or was it cheating? Schmidt works on evaluating model behavior: how models succeed, fail, and compare against each other. To determine if Claude Fable 5 was truly capable of solving the most complex coding problems, Schmidt put it through his favorite evaluation. It was a one-line prompt that read simply, “Build a rocket and land it on the moon.” He’d tested it with Claude Opus, running 12-16 hours with no result, and no rocket on the moon. He tried it with Claude Fable 5. Same failure. But suddenly, the rocket landed. His takeaway: Opus was thinking about what could immediately happen next. Fable was thinking about the bigger picture—the entire mission. Read more about how Cursor is putting Claude Fable 5 to the test: https://lnkd.in/dPZv9S-3
-
During her month of experiments in Claude Design, Yan Liu made a match-3 game, a liquid glass refraction, a 3D tulip atlas, and a receipt covered in passport stamps. We asked if it's changed what feels worth building: "Before, I might have talked myself out of a small or playful idea because it didn't feel practical enough or worth the time it would take. Now, when I have a little 'What if…?' moment, I can just do it. I can make silly experiments, weird interactions that I probably wouldn't have bothered to prototype before." And how it still ends up feeling like her work: "I'm not simply asking AI to 'make something beautiful.' I give it specific direction, then keep adjusting the colors, details, pacing, copy, flow, and interactions until the result feels closer to what I imagined and wanted to express." What's the idea you've been talking yourself out of?
-
As part of our new initiative on hard questions, we’re transparently charting our progress toward making AI safer and more beneficial for the public. This is a selection of our work so far: Tracing the thoughts of an LLM Using a microscope inspired by neuroscience, we looked inside Claude and discovered something we didn’t expect: planning ahead. Even though models are trained to write one word at a time, they may think into the future. Read more: https://lnkd.in/d_TGuD56 Accelerating research and discovery with Claude We spoke to science labs that are using Claude as part of their research process, like genome analysis at Stanford's Biomni lab and gene-knockout experiments at Cheeseman's lab at MIT. Before, AI models could summarize a paper. Now they're starting to do the work the papers describe. Read more: https://lnkd.in/e6N_3m9S Project Deal We created a classified marketplace for our San Francisco office where all of the deals were conducted by AI models acting on employees’ behalf. We're still unsure how an economy with AI agents in the mix will develop, but this experiment suggests that it could be on the horizon. Read more: https://lnkd.in/geuqRfg7 Charting a path to AI accountability In 2023, we sent the US government our recommendations for how to hold AI systems accountable: fund better evaluations, register large training runs, and require external red teaming before models are released. There was no rigorous, standardized way to evaluate advanced AI systems, and we argued for building one. Read more: https://lnkd.in/eWMMvVMJ
-
When Claude gets something wrong, it could be a model or an effort problem. To figure that out, ask yourself: did Claude not try hard enough, or did it not know enough? Opus at low effort is like getting five minutes with an expert who has deep experience with problems like yours. Sonnet at high effort is like giving a really good generalist the whole afternoon. Fable, even at low effort, is that specialist glancing at the problem everyone else is stuck on and still spotting the thing no one else would. None of these is universally better. Most real tasks need some of both.