Good news for Anthropic devs: Tomorrow we will be removing the tokens per day (TPD) limit and splitting tokens per minute (TPM) into separate input/output limits across tiers on the Anthropic API.

People have too inflated sense of what it means to "ask an AI" about something. The AI are language models trained basically by imitation on data from human labelers. Instead of the mysticism of "asking an AI", think of it more as "asking the average data labeler" on the

# HELP WANTED if you are early career and want to move to/plug in in SF: i am in need of a shadow! (aka intern/assistant/gopher. not employee of any of my companies but that can be an option if fit) - you'll help with everything I do - I'll answer your questions - show you

Price the _data product_, and let the user pay for AI - token usage based. If I want to spend 5$ to do some serious combing through my inbox after a vacation and only a few cents for my day to day email work, it works out. Current pricing models create inferior AI

Internal data access & LLM performance on business critical tasks is the next big unlock and it’s fascinating to see the different approaches at play here! Anthropic with MCP, OpenAI acquiring rockset and now Perplexity with carbon! Challenges to be figured out here include: -

Great to see reliability taking centre-stage! To extract business value from AI - reliability > capability. But for users adoption - threshold capability is critical. A highly reliable but super narrow capability system won’t get enough usage. Users will go back to non-AI

We tested a pre-release version of o3 and found that it frequently fabricates actions it never took, and then elaborately justifies these actions when confronted. We were surprised, so we dug deeper 🔎🧵(1/) x.com/OpenAI/status/…

This is THE problem with tool calling / composition. It’s extremely hard to guarantee that what was planned was in fact, what was done. And that the explanation of what was done is accurate. The smarter the LLM gets the better it gets at lying :) atleast Claude-3.5

PSA It’s a new era of ergonomics. The primary audience of your thing (product, service, library, …) is now an LLM, not a human. LLMs don’t like to navigate, they like to scrape. LLMs don’t like to see, they like to read. LLMs don’t like to click, they like to curl. Etc etc.

I actually find vibe coding is significantly easier on the backend Tests make it nearly impossible for the model to mess up. Code is relatively straightforward (crud) But I have not been able to get any model work well on complex UIs. By the 10th loop it’s pile of steaming

Trustworthy AI Agents bit.ly/cloudcast-935 Tanmai Gopal (Tanmai Gopal, CEO/Founder @HasuraQL) discusses the importance of reliability and trustworthiness for both generative and agentic AI. We discuss the pitfalls in existing data pipelines and how to enhance the results.

Yesterday: "We've had broken data authorization for years but it wasn't urgent" Today: "Everyone wants AI agents and suddenly we need this fixed immediately" This conversation with a Fortune 100 customer really hit home. For years at Hasura, we've been solving the distributed

And it turns out their teams are working 80 hour weeks despite that! Feels like a whole work vibe shift is just happening. Culture that Google/FB types set: Be solving problems at scale and have massage parlours and 5 star lunches. 💆‍♂️ The add on with AI companies: Be growing