Hey @googleaidevs @AnthropicAI I have spent the last 5+ years making things look awesome in the terminal. You could have access to all the expertise for the price of a sponsorship, and help fund open source. #worthashot https://t.co/CeQFcEYXUk

My head-canon is some Google PM insisted on having an easy, streamlined credential. Every Google PM appears to wish the rest of Google didn't exist. Well, the monkey paw curled and they got their wish. You can indeed easily get a GEMINI_API_KEY. It won't help you, but it exists.

Again I'm not even sure I'm understanding the auth situation correctly, after having solved this. But if I'm right then the whole GEMINI_API_KEY situation is worse than useless.

So what worked for me is to run the right combination of gcloud auth login, gcloud auth application-default login, gcloud config set project and gcloud auth application-default set-quota-project commands. Oh, set GOOGLE_CLOUD_PROJECT, and of course unset GEMINI_API_KEY. Trivial.

I've seen thread after thread of people getting immediate rate limit errors on the Gemini API, with no resolutions. So is this the answer? For whatever it's worth, Gemini itself says so -- it could always be wrong about its own auth though, it's working from the same docs as us.

The Gemini API key also doesn't have any sort of default billing project. And with no billing project, you have no quota. And the free-tier rate limit for Gemini 2.5 Pro is 0. So correct me if I'm wrong, but doesn't this make the GEMINI_API_KEY worse than useless?

If you go to the Gemini console at https://t.co/WksKzRrleN, it will cheerfully help you create an API key. You can even select which gcloud project to use the key in. What the key doesn't specify is the _billing_ project.

Was keen to try Gemini Code, and I'm immediately reminded of the infamous "skill issue" tweet. I immediately got a rate limit error, because fool that I am, I tried to use GEMINI_API_KEY to authenticate, which is a total red herring.

Going live now! Join me for the first pilot stream on YouTube.

Now that Explosion is small again, I'm also missing a bit of interactivity in my work. Looking forward to seeing how this goes! I want to do the first pilot streams this Thursday 2pm CET (8am EST) and Friday at 10am CET on our YouTube channel.

Trying something new this year: I'm going to stream spaCy and other open-source development and maintenance twice a week! I hope this will give the community a better inside look while things are in progress. I'll also answer your questions and chat about NLP, OSS and more.

Writing a new talk on "What the history of the web can teach us about the future of AI" 🔮 Wanted to do this for so long because I think there are some great lessons & analogies here. I'll be presenting it for my keynote at @PyConWeb in Berlin on Jan 25 – hope to see you there!

New post: From PDFs to AI-ready structured data 📃✨ A deep dive into document processing, layout analysis and a modular workflow for building end-to-end document understanding and information extraction pipelines using PDFs, Word documents, scans and more.

The first version of my spaCy + Docling integration is here: 📚 process PDFs, Word documents & more 📝 structured text-based output via @spacy_io's Doc 🏷 layout spans for sections, headings etc. 🔮 apply NLP pipelines to PDFs ✂️ chunk your data for RAG

But looking at the national totals doesn't even make sense. If you're in a safe state, it's perfectly logical to symbolically abstain or vote third party if that's the message you want to send. And it's not like the election can be "stolen" by adding a bunch of Californian votes

The main reason the 20 million votes are "missing" is that the graph doesn't include in-progress counts from the West Coast. It's also zoomed-in instead of starting from zero. Graphs often lie like this.

My timeline is full of this zoomed-in graph of incomplete vote counts that "proves" the 2020 election was stolen because "20 million democratic votes came from nowhere". But only 7 US states actually vote for president. Here's how the counts for those states look.

Modal is awesome and a real game changer for serverless NLP 💚 In this tutorial, I show you how you can go from an idea and little data to a fully custom model using https://t.co/j3J6mQ9xJf and @modal_labs, no infrastructure or GPU setup required! https://t.co/YK44VQq1bf

Paris recap – merci for a great #dotAI2024 💛 https://t.co/SVhLokf8dg