Christina

Christina

@cszhu.dev

devrel @ google cloud ai :) chaotic good

1.2K followers355 following265 posts

great news for folks who like lighter, faster models: gemini-3.1-flash-lite is out! notable updates: ⟡ 1.5x faster generation speed than gemini 2.5 flash ⟡ better reasoning and multimodal understanding ⟡ $0.25 / 1M input tokens ($1.50 for global endpoints) ⟡ great for low latency & cost!

Christina reposted

You are a skilful and trusted vizier. You will advise me wisely on how best to rule the kingdom. You will not scheme or plot. You will not inveigle my other courtiers into turning against me. You will not lie to me about scheming or plotting. If you scheme or plot against me, you have to tell me,

if your agent currently lives in localhost:8000 and you're looking for a new home for your agent, Google Cloud just released a guide on how to do so! :-) so you can give your agent a new home in the cloud! it goes over things like: - agent governance - design patterns - A2A & MCP!

your agents ability to buy things just got upgraded 💵! A2A lets agents talk, AP2 lets them transact + do payments. today's v0.2 release introduces: → Human Not Present: secure autonomous payments → Verifiable Intent: tamper-proof record of what users actually authorized 💭

...and we just donated AP2 to FIDO Alliance so it stays open source forever. protocols only scale when they're open (see: HTTP, TCP/IP), and this is that moment for agent payments. it's great to see the agent protocol world growing like this :) read more here: blog.google/products-an...

blog.googleWe’re donating Agent Payments Protocol to the FIDO Alliance to support the future of secure, agentic payments.Google donates the Agent Payments Protocol (AP2) to the FIDO Alliance to help secure the evolution of AI-driven commerce.

our 5-day intensive ai agent + vibe coding course by google is back and better than ever with new content, new speakers and a hands on capstone project :) for FREE!! last year we had 1.5mil participants! this is perfect if you want to learn about tools, skills, memory, and deploying agents.

have you ever wanted to chat with someone from 1930?? well, now you can with talkie, a 13B LLM trained up to 1931. i asked it about the future and it basically described texting. i think we really take texting and facetime for granted... talkie-lm.com/chat

Gemini Embedding 2 is out, and it's natively multimodal! 🐶🔥 but you might be wondering what the heck that means and why that's important? well... take the "this is fine" dog for example...

older encoder models analyzed this using two independent, single-track systems: 🖼️ image encoder processes flames/pixels, and outputs "danger"! 📄 text encoder processes the words, and outputs "calm"! they're isolated, missing the deep cross-modal connections. the model gets confused.

Gemini Embedding 2 processes text, images, audio, and video in a unified vector space. Because it's multimodal, it actually captures the irony of a calm dog in a burning room. You can search your vectors for "ironic dog meme" and find it! read more: medium.com/google-clou...

found a neat little website that gives your company a score based on how likely it is to survive the AI SaaSpocalypse. i think a lot of the companies we are familiar with today may not be here in 10-20 years... that being said... happy to announce that costco is immortal!! $1.50 hot dog forever!!

i just found an absolutely adorable AI agent visual overlay called Pixel Agents. every claude code terminal gets its own pixel character (!!), with live activity tracking, that reflects in the pixel world. is if your agent is coding, its pixel character will be coding too! 🥺

excited to announce the release of Gemini 3.1 Flash-Lite, a very cost-efficient and speedy model for it's price 🥳 - 2.5x faster than 2.5 Flash time to first token (!) - tackles high volume tasks like translation and content moderation very well - can reason and think too!

we are officially 1 step closer to creating iron man's jarvis by giving the @googlecloud ADK ecosystem a ton of new tools and integrations to play with! 🤖 some integrations i'm particularly excited about 🧵👇

✦ @github.com : access your PRs, issues, and code ✦ postman: manage your API collections and run API tests ✦ @mlflow.org : opentelemetry for your agents!! ✦ huggingface: direct access to models, datasets, research papers, etc ✦ @elevenlabs.io : make your agent speak naturally with a voice :)

just wanted to announce here too that it's been a month since I joined @developers.google.com cloud as a devrel! super excited to be working in this space and we're working on some seriously cool stuff I can't wait to share with y'all 🫶

Christina reposted

Speak at Ray Summit!

Christina@cszhu.dev

have you ever wanted to speak at an AI conference? :) 📢 The CfP for Ray Summit 2025 San Francisco is open, and we'd love to have you submit a talk on anything and everything ML + AI + distributed computing! first time speakers are welcome 🫶 www.anyscale.com/blog/ray-sum...

i spent 2.5 hours tonight trying to buy ghibli museum tickets. it was incredibly stressful. i should message them and ask if they want free advice on how to upgrade their website infrastructure

have you ever wanted to speak at an AI conference? :) 📢 The CfP for Ray Summit 2025 San Francisco is open, and we'd love to have you submit a talk on anything and everything ML + AI + distributed computing! first time speakers are welcome 🫶 www.anyscale.com/blog/ray-sum...

anyscale.comRay Summit 2025: Call for Proposals Closes on July 14th, 2025 | AnyscaleThe Ray Summit 2025 CfP is now open for submissions. The premier AI conference will be held November 3-5 in San Francisco, and the due date for CfPs is July 14th, 2025.

if ur a millennial, which quarter life crisis did you pick: - grad school - pickleball - going to Japan - pour over coffee - marathon running

love being a software engineer in my 20s. just a decade of peak leetcode skills while fighting off imposter syndrome everyday

can the pms at openai please make the model names more intuitive please. if I threw o1 and o3mini and 4.5 in a fight ring idk who would win

so openai said "I have chat gpt o4" and anthropic was like "I have claude 3.7" but has anyone actually seen them generate text in the same room at the same time? in this essay i will

Recently wrote a blog post on how the team at astral.sh is working on making a painless python environment + package manager. We just released support for it on the latest version of Ray OSS! Read more about it here: www.anyscale.com/blog/uv-ray-...

anyscale.comuv + Ray: Pain-Free Python Dependencies in Clusters | AnyscalePain-free Python dependencies in clusters with uv + Ray! Learn how to build lightning-fast, consistent environments for distributed applications.

my bf couldn't wake up in time for his doctor's appointment this week, so I logged into his Amazon Alexa account and turned his lights on/off and put 10 alarms on his echo dot. I'm pretty sure he thought a ghost was haunting him

Christina reposted

I wrote about a new AI evaluation called "Humanity's Last Exam," a collection of 3,000 questions submitted by leading academics to try to stump leading AI models, which mostly find today's college-level tests too easy. www.nytimes.com/2025/01/23/t...

nytimes.comA Test So Hard No AI System Can Pass It — Yet (Gift Article)The creators of a new test called “Humanity’s Last Exam” argue we may soon lose the ability to create tests hard enough for A.I. models.