ChatGPT is coming soon to the Azure OpenAI Service, which is now generally available, as we help customers apply the world’s most advanced AI models to their own business imperatives.
That feeling when you get a coding answer in ChatGPT, then Copilot autocompletes the same in VS Code - and it works! In a couple of years, kids will think that's how it's always been. Amazing times we live in!
Microsoft is looking for ML practitioners to learn how well ML tools work for you. Fill out a short survey and if you are selected, we will reach out to schedule an interview! There will be a gift for interview participants.
Do you ever think "what was that tweet about XYZ?". I indexed "interesting tweets" that I had saved and asked about prompt injection. GPT3.5 decided there is something called "
@goodside
's prompt injection" 😀
I code
@sphero
RVR with VS
@Code
remote and send it driving commands with
@blueswireless
. VS Code is definitely the easiest way to develop on headless RPi!
This weekend I learned that VS
@Code
can run on a
@Raspberry_Pi
, and that its extension packs make connecting to a headless RPi *super* simple. 🤯
👇Check out how to do it with the Remote Development extension pack () in
@JimBobBennett
's tutorial below:
Hello GPT-4o on Azure!
We're excited to make gpt-4o available to all Azure OpenAI customers for early access today: . Stayed tuned for news on gpt-4o and more at Microsoft Build next week!
Big congrats to our friends at
@OpenAI
on the launch!
The GPT-4 preview is now available in Azure OpenAI Service. Create solutions that can generate and understand text in multiple languages, craft contextually appropriate responses, and more. Get the details →
#AzureOpenAI
#GPT4
Introducing the Batch API: save costs and get higher rate limits on async tasks (such as summarization, translation, and image classification).
Just upload a file of bulk requests, receive results within 24 hours, and get 50% off API prices:
🦜🔗v0.0.44
⏲️ Timeout argument for OpenAI LLM - Sam Bean,
@YouSearchEngine
💙 Azure OpenAI LLM integration -
@gojira
👿 Hypothetical Document Embeddings, based on - h/t
@luyu_gao
For docs on Azure OpenAI, see
@humanloop
Hey this looks really awesome! Would love to have
@humanloop
integrate with Azure OpenAI to help our customers (). How can we help make that happen?
GPT-4 Turbo with Vision is now generally available in the API. Vision requests can now also use JSON mode and function calling.
Below are some great ways developers are building with vision. Drop yours in a reply 🧵
@jheitzeb
Hey Joe! The models, inference code, and GPU SKUs are mostly the same so performance should be roughly comparable. Azure has enterprise features like private networking, compliance certification, and privacy.
Excellent guide from
@HamelHusain
on evaluation driven development for LLM applications. So important and yet not nearly enough tokens talking about it
I often hear that evals are the most confusing part of creating LLM AI products. It's a shame b/c IMO, domain-specific evals are the most important part of an AI product!
I've written a detailed blog post with real examples on how to do this (1/3)
excited to announce Structured Outputs -- our newest feature in the api. model outputs will now reliably follow your exact json schemas, matching the parameters and types accurately.
schema reliability is now 100% with this feature enabled, up from 86% on our past gpt-4o model
Hi, it's me again - here to say you can still be early on
@runwayml
- We are continuing to assemble a very robust, full-feature video editor built entirely on the web. Smooth editing and hours of "traditional" rotoscoping done in seconds - the future is now 🥲
After 2 years, Practical Deep Learning for Coders v5 is finally ready! 🎊
This is a from-scratch rewrite of our most popular course. It has a focus on interactive explorations, & covers
@PyTorch
,
@huggingface
, DeBERTa, ConvNeXt,
@Gradio
& other goodies 🧵
@jeremyphoward
@github
Codespaces is excellent. How are you thinking about this relative to Jupyter (esp. notebooks in Codespaces)? Would love to hear your thoughts!
GitHub Copilot is amazing. It's only the first widely available product using OpenAI Codex models. There is amazing work happening in product and research around AI + code. I can't wait to see what comes next!
I've just released llamafile v0.8 which features LLaMA3, Mixtral 8x22b, and Grok support. It goes 25x faster than ollama at running LLaMA3 70B on CPU. My new tensor multiplication kernels let llamafile eval MoE models 2x faster than llama.cpp
Python in Excel ... now with Copilot! Excel is one of the world's most widely used collaborative canvases. People have been sharing, riffing, reviewing and publishing ideas using Excel for decades now. Now that we have Python in Excel as a foundation, we can build on top of it to
The
@OpenAI
Python library by
@StainlessAPI
@RattrayAlex
is done nicely. It really increases productivity with OpenAI and Azure OpenAI. It even makes it easier to use REST when you need to with the client helper methods for url, headers etc!
Introducing two more model updates:
• Updated GPT-4 Turbo preview, completing tasks like code generation more thoroughly and fixes a bug with UTF-8 encoding
• New GPT-3.5 Turbo next week, with input prices reduced by 50% and output prices reduced by 25%
Just wanted to sing some praise for Github Codespaces . It's not available to individuals yet (esp GPU VMs), but it is by far the easiest way I've seen to "just get a GPU in the cloud" - from one button on a Github repo to an open VS Code few seconds later
# On the "hallucination problem"
I always struggle a bit with I'm asked about the "hallucination problem" in LLMs. Because, in some sense, hallucination is all LLMs do. They are dream machines.
We direct their dreams with prompts. The prompts start the dream, and based on the
It turns out training Chat AI's take a lot of work. I'm finding some of these new chatbots are kind of argumentative & dig in their heels. I'm really proud to work with
@OpenAI
because they've really taken care to release an approachable experience!
This week, we showed how altering internal "features" in our AI, Claude, could change its behavior.
We found a feature that can make Claude focus intensely on the Golden Gate Bridge.
Now, for a limited time, you can chat with Golden Gate Claude:
Ray is a powerful ML framework, but with great power comes massive documentation. How can we make it more accessible?
Now, using
@LangChainAI
and Ray, we can build and deploy a doc search engine in about 100 lines of code -- with a self-hosted LLM! 1/n
1/ Can A.I. do our literature reviews for us? Stop everything and try , an amazing new tool that uses large language models to answer research questions via empirical research- in the video below I ask it "Does social media negatively impact mental health?"
We're pleased to announce the launch of OpenAI’s latest model, GPT-4o-2024-08-06, now available on Azure AI. New features enhance your AI experience on Azure and improve developer productivity through Structured Outputs:
Azure OpenAI is the fastest growing service in Azure's history with more than 4500 customers. Thanks for using Azure and OpenAI! (at 22:30 in this keynote)
Ever wondered how nondeterministic GPT-4 is even with greedy decoding (T=0)? I built a website that asks GPT-4 to draw a unicorn every hour and tracks if the results stay consistent over time (spoiler alert: they don't! 🦄).
Explore the findings:
Excited to share a little behind the scenes of the GPT Builder feature available inside the
@ChatGPTapp
👀
GPT Builder is actually itself a GPT using instructions and a custom action. This is useful to read if you are building GPTs:
It's a real revelation running mixtral on laptop with one llamafile download. Before ChatGPT, going long on LLMs & prompt engineering felt like an outlier. A short year later a large community has made it possible to run a really amazing model out of the box!
@karpathy
Congratulations! I have fond memories of waking up in the middle of the night to solve cs231n homework problems. Apparently I was doing numpy in my sleep. I can't wait to see what Eureka moments your new classes will trigger!
I don't know what I'm going to do with it but I just bought a
@blueswireless
Notecard! It took me a few minutes, I eventually connected the dots on the Lotus Notes reference 😃
📔 Thankful this evening for the notebook review experience in
@code
's
@GitHub
PR extension:
So very much nicer than raw JSON!
Are there any other features in this review experience you'd like to see (ex: being able to leave comments on multiple lines)?
@jheitzeb
Thanks for the opportunity to learn what builders are doing in Seattle with LLMs! It was really a lot of fun. Anyone I missed please reach out here or LinkedIn!
Cog is Docker for machine learning. It lets you package your machine learning model in a standard, production-ready container.
We’ve been using it internally for Replicate, and now it's ready to use for your own deployments.
At Microsoft, we’re expanding AI capabilities by training small language models to achieve the kind of enhanced reasoning and comprehension typically found only in much larger models.