Staff Engineer @ Google + Recommendation Systems Professor @ Columbia University
New York, USA
Joined November 2017
- Tweets406
- Following98
- Followers763
- Likes75
I needed a document from Paris, it drafted a letter in French using other documents it found on my computer and brought me to payment page that prints and mails it.
OpenAIs new embedding seems to really prefer earlier text. Cos(emb(t1), emb(t1+t2)) > Cos(emb(t1), emb(t2+t1)) ~97% of the time of the 1000 examples I tried. Interesting retrieval consideration
With an A100 you can already get this down to ~250ms for 1 iteration
Introducing SDXL Turbo: A real-time text-to-image generation model.
SDXL Turbo achieves state-of-the-art performance with a new distillation technology, enabling single-step image generation with unprecedented quality, reducing the required step count from 50 to just one.
The code, research paper, and weights for non-commercial use are now available on our website.
You can test SDXL Turbo on Stability AI’s image editing platform @Clipdropapp, with a beta demonstration of the real-time text-to-image generation capabilities.
Learn more: stability.ai/news/stability-…
I wonder if sampling of these tokens is the same as others? I’d imagine you want to be greedier with tools so it hallucinates less?
It emits special words, e.g. <|BROWSE|> etc. When the code "above" the LLM detects these words it captures the output that follows, sends it off to a tool, comes back with the result and continues the generation. How does the LLM know to emit these special words? Finetuning datasets "teach" it how and when to browse, by example. And/or the instructions for tool use can also be automatically placed in the context window (in the “system message”).
What errors/mistakes/hallucinations happen 1 in 100, 1000, 10k and so on? Going to be a whole net set of outages over the next couple years
From my testing, 3.5 gets very confused if the function has a bug in it. Example of write/read to file, read file return random contents. it'll keep trying. 3.5 & 4 are convinced that it needs to re-write instead of re-read. It trusts us too much.
GPT-4 and GPT-3.5 Turbo models in the API now support calling your custom functions, allowing the model to use tools you design for it. Also — reduced pricing & new model versions (including 16k context for 3.5 Turbo): openai.com/blog/function-cal…
Had Sydney/#ChatGPT/#gpt4 create a #MarchMadness bracket: @ZagMBB you are the AI's favorite, good luck. Included are reasons why it thinks Gonzaga will beat @UA_Athletics. CC @MarchMadnessMBB
play.ncaa.com/mens-bracket-c…
I use TikTok and IG almost exclusively for food and parenting content. My feeds are usually pretty identical between the two. This week I’m getting a ton of pro TT anti-Congress content even as I continue to scroll past them.
Does @OpenAI or the users pay Wolfram?
Who is paying for these plugins? Will OpenAI add some token system? A more interconnected “App Store”? The AI chooses which plug-in to use if multiple options are given?
2.5 months later...in collaboration with @openAI , ChatGPT gets its "Wolfram superpowers"!
writings.stephenwolfram.com/…
ALT Diagram showing January cover for bringing computational knowledge superpowers to chatGPT then an arrow to the new Wolfram plugin for ChatGPT over an image of the plugin working inside ChatGPT for an example of Jupiter's moons complete with accurate data and visualizations.
Had Sydney/#ChatGPT/#gpt4 create a #MarchMadness bracket: @ZagMBB you are the AI's favorite, good luck. Included are reasons why it thinks Gonzaga will beat @UA_Athletics. CC @MarchMadnessMBB
play.ncaa.com/mens-bracket-c…