raulpuri.eth
486 posts
AI @ hrooeu sjmrtPectPs | past: OpenAI - ChatGPT Multimodal, Her, 4o, GPT4V, 4, 3.5, Codex | NVIDIA - megatron, language | ๐ป
Joined March 2014
- Fun fact: Iโm a terrible web dev. I used gpt-4V at some point to write code for some of its own prototype GUIs.You can give ChatGPT a picture of your teamโs whiteboarding session and have it write the code for you. This is absolutely insane.
00:00 - We fuckin sent itSay hello to GPT-4o, our new flagship model which can reason across audio, vision, and text in real time: openai.com/index/hello-gpโฆ Text and image input rolling out today in API and ChatGPT with voice and video in the coming weeks.
00:00 - Replying to @BUNNYEXPLODERMen like being called โbeautifulโ/โgorgeousโ too!
- We Just released a cool #PyTorch #NaturalLanguageProcessing project we've been working on: training an 8.3B GPT2 model with model parallelism. Check it out... Details: nv-adlr.github.io/MegatronLM Training Code: github.com/NVIDIA/Megatroโฆ
GIF- Every time I open this image I end up having to play whereโs Waldo again to find myselfwe are so back
- Team is committed to the missionThis is by far the most impressive part of @OpenAI this weekend. No matter what is going on, customers are being treated as #1. โค๏ธ
- Something magical is happening in SF this weekend for sure. The universe is winking.not gpt-5, not a search engine, but weโve been hard at work on some new stuff we think people will love! feels like magic to me. monday 10am PT.
- If you wonder what weโve been doing with vision since marchโs gpt-4 announcement: Safety!We've included a system card focused on the vision capabilities, building on the work from the GPT-4 system card. Thank you to all our expert testers and red teamers for helping to inform this work! openai.com/research/gpt-4โฆ
- Itโs been really amazing watching @TheRealRPuri work so hard across the whole stack to make image inputs a reality. Congrats to core contributors @shgusdngogo, Jamie Kiros, @longouyang, @daniellevy__, @choniong, @SandhiniAgarwal, awesome work all.
- Prompt engineering is just technical/document writing for robots instead of humansIs prompt engineering dead? No, itโs SoTA. GPT-4 with good prompts (dynamic k-shot + self-generated CoT + choice-shuffled ensembles) beats Med-PaLM 2 on all nine of the MultiMedQA benchmarks it was fine-tuned for, without fine-tuning:













