From a full day of testing with the new gpt-4-turbo, a few things stand out: It's much less lazy and more willing to output large chunks of complete code. It seems better at reasoning. Plus, function calling with vision is going to be a huge unlock.
PROMPT ENGINEERING
-
Reverse Engineering System Instructions for REST API
By
–
The REST API documentation doesn't cover system instructions yet, so I had to reverse engineer how to use them from the Python client library – notes on that here https://
github.com/simonw/llm-gem
ini/issues/6#issuecomment-2046460319
… The JSON looks like this: -
GPT-4 Vision JSON parsing and prompt engineering techniques
By
–
Were you asking GPT-4 Vision for JSON and then parsing what came back – did you have to do anything special to avoid "Here is the JSON for that:" prefixes and suchlike?
-
Google Gemini 1.5 Pro Now Available in 180+ Countries
By
–
Google’s Gemini 1.5 Pro model is now available in 180+ countries via the Gemini API.
— Rowan Cheung (@rowancheung) 10 avril 2024
The company also introduced native audio understanding, system instructions, JSON mode, and more.
Try the 1.5 Pro for free with my tutorial below: https://t.co/zih5X28NjyGoogle’s Gemini 1.5 Pro model is now available in 180+ countries via the Gemini API. The company also introduced native audio understanding, system instructions, JSON mode, and more. Try the 1.5 Pro for free with my tutorial below:
-
Overcoming AI Output Token Limits Through Strategic Prompting
By
–
Yeah the maximum number of output tokens is an important factor – but you can overcome that by promoting it for more content and including the content it's already given you
-
Discovering Prompts Behind AI Model Research Paper
By
–
I'm trying to figure out what prompts they used. I think it's these – this is from a paper that their paper links to https://
github.com/lilakk/BooookS
core/tree/main/prompts
… -
Vendy: GPT Persuasive Vending Machine for Problem Solving
By
–
If you want to try to interact with a persuasive vending machine, here is Vendy, a little GPT I threw together. It knows that the solution to any problem is probably buying a refreshing lemonade, and is here to help. https://
chat.openai.com/g/g-LMszzSJYv-
vendy-the-friendly-vending-machine
… -
GPT-4 Vision for OCR: Practical Implementation Guide
By
–
I say "OCR" but it was actually GPT-4 Vision itself that I was using for that – my prompt was "Run OCR and return all of the text in this image, with newlines where appropriate"
-
Prompt Engineering Strategies for GPT and Claude Models
By
–
I've not tried fine-tuning any models at all – I've been focusing on figuring out how to best prompt the larger models like gpt-3.5-turbo or claude-3-haiku, mainly through figuring out the best examples to give them
-
Vision Models Function Calling Eliminates Redundant OCR Workflows
By
–
Fantastic – I've been looking forward to function calling for vision models Means I can ditch a nasty little hack I was running where I'd pipe input through Vision and tell it to run OCR, then pipe the result through Turbo to get structured data out of it