ChatGPT Gets Its "Wolfram Superpowers"!
chatgpt-pluginswolfram-alphawolfram-languagetool-usecomputational-knowledge
Abstraction: ChatGPT plugin connecting to Wolfram Alpha and Language
Key points:
- Mar 23, 2023: Stephen Wolfram announces the Wolfram plugin for ChatGPT (via OpenAI's plugin launch), giving the LLM "computational superpowers" ~2.5 months after his January proposal.
- Argues an LLM alone can't do reliable nontrivial computation or produce correct data; the plugin lets ChatGPT call out for actual computation, curated real-time data, and original visualizations.
- Plugin has two entry points: Wolfram|Alpha (forgiving natural-language input, via a special "for LLMs" API endpoint) and Wolfram Language (precise, more powerful, but must be exactly right). ChatGPT can re-run code and self-correct on errors.
- Lets users "factify" ChatGPT output and see the exact query sent / raw response ("Used Wolfram" box) to distinguish facts from hallucination.
- Wolfram frames prompt engineering as "animal wrangling" / communicating with an "alien intelligence" (e.g., all-caps insistence on camelCase); positions Wolfram Language as the ideal human-AI collaboration language.
Connections: Stephen Wolfram · Chatgpt · Openai · Wolfram Alpha · Wolfram Language · Tool Augmented LLMS · Prompt Engineering · Large Language Models
Source: https://writings.stephenwolfram.com/2023/03/chatgpt-gets-its-wolfram-superpowers/