OpenAI CEO's Testimony + Prompt Engineering Tips + AI News From This Week

Using AI

18-05-2023 • 53 mins

The implications of Sam Altman's testimony plus prompt hacking / prompt engineering tips. EU AI Act, Open Source OpenAI Models and much more.


Links

Watch this on Youtube


AI News this week Sam Altman’s abridged testimony (Twitter):

Senate sub-comittee hearing in full:

Sam Altman Testimony on TikTok

Genie AI Linkedin


Mentions:

Sparks of AGI Microsoft Paper

OpenAI open source model

EU AI Act


Crazy things this week

https://mealpractice.com/generate - Effortless meal planning with AI-generated recipes

Elon saying OpenAI wouldn’t exist without him.


Discussion Points

  • AI generated opening statement by Richard Blumenthal
  • Sam Altman (OpenAI), Gary Marcus (Uber, Robust.ai), Christina Montgomery (IBM)
  • Senate is aware they don’t understand this enough to legislate on it
  • OpenAI is making significant efforts towards safety
  • OpenAI advocates for a committee to regulate AI development for large companies
  • Election fraud seems to be a major concern for OpenAI
  • New, better jobs will be created according to OpenAI


Prompt hacking tips

  1. Be succint and specific!

  2. Prime the language model by providing a persona for it.

  3. Asking the model to think through its response step-by-step “Let’s think step-by-step”

  4. Providing clear context: Offer concise background information for guidance. Use Cases: Definition explanations, historical event summaries, concept descriptions.

  5. Specifying the output format: Indicate desired answer structure explicitly. Use Cases: Generating lists, step-by-step instructions, summarizing long texts.

  6. Using explicit instructions: Request specific detail or critical thinking. Use Cases: Debating pros and cons, analyzing biases, evaluating arguments.

  7. Redundancy and rephrasing: Reinforce information by reiterating questions. Use Cases: Clarifying ambiguous topics, extracting specific details, verifying facts.

  8. Temperature and token settings: Adjust randomness and output length. Use Cases: Creative writing, focused summaries, generating multiple response variations.

  9. Iterative refinement: Refine prompts based on previous responses. Use Cases: Troubleshooting, problem-solving, narrowing down complex topics.

  10. Prompt engineering: Craft effective prompts using templates or examples. Use Cases: Analogies, translating complex topics into simple explanations, generating structured responses.

  11. Advanced techniques like asking the language model to think step by step

Shout out to DeepLearning.AI’s free prompt engineering course: https://www.deeplearning.ai/short-courses/chatgpt-prompt-engineering-for-developers/


Links to us on other podcast platforms

Spotify - https://open.spotify.com/show/5Lbw5IlX0z22rqzQheZ5My
Apple Podcasts - https://podcasts.apple.com/us/podcast/using-ai/id1683562982
Google Podcasts - https://podcasts.google.com/feed/aHR0cHM6Ly9hbmNob3IuZm0vcy9kZjg0ODk2Yy9wb2RjYXN0L3Jzcw
Amazon Music - https://music.amazon.co.uk/podcasts/d68cc64d-9854-43b2-b72c-c54486f0cb06/using-ai
Castbox - https://castbox.fm/channel/id5409720
Overcast - https://overcast.fm/itunes1683562982/using-ai
Pocket Casts - https://pca.st/8q41hgrb
Radio Public - https://radiopublic.com/using-ai-GMvYqZ
Stitcher - https://www.stitcher.com/show/1066264
RSS - https://anchor.fm/s/df84896c/podcast/rss

--- Send in a voice message: https://podcasters.spotify.com/pod/show/using-ai/message

You Might Like

Darknet Diaries
Darknet Diaries
Jack Rhysider
Fortnite Emotes
Fortnite Emotes
Lawrence Hopkinson
Hard Fork
Hard Fork
The New York Times
Double Tap
Double Tap
Accessible Media Inc.
Waveform: The MKBHD Podcast
Waveform: The MKBHD Podcast
Vox Media Podcast Network
The Vergecast
The Vergecast
The Verge
Acquired
Acquired
Ben Gilbert and David Rosenthal
TechStuff
TechStuff
iHeartPodcasts
This Week in Retro
This Week in Retro
Neil from RMCretro - The Cave, Chris from 005 AGIMA and Dave
Hacked
Hacked
Hacked