Using AI

This week we will delve into the fascinating crossroads of AI and the future of work!

I’m joined by ML Research Scientist Alex Papadopoulos Korfiatis and AI Startup Founder Rafie Faruq.

Links:

MacWhisper: ⁠https://goodsnooze.gumroad.com/l/macwhisper⁠

Whisper Transcription by OpenAI: ⁠https://openai.com/research/whisper⁠

Hinton leaves Google: ⁠https://www.theguardian.com/technology/2023/may/02/geoffrey-hinton-godfather-of-ai-quits-google-warns-dangers-of-machine-learning⁠

Google leaked document about open source LLMs: ⁠https://www.semianalysis.com/p/google-we-have-no-moat-and-neither⁠

Labor Market Impact Potential of Large Language Models: ⁠https://imgur.com/a/qBDSYRv⁠ (Full paper: ⁠https://arxiv.org/pdf/2304.06123.pdf⁠)

IBM to freeze hiring as CEO expects AI to replace 7,800 jobs: ⁠https://www.aljazeera.com/economy/2023/5/3/ibm-to-freeze-hiring-as-ceo-expects-ai-to-replace-7800-jobs⁠

AI research chatbot which interviews people to learn about their views on upcoming elections: https://canvasbot.understood.chat

We constrain this conversation to the technology available between today and no further ahead than 18 months.

We discuss how and why creative-roles have been affected most so far, e.g. concept artists and writers.

We discuss how democratizing access to AI will lower the barrier to entry for new applications, and how there is a huge underground open-source AI movement.

We discuss roles such as Journalism which may be at least partially immune to these changes.

We cover how LLMs can provide the median information on a subject today, but future models can easily have specific tilts or opinions.

We cover open-source transcription tool MacWhisper vs paid silicon-valley startup sweetheart Otter.AI (powered by Google Speech to text) and how open-source is winning.

Alex discusses AI for medical transcription in his previous role in Babylon Health, and how important AI safety is in this subject.

This gets us on to high risk and expert roles - where you need responsibility and connection - and how they may be the last to really shift.

We cover how with legal notes you don't have to be 100% correct, you just need to make logical arguments that are legally sound - so our job in creating an AI assisted legal services is easier than medical AI's job.

The clerical parts of many roles will start to go in droves over the next 18 months, as many people hate this part of their job and it's easy for LLMs.

Links:

Google C4 dataset analysed by The Washington Post: https://www.washingtonpost.com/technology/interactive/2023/ai-chatbot-learning/

Snapchat’s AI flop https://www.businessinsider.com/snapchat-ai-tool-chatbot-total-fail-with-users-app-store-2023-4

AI Pizza commercial: https://www.youtube.com/watch?v=qSewd6Iaj6I

Call Annie (Free video call with AI avatar Annie): https://callannie.ai/

Join ML Research Scientist Alex Papadopoulos Korfiatis and AI Startup Founder Rafie Faruq, a Philosophy and Ethics expert, as they explore the challenges and opportunities in creating ethical AI.

AI Ethics is a massive subject. We barely touched on each of these, but will do more on each in future episodes.

AI Bias (e.g. ai pictures of lawyers are usually white males)

AI Safety - (lots of examples)

AI Explainability - AI showing its working and being transparent

Copyright & IP in AI - and at scale (who owns the prompt? Who owns the response?

Quality of AI - Used to be much more of an issue, but quality is improving rapidly.

AI regulation - Is any company taking proactive steps? Is any country clearly 'ahead' on this area? We all know that regulating artificial intelligence will not be easy. Would we have regulated Social Media differently if we knew the effect it would have on society?

Accountability and Responsibility in AI - i.e. Who is responsible for the output. The user prompting, or the creator of the model? What about the businesses hooked in to the model via API and layering a UI on top?

Links to us on other platforms:

I think just go for it re podcast - and I’ll share the link to all the places people can listen in a comment:
In order of platform popularity:
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

This week we are discussing Google Search and Large Language Models.

We'll be exploring how LLMs and Transformers are encroaching on the world of search engines. We discuss how AI is changing the way we find information and interact with the internet from BabyAGI to AutoGPT. We touch on Filter bubbles, Hallucinations (and how nobody has solved this yet).

News from this week in AI

Crazy things we saw this week

Autogpt (GPT-4 running fully autonomously) - https://twitter.com/mattcduff/status/1642528658693984256

Photographer admits prize-winning image was AI-generated https://www.theguardian.com/technology/2023/apr/17/photographer-admits-prize-winning-image-was-ai-generated

https://callsam.ai/ • Call Sam - call and speak to an AI about absolutely anything. Fun thing to try out

Some questions from this episode

What do you think the future of Google Search will look like in the next 5-10 years?

There has been some concern about AI and LLMs creating "filter bubbles" that limit the diversity of information users are exposed to. What are your thoughts on this issue, and how can it be addressed?

Next week's episode - AI Ethics