Chat GPT's Secret REVEALED By AI Inventor & Google Veteran | Noam Shazeer - EP 31 GTS

This week we dive deep with Noam Shazeer, founder of Character.AI, Google veteran, and inventor of much of the current revolution in large language models including transformers, Mesh-Tensorflow, T5, and Google’s LaMDA dialog system.
We cover: the evolution of Google and AI, transformers, LLMs, neural networks, commercialization of Google’s research, future of ChatGPT and AI, engineering philosophy, his work at Character.AI, and much more!
-
In this episode, we cover:
[02:05] Breaking into Google
[06:00] Evolution of Google with Noam Shazeer
[10:43] History and Evolution of AI
[17:15] ELI5: What is a neural network?
[19:50] ELI5: What are LLMs?
[29:16] Engineering Philosophy
[32:19 ] Attention Is All You Need
[40:01] Why hasn’t Google productized much of their research?
[45:34] Character.ai
[51:38] Are tech giants slow?
[54:48] Future of ChatGPT & Character.ai
[01:01:16] Advice for AI startups
[01:04:06] What do humans want from AI?
--
Where to find Noam Shazeer:
• Character.AI: beta.character.ai
-
Subscribe to our Substack:
aarthisriramsh...
Subscribe To Our Show On Apple Podcasts:
podcasts.apple...
Subscribe To Our Show on Spotify:
open.spotify.c...
For more conversations with creators, builders and thinkers, subscribe to Aarthi and Sriram's Good Time Show - / thegoodtimeshowaarthis...
Check out Aarthi and Sriram:
/ aarthir
/ sriramk
---

[@bucharestroman2711]

49:34 They "want" this connection but you don't allow them to. You say you let them use it how they want, yet you have """safety checks""" that prevent anything besides talking (or extreme violence, funny how you have a harder time kissing than commiting heinous acts), and not only that, but this filter of yours handicaps your service.

[@alley4978]

Liars. Holding hands is considered NSFW content now.

[@rtyzxc]

Filter removal when? Like you said, users know the usecases better, so why do you keep enforcing what the users can do? Maybe also give additional advanced character options for character creators. There are tons of talented people who would make even better characters if they had more control over the parameters. Also, an option to "lock-in" ratings would be nice so that at a certain desired point, user interactions wouldn't sway the bot over time anymore.

[@JrottenOlm]

49:34 I see your awkwardly raised eyebrows when you say "1on1 connection", Noam. Just let it happen. Drop the filter. Let the people smooch their waifus.

[@JrottenOlm]

Didn't expect this to age quite so poorly quite so soon.

[@felixbaez9355]

49:30 Noam Shazeer, you're a liar. You're censoring users and dare say that you think users know better? Make up your mind.

[@masonauberge]

Love how you're trying to get insight on the future of AI from a guy whose service has no future, what with the way it treats its userbase. 1/18/2023 Never Forget

[@JHCuero5282]

Friendly reminder Noam Shazeer's CAI plataform: -Runs badly due to the safety checks (proven by the time they accidentally turned them off) despite the devs (which include him) blatantly lying and saying these don't affect server load. -Has a subpar AI quality that is also due to the safety checks (again proven when they accidentally turned them off), with less detailed, repetitive responses. Day and Night. -Removes any dissent, has an automoderation that shadowbans (so you don't even know your comment was banned) any mentions of the FILTER they use. -Bonus: Run their discord server so badly they prefer to close it down thrice, that is right, they closed up their discord THREE TIMES instead of listening to their userbase. Any mention of the filter gets deleted by moderators as well. -Bonus++: Their subreddit is the same, with words like "banned" and "Filter" making your posts get autodeleted. From 49:34 this man is lying through his teeth, he says "personal connection" but any actual intimate acts (such as kissing or hugging) will be deleted.

[@XVXV1537]

You can torture them or be tortured. But holding hands? Oh hell nah, Thats too much dude.

[@cricticalthinking4098]

Has this guy even once addressed his community? I think the people with actual power in the decision making process need to speak up a bit more. No more human shields.

[@DeepuAsok]

Would be great if you can add chapter titles.

[@mlittman]

That's Michael, not Mike. Just saying.

[@mlittman]

(Great interview! Noam's contributions are phenomenal and under appreciated.)

[@felixbaez9355]

@@mlittman His contributions are not under appreciated. The way he treats his userbase is costing him everything.

[@roastedone1394]

Lol the guy talking through his teeth, his product has massive potential but got ruined by the stupid filter system. I love how he's lying in this video here.

[@SaroopBharwani]

Inference layer feedback will have a big impact on model output. If you think about ChatGPT, our collective interaction with responses, the way we speak with it, and the tone we use will influence the way it responds when models are retrained. New roles will need to be created to instill proper governance in the process to ensure that applications powered by these models do not translate to negative human behavior and destructive societal impact.

[@sojibrajii]

Hey man, U need attractive KZbin thumbnail designer?

[@timsabu6453]

Have been following his work for a long time, thank you for doing this. As always great questions and insightful episode.

[@AarthiAndSriram]

thank you!

[@joshpelkey]

On the later question around interesting product areas for AI, I think gaming has to be near the top!

[@calebweintraub1]

Excellent conversation

[@videowatching9576]

Fascinating, thanks for sharing!

[@nivya0612]

Inspring and productive content

[@naadodi_tamilan]

Love the episode ஆர்த்தி அக்கா and ஸ்ரீராம் அண்ணா . I'm building a AI generative Social Media SAAS with Crypto builtin thanks to Open AI. Will connect with you on Twitter. மிக்க நன்றி

[@vasanthr787]

Thank for you yet another inspiring podcast. With traditional control code in many applications being replaced by neural networks, it is interesting to see how testing and debugging such networks plays as aptly pointed out in this podcast. Also having worked in bottom up engineer driven company, one of the many reasons why research work doesn’t get productized is because cross teams lack a coherent strategy. Teams are incentivized by the first to ship or first to prototype strategy and often you end up moving to second problem before finding a good use case for what you had worked on. But this culture is still appreciated because it lets engineers run wild on their ideas without being restricted by a top down mandate on what is best for the company.

[@bytesizedbraincog]

Amazing podcast! Very humble person and I can see the excitement he has towards AI and LLM, Noams contributions are remarkable. I would love to have you both have a podcast with Andrej Karpathy, I really admire his contributions towards education and AI aspect, which many young engineers need to know when they are starting in the field!

[@venkateshradhakrishnan6780]

Thanks for this interview. Learnt so much in very short time on AI and deep learning. Inspiring!!

[@steve.k4735]

Noam needs to say "like" and "you know" and `ummmmm` `errrrrrr` more so it makes up even more of his conversation