OpenAI and Elections | Karpathy and Simulations | Anthropic and Sleeper Agents | XKCD and Binoculars

Get on my daily AI newsletter 🔥
[News, Research and Tutorials on AI]

00:00 Karpathy and Simulations
06:09 OpenAI and Elections
12:36 Anthropic and Sleeper Agents
16:39 XKCD and Binoculars

My AI Playlist:

[BTFranklin]

I really appreciate your interesting and well-thought-out takes on these subjects. I watch several different AI News "influencers" but I think you have a distinct voice. (Not audio, but positional.)

[countofst.germain]

This video is philosophical gold. So much to think about here.

[Kopp]

Wes, you definitely need to add a sleeper phrase to Natural20. It is the perfect hiding spot since it is mostly educational content and would likely be scraped for training inputs. I look forward to a video where you add it in!

[adfaklsdjf]

i wish i shared your optimism about potential for abuse. i think the tech will cut both ways; some things get worse while others get better, and i'm not confident ground truth will net-benefit. in 2003, i thought the internet would help people converge on truth...

[OriginalRaveParty]

I've already seen Twitter posts from visual artists, and had two experiences myself, of rubes accusing artists of "faking" their work using AI. It's a fairly soul destroying insult to those who have put thousands of hours of practice into their craft.

[Bronco]

Good point Wes. Humans have been lying for thousands of years and suddenly people are acting like AI could lie is the end of humanity??? Lol

[Gome.o]

Can you please put links to the videos you use in the description?

[Yocoda]

Links to the lex interview and the other Karpathy video???

[lenderzconstable]

One thing that worries me is the people and business who are in denial and who choose not to adapt, as their business and possibly much of what they have will be taken away

[claudioagmfilho]

🇧🇷🇧🇷🇧🇷🇧🇷👏🏻, Amazing video, thanks for sharing!

[Aquis.Querquennis]

Nice metaphysical and existential approach. It reminds me of the only two Tetris players who managed to finish it in a predictable but surprising way at the same time. And also from us as observers of the quantum world that works only when we interact with a small part, leaving out the rest of the Universe. Paraphrasing Gabriel in The Prophecy: way back, before you were born, I told you a secret, then I put my finger there and I said Study your math, kids. Key to the universe!

[apache]

any ideas for what to do with a sleeper agent in a LLM? I could easily submit that as fake training data to openai

[fippe]

Honestly feel like we have already found a "glitch", splitting the atom feels very glitchy

[SpinX]

Could be wrong, but pretty sure there will only be one generation of AGI, anything more advanced than that would be ASI.

[chrisanderson]

Scientists:" Let's find the universal buffer overflow and blue screen the universe!!"

[djmccloskey]

On you not seeing the threat LLMs pose for elections, I believe you are fundamentally wrong to minimize this point. I agree that social media has already provided great tools for this kind of disruption, however, it is not just the proliferation that is afforded by LLMs that is a concern, it is the ability to use their understanding of language itself to create many variants of a post, asking it to select the language that is most triggering based on a topic or movement. This means it can act as an amplifier for the human driven features of the social media platform with respect to sowing doubt, dissent and polarization. LLMs essentially make it possible to probe the words and style of speech which most triggers, and to multiply.

[keeganpenney]

A very ' Report On The Barnhouse Effect' concept.

[TheCaphits]

Do you ever watch Rational Animations?

Lots of good (but old) AI safety stuff.

[kajsing]

Funny, I use the same comic to illustrate how far AI was today, in a presentation the other day. :)

[Thedeepseanomad]

I somehow doubt we will find many relevant exploits in the relationship between the forces of nature. I imagine it will be like trying to find edges of significant size on a very smooth sphere.