วีดีโอ

🔥🔥🔥

So many tools in that direction. Many focused on specific aspects (mostly scientific research, probably because it has somewhat higher quality sources by default and also more difficult questions), Elicit already mentioned, another one is undermind, which I find absolutely awesome and they have exactly that interface: you talk with LLM first which helps you to narrow down your request and then set it off to crawl the papers for minutes and come back to a whole report with timeline, sources, citations, key people, etc. Also, what I always wanted from search and what was mentioned briefly: citation graphs, i.e. when we have some influential piece, like Founder Mode, I would like to see reactions to that, scientific literature style -- who mentions it, criticizes it, builds upon it, etc. Another, aspect that wasn't touched: how much of information is in the videos. Like, will Exa (or any other tool) will be able to ever find the list of emails Swyx once flashed on a stream when I ask to find his contact information? Will it be smart enough to try to find it if I ask for contacts of people who interviewed people who like search? :) Will exa ever find me as a person interested in search based on that comment to add to a pool of recruits? Anyway, great interview, great work and hope we'll get to yes to all the above questions and superhuman search :)

When you fire up ComfyUi , during the book sequence you will eventually hit a quote. These quotes speak about the pursuit of art, being an artist, being brave and taking action. Reading them, I knew this software was special.

Thank you.

I just did a search on looking for specific posts on both Perplexity and Exa API, I got exactly what I am looking on Perplexity but 0 on Exa API. So, I am not sure how much I can trust this neural search. Maybe keyword is probably better search which even costs less. Even DeepSeek V3 chat was better. It seems like the limitation is what does their search index have at this point.

Great great talk and discussion! I'm excited to see what the triumvirate of Tri Dao, Albert Gu and Dan Fu are up to in the future. To me they are the next generation of Ilya, Alec and Vasvani et al.

Did you post that doc somewhere?

This was great. I'm amazed by how quickly the open source democratization of AI is going. A lot of people are investing a lot of money in AI to control something that is not controllable.

Now i want to see how automatic1111 kohya and forgeui looks like :D

Wait, is not an anime girl?

Perhaps the solution is to combine all the architectures into one big Frankenstein model, and name the paper Diversity Is All You Need. I'm just spitballing, but it seems funny enough to be true.

Thank you very much man. I love ComfyUI.

AMD, hook this guy up with some hardware.

Comfortable

I myself too got addicted to generating "images" back in the automatic1111 days.

Thank you for your service, comfy. Love your project, and after investing some time, I got used to it after 1111. : ) Hope it will be around for years to come.

Amazing! Thank you for the interview.

now that Comfy has been on camera, i wonder if he would ever consider doing his own videos about comfyui?

Thank you for the interview. Too bad the lack of preparation from the interviewers. It could be more informative with better questions and knowledge from them. But thank you!!

Synatic data works from .... synthetic capability haha get it ...

This guy is a legend.

Finally, we see the big brain behind Comfy.. Thank you very much for the interview...

Great Interview

Amazing interview!! Thanks!!

Gradio also seems to send way to many http requests, like if you change a single drop down, often it would need a whole http request. Which kinda totally defeats the concept of a frontend UI.

That was revealing ^^ nice to put a personality to the name I type into my terminal so often. This interview feels historic.

<3

Great interview.

Long time podcast listener but am loving the TH-cam versions, the extra work is appreciated!

A coding forward agent will scale much better as models get better too.

the RWKV approach makes so much sense with what we're seeing with distillations, why not just run in through an RNN

Great coverage. More of these shorter pieces

where can i get a hold of the google docs ?, it would be really helpful

Consciousness is. We’ve got it backwards. Consciousness is the thing that exists and the stuff in the World is all properties of it. There is only one consciousness and we are an outcropping of it. The thing that animates us is consciousness. Consciousness is life. They are one and the same. Consciousness has amassed the stuff of this observable universe by layering that spark, by stacking it. The steps taken at each point of transition or stage are neither geometric nor exponential. Logarithms will not suffice to describe it, nor the complex field or fractal dimensions. That is because there is an inner quality that is doubling- and for the word doubling, which is a specific quantization, we may substitute an overflow of life energy(or call it probability distributions), if you want to call it that. Life-consciousness has been compounding while we transit or go through the process of something that we perceive as temporal flow. The metric for this compounding, this complexification is determined entirely by the medium under the knife. It is a meta level operator. We experience it as Darwinian evolution in the context of motile organic molecules and the compounding of these. Time and space are models built by consciousness stepwise, but these steps are more like the drag of a snail because they are continuous, not discreet. We have to re-root our conceptual reality. The things we think we are and have are just projections of what actually is. So it is clear we are a unitary thing like sprites growing on a giant stalk. There is no graviton. Gravity is the lowest rung on the ladder, it is the first step of consciousness. Gravity is the shape of the universe. It is the simplest, most direct, most fundamental exposition of life-consciousness. So now we see that it is gravity-life-consciousness, because these are one and the same. They are like the incarnations of Hindu gods- all manifestations of the same essential being. Reasoning itself is the emergent property. Where did the desire to reason come from. And how do you explain the feeling of the experience.? How this works is beyond the human brain‘s ability to understand, and one has the bear in mind, that the mind has to harness the brain into its creative acts. Biology or technology, it’s beyond dimensionality. Try to imagine if the Meta level of things had a Meta itself. We are talking about something orthogonal to dimensionality. Nevertheless we can make sense of the observable universe. The Social mind is more powerful than the other minds combined because it can override them with the weight of its conclusions against theirs. What the Sociel mind decides is final until extenuating circumstances intervene.

Nathan's recap is impressive. His focus on the nuances of o1 and o3 iterations, and reinforcement finetuning, truly enriches our understanding. Check out his work on Interconnects!

how can we collaborate

hey guys, you mentioned 1:02:00 that we can find the slides in the description. however, they're not there rn. can you share the link? thanks!! awesome wrap!

Congrats on 100 episodes! Reflecting on AI's rise is crucial. The journey of AI engineering and legal/ethical issues sets the stage for future trends.

15.25 what about R1 from deepseek and their latest version o deepseek 3?

Hey what about the R1 from Deepseek 😅

audio and newsletter: www.latent.space/p/2024-review

First! 😊

ah weeding out those quadratic time functions from the codebase

great interview, guys. It's a true pleasure listening to Erik talk in very human terms about the subject. also big props to the interviewers for not interrupting ;)

very cool

Here after Sam Altman's tweet about Alex Radford.

super talk!

really good talk, thanks! Super helpful to see the whole recap of what got us here and what's next on the horizon!

22:45😂

Long context is absolutely needed for software code bases. Alternative is RAG but there are still issues with RAG when it comes to software.