- Видео 2
- Просмотров 7 523
Fernando Rojo
Добавлен 2 май 2012
#1 on ARC-AGI?? 53.6% with Claude + Evolutionary Test-Time Compute
Here's our livestream from the announcement of ARC-AGI 2024 release, where Jeremy Berman sits down with Fernando Rojo to discuss how he got the highest score on the public ARC-AGI benchmark. Learn how Claude Sonnet powered an evolutionary algorithm and follow the internal technical details.
Source code: params.com/@jeremy-berman/arc-agi
Substack write-up: jeremyberman.substack.com/p/how-i-got-a-record-536-on-arc-agi
To find more open source models, check out params.com: params.com
Source code: params.com/@jeremy-berman/arc-agi
Substack write-up: jeremyberman.substack.com/p/how-i-got-a-record-536-on-arc-agi
To find more open source models, check out params.com: params.com
Просмотров: 1 343
Видео
How I Build React Native Libraries
Просмотров 6 тыс.Год назад
Burnt: github.com/nandorojo/burnt Zeego: github.com/nandorojo/zeego GitHub: github.com/nandorojo Galeria: x.com/FernandoTheRojo/status/1703880991628988646 My Twitter: x.com/fernandotherojo
I’m not entirely sure how to incorporate this next argument into a general framework about deduction, but one thought that comes to mind is the unique nature of certain tokens. Take, for example, the token representing "one" (assuming there's only one such token). In my view, this represents a higher-order concept-a distinctly abstract idea of unicity. For us humans, this specific token seems to transcend being just a word; it embodies the foundational concept that allows us to count, sum, and ultimately perform arithmetic. My point is that, while a degree of uncertainty might make sense for tokens associated with other areas of semantics, I think LLM architectures should be grounded in the fact that certain tokens-like "one"-carry an intrinsic certainty. Similar to how transformers embed the sequential nature of language into the attention mechanism, perhaps there could be a way to encode the idea of unicity into the model. Maybe some form of specialized module to help it grasp cycles or abstract concepts of singularity? Just throwing some ideas out there haha
On the latter, I wonder if you could train on sentences trasformer masked embbedings or similar.
Oh. I was just about to say it would be good if we would see the screen. Then they realized they missed to enable it 😃
Video out of date after 24 hours
Signs of the singularity curve. It's about to get extraordinarily steep.
Wow, horrible timing
OpenAI's o3 uses a STUPID amount of computing power.
@@kellymoses8566this. It's not horrible timing; this is a way more compute efficient approach
This is still valuable
I’d love to see how the ARC Evolutionary strategy we discussed performs with o3
Man do I feel dumb for not thinking about just using Swift libraries instead of hand rolling everything😅
@@demidaniel9253 You shouldn’t feel that way! It’s ultimately an unintuitive move and requires getting out of your comfort zone, but it can pay off nicely and be simpler at times.
God level
Hi, me and my team is planning to build a react native UI Library and i learned a lot from this video. It would be an honor to get advice from you tbh. You are awesome man!
what would you do if you needed to put swift or java in code with react native
check out my latest app.js talk!
@@fernandorojo6311 where do I find that
Hey, can you tell me about work flow of react native library development efficient way? I'm working on react native design system and I created a library and use yarn link for local development and it hard/full reload always (not hot/live reload) even small change. I tried lot of things but still no luck
Tbh I just use a local example and then publish and install in my app. Metro doesn’t like yarn link. But Onestack uses vite so yarn link might work there
Thanks man, I just started creating an expo module...tired of thr issues. You motivated to try again.
Much needed video thanks man ❤
After your video I feel dumb 😅
Same feeling
Make more videos
I was going to start using solito starter or installation. It starts with next13. Do I have to migrate to nextjs14 myself? I've looked up all the procedures such as solito upgrade to v4, but I'm wondering if I have to migrate it myself.
I think you can simply do 'yarn add next' and that's it. You can also contribute to the starter to upgrade it!
@@fernandorojo6311 already tried but it doesn’t change the “page router ” to “app router”.
this is great
Hey Fernando, really excited about Solito! Wish there was a full tutorial for setting up the complete stack and an example project
check out my 2022 Next.js conf that does this. Solito also has many examples!
U r a guy how goes through the best practices and say " Naa Bro , I am gonna do my own things "🕸️
This is awesome. One of the few videos that actually goes into the weeds of how real production software is made.
Thank you so much for taking the time to explain how you went about creating these two libraries. The information you provided here is very valuable for someone who is starting out. It's also nice to see your presence here on RUclips. I follow your work through other platforms and your talks but this is so much better and explanatory. I appreciate what you do and I am sure you will have more subscribers here soon too.
Happy to hear that. I’m planning on making more videos that get deep into the details.
Zeego's implementation is so awesome! 😲