Yolin
@nilsengu
Lifelong student - most interested in emergence, complexity, information theory and computation. constantly updating my theory of how the world works 🌎
This is a profound question. And one I have been obsessed with researching the last few months. The latest turn from the industry, after o1, is all about trying to crack reasoning and generalizing RL. Which will be useful and important in its own way for practical reasons but is…
What is it that specifically separates human general intelligence from LLMs? They're definitely is a big difference. LLNs are not AGIs. Not even in principle. But most of the answers I've seen to this question are pretty weak. We need a better answer.
Also there’s the sneaky asymmetry wherein the costs are much more visible, more socially acceptable to discuss with friends, and insidiously lend themselves to quantification. Whereas the benefits are what Wolfram would call, computationally irreducible, because they create…
Your periodic reminder that everything seems terrible on a cost analysis. What are the benefits? If the benefits of having children are vague or emotionally unreal to you, why might that be? How might you fix it?
Why is this an ad
Official results are in - Gemini achieved gold-medal level in the International Mathematical Olympiad! 🏆 An advanced version was able to solve 5 out of 6 problems. Incredible progress - huge congrats to @lmthang and the team! deepmind.google/discover/blog/…
The only satisfaction MWI gives me is knowing that at least some version of myself is going back to every bookmark I have ever saved
I noticed this very clearly about a month or so back. Alignment researchers need to take into account the emergent dynamics in society when we combine the traits of very high degrees of agreeableness and, at least the perception of if not the reality of, hyper intelligence and…
Sycophancy is a much deeper problem than just the recent flawed 4o update that was rolled back from OpenAI. LLMs will need different personas to be effective for different goals, a teacher AI should be more disagreeable than a friend AI, and ICL should be minimized where pushback…
Jagged intelligence is a good framing. I suspect we will come back to this idea of smooth vs jagged more and more as in the coming years.
New blog post about asymmetry of verification and "verifier's law": jasonwei.net/blog/asymmetry… Asymmetry of verification–the idea that some tasks are much easier to verify than to solve–is becoming an important idea as we have RL that finally works generally. Great examples of…
Interestingly he was the one who introduced the concept of autopoiesis to biology. Another polymath that saw the pattern behind patterns
Francisco Varela
So let me get this straight, Windsurf got acquired twice by two separate companies in two separate deal structures? It’s like Descartes mind-body split but for tech m&a
Cognition has signed a definitive agreement to acquire Windsurf. The acquisition includes Windsurf’s IP, product, trademark and brand, and strong business. Above all, it includes Windsurf’s world-class people, whom we’re privileged to welcome to our team. We are also honoring…
Notice an asymmetry in how language handles time. For the past, we default to the neutral word “memory.” Words like nostalgia and regret that add emotional valence come up much less frequently. However, when we speak of the future, we almost exclusively frame it with emotional…
I disagree but I enjoy your counters :) To be clear, I am critical of modernity but in the tradition of McGilchrist, Tarnas or Charles Taylor. I think there is objective truth even though it’s not the same as absolute truth, and the pursuit of which should also integrate the…
Both Rorty and Popper got this but Rorty, was only a fallibilist and not a critical fallibilist like Popper because he was more aware of the fundamental limitations of abstractions and representations.
However, the goal of irony is an intermediary goal. It helps us notice the gap between representation of reality and reality, the gap that allows for meaning making. The thing though is that meaning can only be asymptotically approached to begin with. We will never have a society…
Creativity samples the space of “the adjacent possible” of Stuart Kauffman, however; to be useful and coherent, it also needs to do data retrieval. It’s a both/and
AI is showing us that creativity sits in an entirely different category to data retrieval.
Time to introduce another term for you to imagine towards a (biased) understanding! The landscape of affordances and (more completely imo) the pragmatic landscape. Just imagine affordances as the actions available in the world. Seeing things or systems as not merely in what…
Re: “The real culprit is Lina Khan” So just to be clear, you’re saying tech leaders can’t be trusted to act ethically on their own—because, surprise, they optimize for themselves within whatever constraints they’re given, including our current regulatory and enforcement…
My steel man here is that: 1. Employees who haven’t reached a 1 year vesting cliff don’t have that strong of a claim around “I built this with sweat and need a liquidity event” and we don’t know the tenures of all the left-behind employees. 2. The real culprit here might be FTC…
Someone explain to them what model weights are
Imagine paying $118k for a set of man-made numbers