olirex99

I am not sure on the experiments. For example let’s imagine that Astra is asked to improve the performance of a network.

At first Astra will look at the weights, and try to understand the model behavior. Then if asked he will start to optimize the weights. Now my question is: how can we be sure that inside the thinking process Astra is not hiding a gradiant descending computation?

I do not have social, and this is what hurts me most. People all say the same and think the same, even my friends.

How can I enjoy this world?

Thanks for the suggestion!

I define God as the plateau of intelligence. You maybe haven’t understood my question, but I am asking we are living in a more brother loop in which less intelligent species build more intelligent species. If the answer is yes, would this mean that AI species will build at a certain point (self-improving) another kind of species?

PS: I know that I use species improperly.

Also the definition of artificial will not make sense at all at a certain point. If for example AGI builds a more intelligent species, could we define it “artificial”?

The plateau of intelligence. And now you could ask me to define intelligence and I would stick with the common definition:

Intelligence is the mental capacity to learn, understand, reason, and solve new problems.

Some nice ideas on how people will consume application and software in the future. Let me know if you agree or not, especially about the “open community software” as a substitute to the “open source software”.

I think that this idea is very nice and can be recap in this simple formulation:

open source software : pre LLMs world = open community software : post LLMs world

I suggest you to give a look to the MCP protocol for hardware that is being proposed by Anthropic. The hardware will be the next harness of LLMs, they will be able to operate machines to reinforce their theories.

I still think that a major problem is that biological processes are not “fast” as coding, but they are verifiable. If during post processing we are able to give enough harness to test and verify this kind of environment (maybe via simulation and real data) we will for sure achieve incredible performance also in this domain.

I suggest you to give a look to the MCP protocol for hardware that is being proposed by Anthropic. The hardware will be the next harness of LLMs, they will be able to operate machines to reinforce their theories.

I still think that a major problem is that biological processes are not “fast” as coding, but they are verifiable. If during post processing we are able to give enough harness to test and verify this kind of environment (maybe via simulation and real data) we will for sure achieve incredible performance also in this domain.

Seems like that anyone can now prove math conjecture. Maybe someone already prove some math problem and is not even aware of it.

I am really curious to know how Karpathy would update this survival guide in the 2026. Hope to hear something from him!

Hello HN! This is my first open source project, Git Guide. It is a community-curated travel directory designed to replace SEO spam and fake reviews with a transparent commit history.

The project relies entirely on Markdown files and GitHub Actions—no database or servers. Users propose places via Issues, and once a voting threshold is met, a script automatically validates the location and commits it to the repository. This ensures total transparency without black-box algorithms.

I would love your feedback!

Maybe we can get in touch, I am curious to have a more senior perspective on the indoor localization!

All the new car have the keyless option and use UWB to open and close.

I am pursuing a PhD in indoor localization, and UWB is still far superior. That is the reason why major phone companies still include a UWB chip and are not switching to BLE 6.0.

I have compared them, and because BLE is a narrowband signal, it is highly susceptible to Non-Line-of-Sight (NLOS) conditions compared to UWB.

I also attended a prototype presentation by a large European silicon company. I noticed that even in their demo, BLE did not achieve 30 cm accuracy, but rather hovered around 1 m.

I have only tested PBR and RTT ranging with a simple Kalman Filter, so maybe someone has found a clever combination of these data sources (I hope).

Companies are doing some crazy stuff to block other devices. One of the worst things that I have seen is that on MacOS you cannot use the AirPods in a Microsoft Teams call.