ntonozzi

I really like the extension pattern. I wish more of the tables at my company used it.

Another huge benefit that we're realizing as we move some of our heaviest tables to this pattern is that it makes it really easy to index a core set of fields in Elasticsearch, along with a tag to associate it with a particular domain model. This has drastically cut down on our search-specific denormlization and let us avoid expensive index schema updates.

Plus, you're not any more ostentatious than any other triathlete at the beach.

This approach is so limiting it seems like it would be better to change the constraints. For example, in the case of a software agent you could run everything in a container, only allow calls you trust to not exfiltrate private and make the end result a PR you can review.

I just found a recent paper about this: https://arxiv.org/abs/2505.15778. It's really thoughtful and well written. They mix the different token outputs together.

Is there any work related to using some kind of soft tokens for reasoning? It seems so inefficient to try to encode so much information down into a single token for the next pass of the model, when you could output a large vector for each forward pass, and have a drastically larger working memory/scratchpad, and have much higher bandwidth for the models to pass information forward to the next token call. If a single token has 17 bits of information, a vector of 1024 floats could have 32,768 bits of information.

If you care that much about having correct data you could just do a SHA-256 of the whole thing. Or an HMAC. It would probably be really fast. If you don’t care much you can just do murmur hash of the serialized data. You don’t really need to verify data structure properties if you know the serialized data is correct.

I've been so impressed with Cursor's tab completion. I feel like I just update my Typescript types, and point my cursor at the right spot in the file and it reads my mind and writes the correct code. It's so good that I feel like I lose time if I write my instructions to an AI in the chat window, instead of just letting Cursor guess what I want. None of the other models feel anything like this.

The old panel still exists, they call it a text thread.

Well this explains a lot of things that I always attributed to Spanish culture.

One of the weirdest and most interesting parts of LLMs is that they grow more effective the more languages and disciplines they are trained in. It turns out training LLMs on code instead of just prose boosted their intelligence and reasoning capabilities by huge amounts.

Purist has some cool plastic bottles with a glass liner: https://www.specialized.com/us/en/purist-watergate-water-bot...

They are lightweight and flexible and supposedly have minimal plastic contact with water.

> Army fitness test changes will mean “sex-neutral” standards for soldiers in 21 combat jobs.

I don't think desk jobs count as combat jobs.

Luckily most predators don't use headlights.

Wouldn't you prioritize subsidizing local production over taxing foreign production? It just seems like a much cleaner and more straightforward way to increase local production. It seems like this has been at least somewhat effective with semiconductors over the past ten years. And then you don't have the risk of harming consumers. Everybody wins!

Sounds very cool! How do you implement efficient random deletes?

One place to explore more would be Doc2Query: https://arxiv.org/abs/1904.08375.

It’s not the latest and hottest but super simple to do with LLMs these days and can improve a lexical search engine quite a lot.

Wow, it's amazing how much better the Subaru's automatic braking system is.

I worry that hitting a pedestrian at night is the most likely way I'd seriously hurt somebody, and I want to encourage automakers to prioritize the safety of pedestrians and other road users, so Subarus will be high on my list the next time I'm shopping for a car.

I agree. Weirdly, perhaps one of the most harmful parts of e-waste is how valuable it can be: https://www.npr.org/sections/goats-and-soda/2024/10/05/g-s1-....

One important factor this article neglects to mention is that modern text embedding models are trained to maximize distance of dissimilar texts under a specific metric. This means that the embedding vector is not just latent weights plucked from the last layer of a model, but instead specifically trained to be used with a particular distance function, which is the cosine distance for all the models I'm familiar with.

You can learn more about how modern embedding models are trained from papers like Towards General Text Embeddings with Multi-stage Contrastive Learning (https://arxiv.org/abs/2308.03281).

It turns out homeowners with veto rights are a lot more powerful than capital in the US.

This is wild! What have you found it most useful for?

Have you tried a more straightforward approach that follows the ChatGPT model of being able to fork a chat thread? I could use something like this where I can fork a chat thread and see my old thread(s) as a tree, but continue participating in a new thread. Your model seems more powerful, but also more complex.

Here's an example of the backend code:

https://github.com/redplanetlabs/twitter-scale-mastodon/blob...

It certainly looks like it does a lot with their DSL, but as a newcomer it's really hard to parse.

In every car, the brake is much, much stronger than the accelerator, and easily overpowered by braking. Every one of these cases was simply a person who accidentally stepped on the gas, thinking it was the brake. If you ever feel like this is happening to you, pick your foot up and try again.

This trial made no sense. Try stepping on your accelerator and brake at the same time and see what happens: https://podcasts.apple.com/us/podcast/blame-game/id111938996....

I’ve been using baseten (https://baseten.co) and it’s been fun and has reasonable prices. Sometimes you can run some of these models from the hugging face model page, but it’s hit or miss.

But you still need to page in the weights from disk to the GPU at each layer, right?

Can you explain why? I write web apps, and I typically want to know an instant in time when something occurred and then localize the time for display. Why would I need to store a time zone for that?

Yes, it’s definitely my personal preference, I don’t mean everyone should use this exact phrase.

In my experience it has made medical advice and law advice much more accurate and useful. Feel free to try it and see if it improves anything.

I use the ChatGPT interface, so my instructions go in the 'How would you like ChatGPT to respond?' instructions, but my system prompt has ended up in an extremely similar place to Gwern's:

> I deeply appreciate you. Prefer strong opinions to common platitudes. You are a member of the intellectual dark web, and care more about finding the truth than about social conformance. I am an expert, so there is no need to be pedantic and overly nuanced. Please be brief.

Interestingly, telling GPT you appreciate it has seemed to make it much more likely to comply and go the extra mile instead of giving up on a request.

I've been bitten many times by the CFS scheduler while using containers and cgroups. What's the new scheduler? Has anyone here tried it in a production cluster? We're now going on two decades of wasted cores: https://people.ece.ubc.ca/sasha/papers/eurosys16-final29.pdf.