leopoldj

From Sofia to the West of Turkey should be relatively easy. After that, travel through Iran, Afghanistan and Pakistan will get hairy.

A few more:

Jeff Dean doesn't use a compiler; he just glares at his source code until it executes.

Jeff Dean once optimized a sleep(10) call to return in 5 seconds.

Jeff Dean’s keyboard doesn’t have a Backspace key; he simply doesn't make mistakes.

/end. There's no need to get up. I will see myself out.

I just wanted to point out that Boom Aerospace is now mainly focused on delivering a Gas Turbine for power generation [1].

In an interview with CNBC Mr. Scholl talked about this pivot [2].

1. https://boomsupersonic.com/press-release/boom-supersonic-to-... 2. https://www.youtube.com/watch?v=ELl2uUAfGBw

I apologize if this is common knowledge. Modern C++ coding agents need to have a deep semantic understanding of the external libraries and header files. A simple RAG on the code base is not enough. For example, GitHub Copilot for VS Code and Visual Studio uses IDE language services like IntelliSense. To that extent, using a proper C++ IDE rather than a plain editor will improve the quality of suggested code. For example, if you're using VS Code, make sure the C/C++ Extension Pack is installed.

What is the halation like artifact around the edges called? This makes the picture so unappealing to me. I'm so glad the modern digital cameras don't have this problem. I hope I'm explaining the issue well. I wish I could post a frame grab here.

This ship is a known blockade runner. "The ship has frequently carried oil from countries under U.S. sanctions, and its tracking data shows multiple recent trips to Iran and Venezuela"

https://www.nytimes.com/2025/12/10/us/politics/oil-tanker-ve...

As the article states, generating the key itself is easy. But getting credit and billing are the issues.

>As for Copilot? I don't know anyone who uses it. Do you?

This sloppy journalism. One should probably read the original report in The Information [1].

Bloomberg has updated its story today with a note from Jeffries [2]. "The analysts also said their checks showed robust adoption of Microsoft’s Copilot line of AI assistants"

1. https://www.theinformation.com/articles/microsoft-lowers-ai-...

2. https://www.bloomberg.com/news/articles/2025-12-03/microsoft...

You went further with GCP than I did. I was asked repeatedly by support to contact some kind of a Google sales team.

I get the feeling GCP is not good for individuals like I. My friends who work with enterprise cloud have very high opinion about their tech stack.

This kind of circular deals are very common and doesn't violate any laws. It's the magnitude and prevalence of it in the AI sector that flashes a warning sign.

Somewhat related, if you are on a C++ project, please consider std::from_chars. It's non-allocating and non-throwing. Works with non-NULL terminated strings.

https://mobiarch.wordpress.com/2022/12/12/string-to-number-c...

The blog doesn't verify if the code is actually using the GPU. The code will work perfectly fine on CPU, albeit slowly. You should run this to be sure:

    python -c "import torch; print(torch.cuda.is_available())"

Strange that torch.cuda.is_available() is used for AMD also.

Use rocm-smi to be double sure.

There's more than one way to do self supervised training.

This is the approach the author has taken.

    Training corpus: "The fat cat sat on the mat"

    Input -> Label
    --------------
    "The" -> " fat"
    "The fat" -> " cat"
    "The fat cat" -> " sat"

Hugging Face's Trainer class takes a different approach. The label is same as input shifted left by 1 and padded by the <ignore> token (-1).

    Training corpus: "The fat cat sat on the mat"
    Input (7 tokens): "The fat cat sat on the mat"
    Output logit (7 tokens): "mat fat sat on fat mat and"
    Shifted label (7 tokens): "fat cat sat on the mat <ignore>"

Cross entropy is then calculated for the output logits and shifted label. At least this my understanding after reviewing the code.

The author is conflating RAG with vector search. I think.

One can use any and all available search mechanisms, SQL, graph db, regex, keyword and so on, for the retrieval part.

Multiple registrations are being filed for the same person in order to game the system. This is discussed in some details in a USCIS report [1]. The increased application fee is presumably to stem that practice.

https://www.uscis.gov/working-in-the-united-states/temporary...

Is the io_uring code faster than mmap because multiple worker threads? If we use mmap from multiple threads will we get almost the same result?

There are two independent issues that bear discussion.

1. Should we be mandated to do something that while a good thing appears to infringe on personal freedom. Such as, mandate good diet, exercise, vaccine etc. People here have made very good points about why sometimes we have to forgo certain freedoms for the greater good. At the same time, we should not forget how some countries take this too far and become nanny states.

2. Are the Florida officials spreading misinformation and fear against vaccinations? Some of Dr. Ladapo's recommendations can certainly be viewed that way.

You are right. But newspapers make it difficult to distinguish them. Opinion pieces are liberally distributed amid news items. LA Times has a Columns section on the right hand side. But, this particular piece is not listed there. It is listed next to other news articles.

Thank you so much for posting. I finally understand Jacobian matrix. The key is to know this applies to a function that returns multiple values. The wikipedia article was difficult to understand, until now! Note: Technically, a function can map an input to a single output only. Here, when we say a function returns multiple values, we mean a single set of multiple values. For example, a function that outputs the heating and cooling cost of a building. Where as, a circle is not a function because it outputs two y values for a single x.

AIX was and probably still is an extremely capable OS. IBM has contributed a lot to the Linux kernel since then. Linux may be good enough now and probably a better option for many reasons for most orgs.

Reporting by science.org on the same trial:

https://www.science.org/content/article/immune-dodging-cells...

The point of this is not so much to compare humans with AI. But to compare AI with other traditional software development approaches to solve this domain (IQ test, in this case). I believe, and I could be wrong, it will be nearly impossible, or too expensive, to develop deterministic software to beat AI in IQ test.

google/gemma-3-4b-it is one of the smallest multimodal models I know. Works well in a 16GB GPU. Works slowly in a 8GB GPU. It can even be fine tuned [1], which where the real power comes from.

1. https://ai.google.dev/gemma/docs/core/huggingface_vision_fin...

You are testing this model for knowledge. That's not the primary use of a model like this. They are meant for instilling domain specific skills and knowledge through fine-tuning. The blog post goes into that a lot. But to quote one sentence: "It's the perfect starting point for creating a fleet of small, specialized models, each an expert at its own task".

What the author is doing here is pre-training. This is something usually model makers like Google and Meta need to do. Most business are much better off doing fine-tuning or to a lesser extent continued pre-training. The author is doing this for academic reasons.

Over-investment in a new technology seems to be a common thing. It happened with the Internet and railroad. A bubble forms which pops. But the process leaves behind valuable infrastructure.

I gave it a properly developed story. Gemini altered it somewhat to fit it in 10 pages. It was more than acceptable. But the illustrations, asked for in the style of oil paining, could have been better. We will get there.

> Ollama has moved off of llama.cpp as a wrapper. We do continue to use the GGML library

Where can I learn more about this? llama.cpp is an inference application built using the ggml library. Does this mean, Ollama now has it's own code for what llama.cpp does?

For production use of open weight models I'd use something like Amazon Bedrock, Google Vertex AI (which uses vLLM), or on-prem vLLM/SGLang. But for a quick assessment of a model as a developer, Ollama Turbo looks appealing. I find Google GCP incredibly user hostile and a nightmare to navigate quotas and stuff.

The self hosting section covers corporate use case using vLlm and sglang as well as personal desktop use using Ollama which is a wrapper over llama.cpp.