aduffy

Very cool! At first glance this looks a lot like SpiceDB, which is also a Go service based on Zanzibar. Excited to see more entrants in the space, can you talk a bit about what makes Permify special?

Yes, the popular fork is called Postgres. You can find many vendors who will let you run it on one node cheaply. It’s also free to self-host.

I’m very curious, are you running the JDK inside the app? What difficulties did you run into trying to embed an external runtime?

From October: https://www.theinformation.com/articles/openai-dropped-work-...

If OpenAI had GPT-5 they would’ve released that instead of 4-Turbo.

OpenAI was allegedly working on their Arrakis model last year that didn’t end up panning out. It’s clear they’re hitting some sort of limit in the scaling laws, or cost efficiency, or both.

It’s really ORC++ as they mention in the video.

I have a thread with Andy Pavlo on the salient bits but basically it grew out of ORC modifications

https://x.com/andreweduffy/status/1778054712517857458?s=46

I remember using GitHub as a teenager, circa 2010. I, not understanding git because I was an idiot, accidentally force pushed and deleted my history.

I emailed the support email explaining the situation, and within a few hours got a reply from Chris saying that he’d fixed my repo, along with some advice about how to avoid this issue in the future.

Facebook invented thrift and gifted it to Apache

The quality of discussion in the NYT comments is exceedingly bad.

In our industry it’s easy to forget that the average person doesn’t understand what’s happening currently in AI and fears the worst, based on the FUD being shouted by mouthpieces who have something to gain by being the gatekeepers of who uses AI and how.

Looks like this coincides with seed funding[1], congrats folks! Did you guys just bootstrap through the last 5 years of development?

[1] https://www.saltwire.com/cape-breton/business/code-testing-s...

The "coup by MSFT" conspiracy theory made no sense. Microsoft has an insanely good deal with OpenAI:

    * Exclusive access to resell OpenAI's technology and keep nearly all of that revenue for themselves, both cloud and services
    * Receive 75% of OpenAI's profits up to $1 trillion

All they had to do is not rock the boat and let the golden goose keep laying eggs. A massive disruption like this, so soon after DevDay would not fit that strategy.

My guess at this point is financial malfeasance, either failing to present a deal to the board or OpenAI has been in financial straits and he was covering it up.

Yes I meant people in government who are responsible for spending money and overseeing programs. NASA and NIH scientists and DARPA PMs are amazing innovators that should be celebrated. All of these people are responsible for an incredibly small piece of government spending.

Recoding America[1] is an excellent book about this problem, but it extends a lot further than just contractors charging inflated rates. The government itself has a culture that is so averse to trying new things and compliance with existing policy that it's actually dangerous to the careers of civil servants to do anything innovative, because that is one of the only things they can actually be fired for.

The solution she proposes effectively would be an expansion of the US Digital Service (which the author spearheaded) into something that approximates the British civil service, which is actually highly functional and effective.

If you actually care about this problem and have a spare Audible credit it's worth a listen.

[1] https://us.macmillan.com/books/9781250266774/recodingamerica

The model discussed in the article is MiniLM-L6-v2, which you can run via PyTorch from the sentence-transformers project[1].

That model is based on BERT and not LLaMa [2].

[1]: https://www.sbert.net/docs/pretrained_models.html

[2]: https://huggingface.co/microsoft/MiniLM-L12-H384-uncased

For several projects I’ve opted for the even dumber approach, that works out of the box with every ORM/Query DSL framework in every language: using a normal table with SELECT FOR UPDATE SKIP LOCKED

https://www.pgcasts.com/episodes/the-skip-locked-feature-in-...

It’s not “web scale” but it easily extends to several thousand background jobs in my experience

I did not in fact RTF (original) A so thanks for bringing that to my attention, updated the post title.

This is exciting, we built a similar tool[1] recently specifically targeted at constraining llama output to match a TypeScript interface.

I firmly believe that output format guarantees are going to be important for real (non-toy) decades for LLMs

[1] https://github.com/ggerganov/llama.cpp/discussions/2494

Kafka replaced ZooKeeper with Kafka itself already a few years ago https://cwiki.apache.org/confluence/display/KAFKA/KIP-500%3A...

See also https://datatracker.ietf.org/doc/rfc9330/ for the IETF RFC

I believe in their culture this is called a “total mad lad move”

Thanks for the feedback. There's an example in the linked GH discussion but let me provide a quick ETE example here, centered around parsing structured information from a hypothetical shipping company email.

1. In the app, write a TypeScript schema for this, one good example could be

  interface DeliveryInformation {
      /* Tracking number for the delivery */
      tracking_number: string;
      /* Status of the delivery, one of "preparing", "out-for-delivery", or "delivered" */
      status: string;
      /* Weight of the package, e.g. "2oz" or "3lb" */
      weight: string;
      /* submission date time representation */
      submitted_ts: string;
  }

If you click Generate you'll see it generate a context-free grammar looking text, which is what llama.cpp reads. Click the download file to save it as grammar.gbnf

2. Grab a model for llama.cpp, e.g. this quantized Llama2 chat model: https://huggingface.co/TheBloke/Llama-2-7B-Chat-GGML/blob/ma...

3. Grab the prompt in this gist (https://gist.github.com/a10y/d926039eee63cc2bcaf6345f9a419e3...) and save as prompt.txt

4. Compile a recent commit of llama.cpp and run the following command

  ./main -m ./models/llama-2-13b-chat/llama-2-13b-chat.ggmlv3.q8_0.bin -f prompt.txt -c 4096 -n 1000 -t 1 --temp 0 --grammar-file ./grammar.gbnf

EDIT: fixed some typos

This article is better https://www.paloaltoonline.com/news/2023/03/03/judge-rules-h...

The crux is that many students would take the 9th grade math course via private courses to then jump ahead a year or two when they get to the high school. The school tried to prevent people from doing this and the parents sued.

I went to this high school ~10 years ago and at the time a handful of kids every year would do this. I'd imagine it's much higher now though.

BBEdit was my first text editor ~15 years ago! Back when I was really obsessive over the small stuff because I wanted to be like the cool kids. I then got into Text Wrangler, vim, emacs, nvim, etc. over the years. I was really into customizing my environment.

Now I just use IntelliJ and VSCode. Don't really have to think about it, plugin system has grown enormously, and I rarely need to spend time futzing with my configs because excellent themes are free and plentiful. Progress!

The conditions that created OpenAI (oodles of money from some of the wealthiest people on the planet) are not reproducible for the vast majority of businesses.

If you have oodles of money from the wealthiest people on the planet to spend on the worlds best researchers, you too can ignore this rule.

At least w.r.t. to the synonym problem, more modern search techniques that rely on language modeling and word representations seem to solve that.

I've also personally struggled a lot with this at $WORK, but some of the new features in RHEL 7.8+ (specifically GA'ing of the fuse-overlayfs graph driver) have actually alleviated a ton of the issues I've seen previously. If you haven't tried that I'd certainly recommend it, YMMV

I've TA'd all of Stanford's intro programming classes, including the first course in programming CS 106A, so a few things:

- unlike a lot of other schools, CS 106A is mostly taken by non-majors (90-something percent of undergrads take at least one CS course before graduation, and for non-majors that means 106A)

- the follow up course, 106B, is in C++ and focuses on OOP, pointers, memory, data structures, etc.

- 106J was crafted as a replacement for 106A by in JS. It had limited enrollment, and word on the street is 106J's first offering this past spring was an experiment and it didn't go all that well, so who's to say if it'll still be around next year.

Maybe they should switch from Java to something hipster and trendy, but there's about 10+ years worth of assignments that will need to be rewritten, including autograding infrastructure for the 600-800 people who take it every quarter.

I personally feel like JS is an awful language with few opinions or guard rails, and thus a bad language to teach people who are looking for programming principles. It works for the web but I think for a course almost entirely consisting of non-majors it's not a good first language.

Since when does anyone post things on sourceforge anymore? It makes me wonder how legitimate this is...

I was in the room when he made the comment, it's not entirely unfair. The libraries are mostly for computers and group meeting rooms. Many of the books I have checked out have had no stamps since the late 90s, and several I've been first to checkout.